timestamp stringdate 2025-10-22 18:08:07 2025-10-22 18:09:23 | end_timestamp stringdate 2025-10-22 18:09:09 2025-10-22 18:10:23 | stage_name stringclasses 1 value | stage_number int64 1 1 | level stringclasses 1 value | message stringclasses 1 value | stdout_content stringclasses 2 values | stderr_content stringclasses 2 values | experiment_name stringclasses 1 value | elapsed_time_seconds float64 60 61.6 | stage_complete bool 1 class |
|---|---|---|---|---|---|---|---|---|---|---|
2025-10-22T18:08:07.516943 | 2025-10-22T18:09:09.153331 | verl_rl | 1 | INFO | Complete log capture for stage: verl_rl | [INFO] Starting stage: VeRL RL training - rl
[INFO] Data preparation succeeded
[INFO] Starting checkpoint monitoring for intermediate uploads...
[INFO] Intermediate checkpoint upload enabled
[DEBUG] Found 0 global_step directories
[DEBUG] Running verl command:
python -m verl.trainer.main_ppo trainer.total_epochs=50 actor_rollout_ref.actor.optim.lr=1e-06 trainer.save_freq=20 trainer.test_freq=20 trainer.val_before_train=True algorithm.adv_estimator=grpo actor_rollout_ref.rollout.n=16 data.train_batch_size=256 actor_rollout_ref.actor.ppo_mini_batch_size=32 actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16 actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 custom_reward_function.reward_kwargs.response_or_sample=sample custom_reward_function.reward_kwargs.simple_format_reward_weight=0.0 custom_reward_function.reward_kwargs.complex_format_reward_weight=0.0 custom_reward_function.reward_kwargs.sample_correctness_reward_weight=0.0 custom_reward_function.reward_kwargs.verdict_correctness_reward_weight=0.0 custom_reward_function.reward_kwargs.reflection_correctness_reward_weight=0.0 custom_reward_function.reward_kwargs.final_answer_in_samples_reward_weight=0.0 custom_reward_function.reward_kwargs.transition_penalty_weight=0.0 custom_reward_function.reward_kwargs.similarity_penalty_weight=0.0 custom_reward_function.reward_kwargs.sample_count_penalty_weight=0.0 custom_reward_function.reward_kwargs.reward_min=0.0 custom_reward_function.reward_kwargs.reward_max=10.0 reward_model.reward_manager=batch custom_reward_function.name=compute_score_batch reward_model.launch_reward_fn_async=True actor_rollout_ref.model.enable_gradient_checkpointing=True actor_rollout_ref.model.enable_activation_offload=True actor_rollout_ref.rollout.gpu_memory_utilization=0.8 actor_rollout_ref.model.use_remove_padding=True actor_rollout_ref.actor.strategy=fsdp2 actor_rollout_ref.actor.fsdp_config.forward_prefetch=True actor_rollout_ref.ref.fsdp_config.forward_prefetch=True reward_model.model.fsdp_config.forward_prefetch=True actor_rollout_ref.rollout.max_num_batched_tokens=16384 actor_rollout_ref.rollout.max_num_seqs=2048 hydra.run.dir=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/hydra hydra.output_subdir=null hydra.job.chdir=False actor_rollout_ref.rollout.tensor_model_parallel_size=1 data.max_prompt_length=512 data.max_response_length=4096 actor_rollout_ref.model.path=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct actor_rollout_ref.rollout.dtype=bfloat16 critic.optim.lr=1e-05 critic.model.path=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct critic.ppo_micro_batch_size_per_gpu=1 algorithm.kl_ctrl.kl_coef=0.001 trainer.logger=[console,wandb] trainer.project_name=jackrl trainer.experiment_name=1022_longcontextrl__0epoch_3args_grpo_rl trainer.resume_mode=disable data.train_files=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/train.parquet data.val_files=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/test.parquet custom_reward_function.path=/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py trainer.default_local_dir=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/checkpoints actor_rollout_ref.model.trust_remote_code=True critic.model.trust_remote_code=True trainer.nnodes=1 trainer.n_gpus_per_node=2
2025-10-22 18:08:32,673 INFO worker.py:1918 -- Started a local Ray instance. View the dashboard at [1m[32m127.0.0.1:8265 [39m[22m
[36m(TaskRunner pid=1355539)[0m TaskRunner hostname: ga003.hpc.nyu.edu, PID: 1355539
[36m(TaskRunner pid=1355539)[0m {'actor_rollout_ref': {'actor': {'checkpoint': {'load_contents': ['model',
[36m(TaskRunner pid=1355539)[0m 'optimizer',
[36m(TaskRunner pid=1355539)[0m 'extra'],
[36m(TaskRunner pid=1355539)[0m 'save_contents': ['model',
[36m(TaskRunner pid=1355539)[0m 'optimizer',
[36m(TaskRunner pid=1355539)[0m 'extra']},
[36m(TaskRunner pid=1355539)[0m 'clip_ratio': 0.2,
[36m(TaskRunner pid=1355539)[0m 'clip_ratio_c': 3.0,
[36m(TaskRunner pid=1355539)[0m 'clip_ratio_high': 0.2,
[36m(TaskRunner pid=1355539)[0m 'clip_ratio_low': 0.2,
[36m(TaskRunner pid=1355539)[0m 'entropy_checkpointing': False,
[36m(TaskRunner pid=1355539)[0m 'entropy_coeff': 0,
[36m(TaskRunner pid=1355539)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=1355539)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=1355539)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=1355539)[0m 'offload_policy': False,
[36m(TaskRunner pid=1355539)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=1355539)[0m 'param_offload': False,
[36m(TaskRunner pid=1355539)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=1355539)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=1355539)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=1355539)[0m 'kl_loss_coef': 0.001,
[36m(TaskRunner pid=1355539)[0m 'kl_loss_type': 'low_var_kl',
[36m(TaskRunner pid=1355539)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=1355539)[0m 'optim': {'lr': 1e-06,
[36m(TaskRunner pid=1355539)[0m 'lr_warmup_steps': -1,
[36m(TaskRunner pid=1355539)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=1355539)[0m 'min_lr_ratio': 0.0,
[36m(TaskRunner pid=1355539)[0m 'num_cycles': 0.5,
[36m(TaskRunner pid=1355539)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=1355539)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=1355539)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=1355539)[0m 'policy_loss': {'clip_cov_lb': 1.0,
[36m(TaskRunner pid=1355539)[0m 'clip_cov_ratio': 0.0002,
[36m(TaskRunner pid=1355539)[0m 'clip_cov_ub': 5.0,
[36m(TaskRunner pid=1355539)[0m 'kl_cov_ratio': 0.0002,
[36m(TaskRunner pid=1355539)[0m 'loss_mode': 'vanilla',
[36m(TaskRunner pid=1355539)[0m 'ppo_kl_coef': 0.1},
[36m(TaskRunner pid=1355539)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=1355539)[0m 'ppo_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=1355539)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=1355539)[0m 'ppo_micro_batch_size_per_gpu': 8,
[36m(TaskRunner pid=1355539)[0m 'ppo_mini_batch_size': 32,
[36m(TaskRunner pid=1355539)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1355539)[0m 'all_ranks': False,
[36m(TaskRunner pid=1355539)[0m 'discrete': False,
[36m(TaskRunner pid=1355539)[0m 'ranks': []},
[36m(TaskRunner pid=1355539)[0m 'shuffle': False,
[36m(TaskRunner pid=1355539)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=1355539)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=1355539)[0m 'use_dynamic_bsz': False,
[36m(TaskRunner pid=1355539)[0m 'use_kl_loss': False,
[36m(TaskRunner pid=1355539)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=1355539)[0m 'hybrid_engine': True,
[36m(TaskRunner pid=1355539)[0m 'model': {'custom_chat_template': None,
[36m(TaskRunner pid=1355539)[0m 'enable_activation_offload': True,
[36m(TaskRunner pid=1355539)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=1355539)[0m 'exclude_modules': None,
[36m(TaskRunner pid=1355539)[0m 'external_lib': None,
[36m(TaskRunner pid=1355539)[0m 'fused_kernel_options': {'impl_backend': 'torch'},
[36m(TaskRunner pid=1355539)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=1355539)[0m 'lora_rank': 0,
[36m(TaskRunner pid=1355539)[0m 'override_config': {},
[36m(TaskRunner pid=1355539)[0m 'path': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct',
[36m(TaskRunner pid=1355539)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=1355539)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=1355539)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=1355539)[0m 'use_liger': False,
[36m(TaskRunner pid=1355539)[0m 'use_remove_padding': True,
[36m(TaskRunner pid=1355539)[0m 'use_shm': False},
[36m(TaskRunner pid=1355539)[0m 'ref': {'entropy_checkpointing': False,
[36m(TaskRunner pid=1355539)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=1355539)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=1355539)[0m 'param_offload': False,
[36m(TaskRunner pid=1355539)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=1355539)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=1355539)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=1355539)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=1355539)[0m 'log_prob_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=1355539)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=1355539)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1355539)[0m 'all_ranks': False,
[36m(TaskRunner pid=1355539)[0m 'discrete': False,
[36m(TaskRunner pid=1355539)[0m 'ranks': []},
[36m(TaskRunner pid=1355539)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=1355539)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=1355539)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=1355539)[0m 'rollout': {'agent': {'custom_async_server': {'name': None,
[36m(TaskRunner pid=1355539)[0m 'path': None},
[36m(TaskRunner pid=1355539)[0m 'num_workers': 8},
[36m(TaskRunner pid=1355539)[0m 'calculate_log_probs': False,
[36m(TaskRunner pid=1355539)[0m 'disable_log_stats': True,
[36m(TaskRunner pid=1355539)[0m 'do_sample': True,
[36m(TaskRunner pid=1355539)[0m 'dtype': 'bfloat16',
[36m(TaskRunner pid=1355539)[0m 'enable_chunked_prefill': True,
[36m(TaskRunner pid=1355539)[0m 'enforce_eager': True,
[36m(TaskRunner pid=1355539)[0m 'engine_kwargs': {'sglang': {'attention_backend': None},
[36m(TaskRunner pid=1355539)[0m 'vllm': {'disable_mm_preprocessor_cache': False,
[36m(TaskRunner pid=1355539)[0m 'swap_space': None}},
[36m(TaskRunner pid=1355539)[0m 'free_cache_engine': True,
[36m(TaskRunner pid=1355539)[0m 'gpu_memory_utilization': 0.8,
[36m(TaskRunner pid=1355539)[0m 'ignore_eos': False,
[36m(TaskRunner pid=1355539)[0m 'layered_summon': False,
[36m(TaskRunner pid=1355539)[0m 'load_format': 'dummy_dtensor',
[36m(TaskRunner pid=1355539)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=1355539)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=1355539)[0m 'log_prob_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=1355539)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=1355539)[0m 'max_model_len': None,
[36m(TaskRunner pid=1355539)[0m 'max_num_batched_tokens': 16384,
[36m(TaskRunner pid=1355539)[0m 'max_num_seqs': 2048,
[36m(TaskRunner pid=1355539)[0m 'mode': 'sync',
[36m(TaskRunner pid=1355539)[0m 'multi_stage_wake_up': False,
[36m(TaskRunner pid=1355539)[0m 'multi_turn': {'completion_callback': None,
[36m(TaskRunner pid=1355539)[0m 'enable': False,
[36m(TaskRunner pid=1355539)[0m 'format': 'hermes',
[36m(TaskRunner pid=1355539)[0m 'interaction_config_path': None,
[36m(TaskRunner pid=1355539)[0m 'max_assistant_turns': None,
[36m(TaskRunner pid=1355539)[0m 'max_parallel_calls': 1,
[36m(TaskRunner pid=1355539)[0m 'max_tool_response_length': 256,
[36m(TaskRunner pid=1355539)[0m 'max_user_turns': None,
[36m(TaskRunner pid=1355539)[0m 'tokenization_sanity_check_mode': 'strict',
[36m(TaskRunner pid=1355539)[0m 'tool_config_path': None,
[36m(TaskRunner pid=1355539)[0m 'tool_response_truncate_side': 'middle',
[36m(TaskRunner pid=1355539)[0m 'use_inference_chat_template': False},
[36m(TaskRunner pid=1355539)[0m 'n': 16,
[36m(TaskRunner pid=1355539)[0m 'name': 'vllm',
[36m(TaskRunner pid=1355539)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1355539)[0m 'all_ranks': False,
[36m(TaskRunner pid=1355539)[0m 'discrete': False,
[36m(TaskRunner pid=1355539)[0m 'ranks': []},
[36m(TaskRunner pid=1355539)[0m 'prompt_length': 512,
[36m(TaskRunner pid=1355539)[0m 'response_length': 4096,
[36m(TaskRunner pid=1355539)[0m 'temperature': 1.0,
[36m(TaskRunner pid=1355539)[0m 'tensor_model_parallel_size': 1,
[36m(TaskRunner pid=1355539)[0m 'top_k': -1,
[36m(TaskRunner pid=1355539)[0m 'top_p': 1,
[36m(TaskRunner pid=1355539)[0m 'val_kwargs': {'do_sample': False,
[36m(TaskRunner pid=1355539)[0m 'n': 1,
[36m(TaskRunner pid=1355539)[0m 'temperature': 0,
[36m(TaskRunner pid=1355539)[0m 'top_k': -1,
[36m(TaskRunner pid=1355539)[0m 'top_p': 1.0}}},
[36m(TaskRunner pid=1355539)[0m 'algorithm': {'_target_': 'verl.trainer.config.AlgoConfig',
[36m(TaskRunner pid=1355539)[0m 'adv_estimator': 'grpo',
[36m(TaskRunner pid=1355539)[0m 'gamma': 1.0,
[36m(TaskRunner pid=1355539)[0m 'kl_ctrl': {'_target_': 'verl.trainer.config.KLControlConfig',
[36m(TaskRunner pid=1355539)[0m 'horizon': 10000,
[36m(TaskRunner pid=1355539)[0m 'kl_coef': 0.001,
[36m(TaskRunner pid=1355539)[0m 'target_kl': 0.1,
[36m(TaskRunner pid=1355539)[0m 'type': 'fixed'},
[36m(TaskRunner pid=1355539)[0m 'kl_penalty': 'kl',
[36m(TaskRunner pid=1355539)[0m 'lam': 1.0,
[36m(TaskRunner pid=1355539)[0m 'norm_adv_by_std_in_grpo': True,
[36m(TaskRunner pid=1355539)[0m 'pf_ppo': {'_target_': 'verl.trainer.config.PFPPOConfig',
[36m(TaskRunner pid=1355539)[0m 'reweight_method': 'pow',
[36m(TaskRunner pid=1355539)[0m 'weight_pow': 2.0},
[36m(TaskRunner pid=1355539)[0m 'use_kl_in_reward': False,
[36m(TaskRunner pid=1355539)[0m 'use_pf_ppo': False},
[36m(TaskRunner pid=1355539)[0m 'critic': {'checkpoint': {'load_contents': ['model', 'optimizer', 'extra'],
[36m(TaskRunner pid=1355539)[0m 'save_contents': ['model', 'optimizer', 'extra']},
[36m(TaskRunner pid=1355539)[0m 'cliprange_value': 0.5,
[36m(TaskRunner pid=1355539)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=1355539)[0m 'forward_micro_batch_size': None,
[36m(TaskRunner pid=1355539)[0m 'forward_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=1355539)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=1355539)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=1355539)[0m 'model': {'enable_activation_offload': False,
[36m(TaskRunner pid=1355539)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=1355539)[0m 'external_lib': None,
[36m(TaskRunner pid=1355539)[0m 'fsdp_config': {'forward_prefetch': False,
[36m(TaskRunner pid=1355539)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=1355539)[0m 'offload_policy': False,
[36m(TaskRunner pid=1355539)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=1355539)[0m 'param_offload': False,
[36m(TaskRunner pid=1355539)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=1355539)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=1355539)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=1355539)[0m 'lora_rank': 0,
[36m(TaskRunner pid=1355539)[0m 'override_config': {},
[36m(TaskRunner pid=1355539)[0m 'path': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct',
[36m(TaskRunner pid=1355539)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=1355539)[0m 'tokenizer_path': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct',
[36m(TaskRunner pid=1355539)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=1355539)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=1355539)[0m 'use_shm': False},
[36m(TaskRunner pid=1355539)[0m 'optim': {'lr': 1e-05,
[36m(TaskRunner pid=1355539)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=1355539)[0m 'min_lr_ratio': None,
[36m(TaskRunner pid=1355539)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=1355539)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=1355539)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=1355539)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=1355539)[0m 'ppo_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=1355539)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=1355539)[0m 'ppo_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=1355539)[0m 'ppo_mini_batch_size': 32,
[36m(TaskRunner pid=1355539)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1355539)[0m 'all_ranks': False,
[36m(TaskRunner pid=1355539)[0m 'discrete': False,
[36m(TaskRunner pid=1355539)[0m 'ranks': []},
[36m(TaskRunner pid=1355539)[0m 'rollout_n': 16,
[36m(TaskRunner pid=1355539)[0m 'shuffle': False,
[36m(TaskRunner pid=1355539)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=1355539)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=1355539)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=1355539)[0m 'custom_reward_function': {'name': 'compute_score_batch',
[36m(TaskRunner pid=1355539)[0m 'path': '/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py',
[36m(TaskRunner pid=1355539)[0m 'reward_kwargs': {'complex_format_reward_weight': 0.0,
[36m(TaskRunner pid=1355539)[0m 'final_answer_in_samples_reward_weight': 0.0,
[36m(TaskRunner pid=1355539)[0m 'reflection_correctness_reward_weight': 0.0,
[36m(TaskRunner pid=1355539)[0m 'response_or_sample': 'sample',
[36m(TaskRunner pid=1355539)[0m 'reward_max': 10.0,
[36m(TaskRunner pid=1355539)[0m 'reward_min': 0.0,
[36m(TaskRunner pid=1355539)[0m 'sample_correctness_reward_weight': 0.0,
[36m(TaskRunner pid=1355539)[0m 'sample_count_penalty_weight': 0.0,
[36m(TaskRunner pid=1355539)[0m 'similarity_penalty_weight': 0.0,
[36m(TaskRunner pid=1355539)[0m 'simple_format_reward_weight': 0.0,
[36m(TaskRunner pid=1355539)[0m 'transition_penalty_weight': 0.0,
[36m(TaskRunner pid=1355539)[0m 'verdict_correctness_reward_weight': 0.0}},
[36m(TaskRunner pid=1355539)[0m 'data': {'custom_cls': {'name': None, 'path': None},
[36m(TaskRunner pid=1355539)[0m 'dataloader_num_workers': 8,
[36m(TaskRunner pid=1355539)[0m 'filter_overlong_prompts': False,
[36m(TaskRunner pid=1355539)[0m 'filter_overlong_prompts_workers': 1,
[36m(TaskRunner pid=1355539)[0m 'image_key': 'images',
[36m(TaskRunner pid=1355539)[0m 'max_prompt_length': 512,
[36m(TaskRunner pid=1355539)[0m 'max_response_length': 4096,
[36m(TaskRunner pid=1355539)[0m 'prompt_key': 'prompt',
[36m(TaskRunner pid=1355539)[0m 'return_full_prompt': False,
[36m(TaskRunner pid=1355539)[0m 'return_multi_modal_inputs': True,
[36m(TaskRunner pid=1355539)[0m 'return_raw_chat': False,
[36m(TaskRunner pid=1355539)[0m 'return_raw_input_ids': False,
[36m(TaskRunner pid=1355539)[0m 'reward_fn_key': 'data_source',
[36m(TaskRunner pid=1355539)[0m 'sampler': {'class_name': None, 'class_path': None},
[36m(TaskRunner pid=1355539)[0m 'shuffle': True,
[36m(TaskRunner pid=1355539)[0m 'tokenizer': None,
[36m(TaskRunner pid=1355539)[0m 'train_batch_size': 256,
[36m(TaskRunner pid=1355539)[0m 'train_files':
[36m(TaskRunner pid=1355539)[0m '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/train.parquet',
[36m(TaskRunner pid=1355539)[0m 'truncation': 'error',
[36m(TaskRunner pid=1355539)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=1355539)[0m 'use_shm': False,
[36m(TaskRunner pid=1355539)[0m 'val_batch_size': None,
[36m(TaskRunner pid=1355539)[0m 'val_files': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/test.parquet',
[36m(TaskRunner pid=1355539)[0m 'validation_shuffle': False,
[36m(TaskRunner pid=1355539)[0m 'video_key': 'videos'},
[36m(TaskRunner pid=1355539)[0m 'ray_init': {'num_cpus': None, 'timeline_json_file': None},
[36m(TaskRunner pid=1355539)[0m 'reward_model': {'enable': False,
[36m(TaskRunner pid=1355539)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=1355539)[0m 'launch_reward_fn_async': True,
[36m(TaskRunner pid=1355539)[0m 'max_length': None,
[36m(TaskRunner pid=1355539)[0m 'micro_batch_size': None,
[36m(TaskRunner pid=1355539)[0m 'micro_batch_size_per_gpu': None,
[36m(TaskRunner pid=1355539)[0m 'model': {'external_lib': None,
[36m(TaskRunner pid=1355539)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=1355539)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=1355539)[0m 'param_offload': False,
[36m(TaskRunner pid=1355539)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=1355539)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=1355539)[0m 'input_tokenizer': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct',
[36m(TaskRunner pid=1355539)[0m 'path': '~/models/FsfairX-LLaMA3-RM-v0.1',
[36m(TaskRunner pid=1355539)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=1355539)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=1355539)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=1355539)[0m 'use_shm': False},
[36m(TaskRunner pid=1355539)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1355539)[0m 'all_ranks': False,
[36m(TaskRunner pid=1355539)[0m 'discrete': False,
[36m(TaskRunner pid=1355539)[0m 'ranks': []},
[36m(TaskRunner pid=1355539)[0m 'reward_manager': 'batch',
[36m(TaskRunner pid=1355539)[0m 'sandbox_fusion': {'max_concurrent': 64,
[36m(TaskRunner pid=1355539)[0m 'memory_limit_mb': 1024,
[36m(TaskRunner pid=1355539)[0m 'url': None},
[36m(TaskRunner pid=1355539)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=1355539)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=1355539)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=1355539)[0m 'trainer': {'balance_batch': True,
[36m(TaskRunner pid=1355539)[0m 'controller_nsight_options': {'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=1355539)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=1355539)[0m 'trace': 'cuda,nvtx,cublas,ucx'},
[36m(TaskRunner pid=1355539)[0m 'critic_warmup': 0,
[36m(TaskRunner pid=1355539)[0m 'default_hdfs_dir': None,
[36m(TaskRunner pid=1355539)[0m 'default_local_dir': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/checkpoints',
[36m(TaskRunner pid=1355539)[0m 'del_local_ckpt_after_load': False,
[36m(TaskRunner pid=1355539)[0m 'device': 'cuda',
[36m(TaskRunner pid=1355539)[0m 'esi_redundant_time': 0,
[36m(TaskRunner pid=1355539)[0m 'experiment_name': '1022_longcontextrl__0epoch_3args_grpo_rl',
[36m(TaskRunner pid=1355539)[0m 'log_val_generations': 0,
[36m(TaskRunner pid=1355539)[0m 'logger': ['console', 'wandb'],
[36m(TaskRunner pid=1355539)[0m 'max_actor_ckpt_to_keep': None,
[36m(TaskRunner pid=1355539)[0m 'max_critic_ckpt_to_keep': None,
[36m(TaskRunner pid=1355539)[0m 'n_gpus_per_node': 2,
[36m(TaskRunner pid=1355539)[0m 'nnodes': 1,
[36m(TaskRunner pid=1355539)[0m 'profile_steps': None,
[36m(TaskRunner pid=1355539)[0m 'project_name': 'jackrl',
[36m(TaskRunner pid=1355539)[0m 'ray_wait_register_center_timeout': 300,
[36m(TaskRunner pid=1355539)[0m 'resume_from_path': None,
[36m(TaskRunner pid=1355539)[0m 'resume_mode': 'disable',
[36m(TaskRunner pid=1355539)[0m 'rollout_data_dir': None,
[36m(TaskRunner pid=1355539)[0m 'save_freq': 20,
[36m(TaskRunner pid=1355539)[0m 'test_freq': 20,
[36m(TaskRunner pid=1355539)[0m 'total_epochs': 50,
[36m(TaskRunner pid=1355539)[0m 'total_training_steps': None,
[36m(TaskRunner pid=1355539)[0m 'val_before_train': True,
[36m(TaskRunner pid=1355539)[0m 'val_only': False,
[36m(TaskRunner pid=1355539)[0m 'validation_data_dir': None,
[36m(TaskRunner pid=1355539)[0m 'worker_nsight_options': {'capture-range': 'cudaProfilerApi',
[36m(TaskRunner pid=1355539)[0m 'capture-range-end': None,
[36m(TaskRunner pid=1355539)[0m 'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=1355539)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=1355539)[0m 'kill': 'none',
[36m(TaskRunner pid=1355539)[0m 'trace': 'cuda,nvtx,cublas,ucx'}}}
[36m(TaskRunner pid=1355539)[0m Registered source: longmult
[36m(TaskRunner pid=1355539)[0m Registered source: countdown
[36m(TaskRunner pid=1355539)[0m Registered source: gsm8k
[36m(TaskRunner pid=1355539)[0m Registered source: arc
[36m(TaskRunner pid=1355539)[0m Registered source: arc_challenge
[36m(TaskRunner pid=1355539)[0m Registered source: arc_easy
[36m(TaskRunner pid=1355539)[0m Registered source: piqa
[36m(TaskRunner pid=1355539)[0m Registered source: mmlu
[36m(TaskRunner pid=1355539)[0m Registered source: mmlu_pro
[36m(TaskRunner pid=1355539)[0m Registered source: csqa
[36m(TaskRunner pid=1355539)[0m Registered source: social_iqa
[36m(TaskRunner pid=1355539)[0m Registered source: strategy_qa
[36m(TaskRunner pid=1355539)[0m Registered source: winogrande
[36m(TaskRunner pid=1355539)[0m Registered source: bbh
[36m(TaskRunner pid=1355539)[0m Registered source: letter_countdown
[36m(TaskRunner pid=1355539)[0m Registered source: acronym
[36m(TaskRunner pid=1355539)[0m using customized reward function 'compute_score_batch' from '/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=1355539)[0m using customized reward function 'compute_score_batch' from '/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=1355539)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=1355539)[0m
Generating train split: 0 examples [00:00, ? examples/s]
[36m(TaskRunner pid=1355539)[0m
Generating train split: 1000 examples [00:00, 2857.54 examples/s]
Generating train split: 1000 examples [00:00, 2788.48 examples/s]
[36m(TaskRunner pid=1355539)[0m dataset len: 1000
[36m(TaskRunner pid=1355539)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=1355539)[0m dataset len: 2450
[36m(TaskRunner pid=1355539)[0m Using critic: False
[36m(TaskRunner pid=1355539)[0m [validate_config] All configuration checks passed successfully!
[36m(TaskRunner pid=1355539)[0m
Generating train split: 0 examples [00:00, ? examples/s]
Generating train split: 2450 examples [00:00, 111664.58 examples/s]
[36m(TaskRunner pid=1355539)[0m Size of train dataloader: 3, Size of val dataloader: 1
[36m(TaskRunner pid=1355539)[0m Total training steps: 150
[36m(TaskRunner pid=1355539)[0m DeprecationWarning: `ray.state.available_resources_per_node` is a private attribute and access will be removed in a future Ray version.
[36m(TaskRunner pid=1355539)[0m {'61839b0d69bf0403b44c2bd98583fb2ffc385d6e620d55bfdb57b8fc': {'CPU': 127.0,
[36m(TaskRunner pid=1355539)[0m 'GPU': 2.0,
[36m(TaskRunner pid=1355539)[0m 'accelerator_type:A100': 1.0,
[36m(TaskRunner pid=1355539)[0m 'memory': 360675969024.0,
[36m(TaskRunner pid=1355539)[0m 'node:10.32.35.179': 1.0,
[36m(TaskRunner pid=1355539)[0m 'node:__internal_head__': 1.0,
[36m(TaskRunner pid=1355539)[0m 'object_store_memory': 154575415296.0}}
[36m(TaskRunner pid=1355539)[0m ('Resource pool to cls: {<verl.single_controller.ray.base.RayResourcePool '
[36m(TaskRunner pid=1355539)[0m "object at 0x1469e0e721d0>: {'actor_rollout': "
[36m(TaskRunner pid=1355539)[0m '<verl.single_controller.ray.base.RayClassWithInitArgs object at '
[36m(TaskRunner pid=1355539)[0m '0x1469e0e72200>}}')
[36m(TaskRunner pid=1355539)[0m colocated worker base class <class 'verl.single_controller.base.worker.Worker'>
[DEBUG] Found 0 global_step directories
[36m(TaskRunner pid=1355539)[0m WARNING:2025-10-22 18:08:54,300:Waiting for register center actor qYM3Zu_register_center to be ready. Elapsed time: 0 seconds out of 300 seconds.
[36m(WorkerDict pid=1360025)[0m Exception raised in creation task: The actor died because of an error raised in its creation task, [36mray::qYM3ZuWorkerDict_0:0:WorkerDict.__init__()[39m (pid=1360025, ip=10.32.35.179, actor_id=d6abb71f5afcb72b0b999a0201000000, repr=<verl.single_controller.ray.base.WorkerDict object at 0x14d7e9325a50>)
[36m(WorkerDict pid=1360025)[0m File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 779, in __init__
[36m(WorkerDict pid=1360025)[0m super().__init__()
[36m(WorkerDict pid=1360025)[0m File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/worker.py", line 161, in __init__
[36m(WorkerDict pid=1360025)[0m self._setup_env_cuda_visible_devices()
[36m(WorkerDict pid=1360025)[0m File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/worker.py", line 233, in _setup_env_cuda_visible_devices
[36m(WorkerDict pid=1360025)[0m raise ValueError("Please don't set ROCR_VISIBLE_DEVICES when HIP/CUDA_VISIBLE_DEVICES is set.")
[36m(WorkerDict pid=1360025)[0m ValueError: Please don't set ROCR_VISIBLE_DEVICES when HIP/CUDA_VISIBLE_DEVICES is set.
Error executing job with overrides: ['trainer.total_epochs=50', 'actor_rollout_ref.actor.optim.lr=1e-06', 'trainer.save_freq=20', 'trainer.test_freq=20', 'trainer.val_before_train=True', 'algorithm.adv_estimator=grpo', 'actor_rollout_ref.rollout.n=16', 'data.train_batch_size=256', 'actor_rollout_ref.actor.ppo_mini_batch_size=32', 'actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8', 'actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16', 'actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16', 'custom_reward_function.reward_kwargs.response_or_sample=sample', 'custom_reward_function.reward_kwargs.simple_format_reward_weight=0.0', 'custom_reward_function.reward_kwargs.complex_format_reward_weight=0.0', 'custom_reward_function.reward_kwargs.sample_correctness_reward_weight=0.0', 'custom_reward_function.reward_kwargs.verdict_correctness_reward_weight=0.0', 'custom_reward_function.reward_kwargs.reflection_correctness_reward_weight=0.0', 'custom_reward_function.reward_kwargs.final_answer_in_samples_reward_weight=0.0', 'custom_reward_function.reward_kwargs.transition_penalty_weight=0.0', 'custom_reward_function.reward_kwargs.similarity_penalty_weight=0.0', 'custom_reward_function.reward_kwargs.sample_count_penalty_weight=0.0', 'custom_reward_function.reward_kwargs.reward_min=0.0', 'custom_reward_function.reward_kwargs.reward_max=10.0', 'reward_model.reward_manager=batch', 'custom_reward_function.name=compute_score_batch', 'reward_model.launch_reward_fn_async=True', 'actor_rollout_ref.model.enable_gradient_checkpointing=True', 'actor_rollout_ref.model.enable_activation_offload=True', 'actor_rollout_ref.rollout.gpu_memory_utilization=0.8', 'actor_rollout_ref.model.use_remove_padding=True', 'actor_rollout_ref.actor.strategy=fsdp2', 'actor_rollout_ref.actor.fsdp_config.forward_prefetch=True', 'actor_rollout_ref.ref.fsdp_config.forward_prefetch=True', 'reward_model.model.fsdp_config.forward_prefetch=True', 'actor_rollout_ref.rollout.max_num_batched_tokens=16384', 'actor_rollout_ref.rollout.max_num_seqs=2048', 'actor_rollout_ref.rollout.tensor_model_parallel_size=1', 'data.max_prompt_length=512', 'data.max_response_length=4096', 'actor_rollout_ref.model.path=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct', 'actor_rollout_ref.rollout.dtype=bfloat16', 'critic.optim.lr=1e-05', 'critic.model.path=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct', 'critic.ppo_micro_batch_size_per_gpu=1', 'algorithm.kl_ctrl.kl_coef=0.001', 'trainer.logger=[console,wandb]', 'trainer.project_name=jackrl', 'trainer.experiment_name=1022_longcontextrl__0epoch_3args_grpo_rl', 'trainer.resume_mode=disable', 'data.train_files=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/train.parquet', 'data.val_files=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/test.parquet', 'custom_reward_function.path=/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py', 'trainer.default_local_dir=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/checkpoints', 'actor_rollout_ref.model.trust_remote_code=True', 'critic.model.trust_remote_code=True', 'trainer.nnodes=1', 'trainer.n_gpus_per_node=2']
Traceback (most recent call last):
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 39, in main
run_ppo(config)
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 69, in run_ppo
ray.get(runner.run.remote(config))
File "/scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/ray/_private/auto_init_hook.py", line 22, in auto_init_wrapper
return fn(*args, **kwargs)
File "/scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/ray/_private/client_mode_hook.py", line 104, in wrapper
return func(*args, **kwargs)
File "/scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/ray/_private/worker.py", line 2858, in get
values, debugger_breakpoint = worker.get_objects(object_refs, timeout=timeout)
File "/scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/ray/_private/worker.py", line 958, in get_objects
raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(ActorDiedError): [36mray::TaskRunner.run()[39m (pid=1355539, ip=10.32.35.179, actor_id=7cb1bdf2dea30cc15ac853f101000000, repr=<main_ppo.TaskRunner object at 0x146a34543e20>)
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 232, in run
trainer.init_workers()
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/trainer/ppo/ray_trainer.py", line 931, in init_workers
self.actor_rollout_wg.init_model()
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 51, in __call__
output = ray.get(output)
ray.exceptions.ActorDiedError: The actor died because of an error raised in its creation task, [36mray::qYM3ZuWorkerDict_0:0:WorkerDict.__init__()[39m (pid=1360025, ip=10.32.35.179, actor_id=d6abb71f5afcb72b0b999a0201000000, repr=<verl.single_controller.ray.base.WorkerDict object at 0x14d7e9325a50>)
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 779, in __init__
super().__init__()
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/worker.py", line 161, in __init__
self._setup_env_cuda_visible_devices()
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/worker.py", line 233, in _setup_env_cuda_visible_devices
raise ValueError("Please don't set ROCR_VISIBLE_DEVICES when HIP/CUDA_VISIBLE_DEVICES is set.")
ValueError: Please don't set ROCR_VISIBLE_DEVICES when HIP/CUDA_VISIBLE_DEVICES is set.
Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.
[INFO] Extracting model from VeRL checkpoint at /scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/checkpoints
[ERROR] No global_step directories found
EXTRACT OUT: False
[ERROR] Stage error: RuntimeError: Model extraction failed
| /scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/huggingface_hub/file_download.py:980: UserWarning: `local_dir_use_symlinks` parameter is deprecated and will be ignored. The process to download files to a local folder has been updated and do not rely on symlinks anymore. You only need to pass a destination folder as`local_dir`.
For more details, check out https://huggingface.co/docs/huggingface_hub/main/en/guides/download#download-files-to-local-folder.
warnings.warn(
Fetching 10 files: 0%| | 0/10 [00:00<?, ?it/s]
LICENSE: 0.00B [00:00, ?B/s][A
LICENSE: 11.3kB [00:00, 68.4MB/s]
.gitattributes: 0.00B [00:00, ?B/s][A
.gitattributes: 1.52kB [00:00, 12.8MB/s]
Fetching 10 files: 10%|█████████████████████▉ | 1/10 [00:00<00:01, 6.40it/s]
README.md: 0.00B [00:00, ?B/s][A
README.md: 4.92kB [00:00, 17.5MB/s]
Fetching 10 files: 70%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▎ | 7/10 [00:01<00:00, 6.91it/s]
Fetching 10 files: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:01<00:00, 9.83it/s]
Fetching 10 files: 0%| | 0/10 [00:00<?, ?it/s]
Fetching 10 files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 2534.17it/s]
| 1022_longcontextrl__0epoch_3args_grpo | 61.636388 | true |
2025-10-22T18:09:23.530270 | 2025-10-22T18:10:23.561198 | verl_rl | 1 | INFO | Complete log capture for stage: verl_rl | [INFO] Starting stage: VeRL RL training - rl
[INFO] Data preparation succeeded
[INFO] Intermediate checkpoint upload enabled
[INFO] Starting checkpoint monitoring for intermediate uploads...
[DEBUG] Found 0 global_step directories
[DEBUG] Running verl command:
python -m verl.trainer.main_ppo trainer.total_epochs=50 actor_rollout_ref.actor.optim.lr=1e-06 trainer.save_freq=20 trainer.test_freq=20 trainer.val_before_train=True algorithm.adv_estimator=grpo actor_rollout_ref.rollout.n=16 data.train_batch_size=256 actor_rollout_ref.actor.ppo_mini_batch_size=32 actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16 actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 custom_reward_function.reward_kwargs.response_or_sample=sample custom_reward_function.reward_kwargs.simple_format_reward_weight=0.0 custom_reward_function.reward_kwargs.complex_format_reward_weight=0.0 custom_reward_function.reward_kwargs.sample_correctness_reward_weight=0.0 custom_reward_function.reward_kwargs.verdict_correctness_reward_weight=0.0 custom_reward_function.reward_kwargs.reflection_correctness_reward_weight=0.0 custom_reward_function.reward_kwargs.final_answer_in_samples_reward_weight=0.0 custom_reward_function.reward_kwargs.transition_penalty_weight=0.0 custom_reward_function.reward_kwargs.similarity_penalty_weight=0.0 custom_reward_function.reward_kwargs.sample_count_penalty_weight=0.0 custom_reward_function.reward_kwargs.reward_min=0.0 custom_reward_function.reward_kwargs.reward_max=10.0 reward_model.reward_manager=batch custom_reward_function.name=compute_score_batch reward_model.launch_reward_fn_async=True actor_rollout_ref.model.enable_gradient_checkpointing=True actor_rollout_ref.model.enable_activation_offload=True actor_rollout_ref.rollout.gpu_memory_utilization=0.8 actor_rollout_ref.model.use_remove_padding=True actor_rollout_ref.actor.strategy=fsdp2 actor_rollout_ref.actor.fsdp_config.forward_prefetch=True actor_rollout_ref.ref.fsdp_config.forward_prefetch=True reward_model.model.fsdp_config.forward_prefetch=True actor_rollout_ref.rollout.max_num_batched_tokens=16384 actor_rollout_ref.rollout.max_num_seqs=2048 hydra.run.dir=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/hydra hydra.output_subdir=null hydra.job.chdir=False actor_rollout_ref.rollout.tensor_model_parallel_size=1 data.max_prompt_length=512 data.max_response_length=4096 actor_rollout_ref.model.path=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct actor_rollout_ref.rollout.dtype=bfloat16 critic.optim.lr=1e-05 critic.model.path=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct critic.ppo_micro_batch_size_per_gpu=1 algorithm.kl_ctrl.kl_coef=0.001 trainer.logger=[console,wandb] trainer.project_name=jackrl trainer.experiment_name=1022_longcontextrl__0epoch_3args_grpo_rl trainer.resume_mode=disable data.train_files=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/train.parquet data.val_files=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/test.parquet custom_reward_function.path=/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py trainer.default_local_dir=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/checkpoints actor_rollout_ref.model.trust_remote_code=True critic.model.trust_remote_code=True trainer.nnodes=1 trainer.n_gpus_per_node=2
2025-10-22 18:09:47,239 INFO worker.py:1918 -- Started a local Ray instance. View the dashboard at [1m[32m127.0.0.1:8265 [39m[22m
[36m(TaskRunner pid=1369074)[0m TaskRunner hostname: ga003.hpc.nyu.edu, PID: 1369074
[36m(TaskRunner pid=1369074)[0m {'actor_rollout_ref': {'actor': {'checkpoint': {'load_contents': ['model',
[36m(TaskRunner pid=1369074)[0m 'optimizer',
[36m(TaskRunner pid=1369074)[0m 'extra'],
[36m(TaskRunner pid=1369074)[0m 'save_contents': ['model',
[36m(TaskRunner pid=1369074)[0m 'optimizer',
[36m(TaskRunner pid=1369074)[0m 'extra']},
[36m(TaskRunner pid=1369074)[0m 'clip_ratio': 0.2,
[36m(TaskRunner pid=1369074)[0m 'clip_ratio_c': 3.0,
[36m(TaskRunner pid=1369074)[0m 'clip_ratio_high': 0.2,
[36m(TaskRunner pid=1369074)[0m 'clip_ratio_low': 0.2,
[36m(TaskRunner pid=1369074)[0m 'entropy_checkpointing': False,
[36m(TaskRunner pid=1369074)[0m 'entropy_coeff': 0,
[36m(TaskRunner pid=1369074)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=1369074)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=1369074)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=1369074)[0m 'offload_policy': False,
[36m(TaskRunner pid=1369074)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=1369074)[0m 'param_offload': False,
[36m(TaskRunner pid=1369074)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=1369074)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=1369074)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=1369074)[0m 'kl_loss_coef': 0.001,
[36m(TaskRunner pid=1369074)[0m 'kl_loss_type': 'low_var_kl',
[36m(TaskRunner pid=1369074)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=1369074)[0m 'optim': {'lr': 1e-06,
[36m(TaskRunner pid=1369074)[0m 'lr_warmup_steps': -1,
[36m(TaskRunner pid=1369074)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=1369074)[0m 'min_lr_ratio': 0.0,
[36m(TaskRunner pid=1369074)[0m 'num_cycles': 0.5,
[36m(TaskRunner pid=1369074)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=1369074)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=1369074)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=1369074)[0m 'policy_loss': {'clip_cov_lb': 1.0,
[36m(TaskRunner pid=1369074)[0m 'clip_cov_ratio': 0.0002,
[36m(TaskRunner pid=1369074)[0m 'clip_cov_ub': 5.0,
[36m(TaskRunner pid=1369074)[0m 'kl_cov_ratio': 0.0002,
[36m(TaskRunner pid=1369074)[0m 'loss_mode': 'vanilla',
[36m(TaskRunner pid=1369074)[0m 'ppo_kl_coef': 0.1},
[36m(TaskRunner pid=1369074)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=1369074)[0m 'ppo_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=1369074)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=1369074)[0m 'ppo_micro_batch_size_per_gpu': 8,
[36m(TaskRunner pid=1369074)[0m 'ppo_mini_batch_size': 32,
[36m(TaskRunner pid=1369074)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1369074)[0m 'all_ranks': False,
[36m(TaskRunner pid=1369074)[0m 'discrete': False,
[36m(TaskRunner pid=1369074)[0m 'ranks': []},
[36m(TaskRunner pid=1369074)[0m 'shuffle': False,
[36m(TaskRunner pid=1369074)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=1369074)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=1369074)[0m 'use_dynamic_bsz': False,
[36m(TaskRunner pid=1369074)[0m 'use_kl_loss': False,
[36m(TaskRunner pid=1369074)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=1369074)[0m 'hybrid_engine': True,
[36m(TaskRunner pid=1369074)[0m 'model': {'custom_chat_template': None,
[36m(TaskRunner pid=1369074)[0m 'enable_activation_offload': True,
[36m(TaskRunner pid=1369074)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=1369074)[0m 'exclude_modules': None,
[36m(TaskRunner pid=1369074)[0m 'external_lib': None,
[36m(TaskRunner pid=1369074)[0m 'fused_kernel_options': {'impl_backend': 'torch'},
[36m(TaskRunner pid=1369074)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=1369074)[0m 'lora_rank': 0,
[36m(TaskRunner pid=1369074)[0m 'override_config': {},
[36m(TaskRunner pid=1369074)[0m 'path': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct',
[36m(TaskRunner pid=1369074)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=1369074)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=1369074)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=1369074)[0m 'use_liger': False,
[36m(TaskRunner pid=1369074)[0m 'use_remove_padding': True,
[36m(TaskRunner pid=1369074)[0m 'use_shm': False},
[36m(TaskRunner pid=1369074)[0m 'ref': {'entropy_checkpointing': False,
[36m(TaskRunner pid=1369074)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=1369074)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=1369074)[0m 'param_offload': False,
[36m(TaskRunner pid=1369074)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=1369074)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=1369074)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=1369074)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=1369074)[0m 'log_prob_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=1369074)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=1369074)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1369074)[0m 'all_ranks': False,
[36m(TaskRunner pid=1369074)[0m 'discrete': False,
[36m(TaskRunner pid=1369074)[0m 'ranks': []},
[36m(TaskRunner pid=1369074)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=1369074)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=1369074)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=1369074)[0m 'rollout': {'agent': {'custom_async_server': {'name': None,
[36m(TaskRunner pid=1369074)[0m 'path': None},
[36m(TaskRunner pid=1369074)[0m 'num_workers': 8},
[36m(TaskRunner pid=1369074)[0m 'calculate_log_probs': False,
[36m(TaskRunner pid=1369074)[0m 'disable_log_stats': True,
[36m(TaskRunner pid=1369074)[0m 'do_sample': True,
[36m(TaskRunner pid=1369074)[0m 'dtype': 'bfloat16',
[36m(TaskRunner pid=1369074)[0m 'enable_chunked_prefill': True,
[36m(TaskRunner pid=1369074)[0m 'enforce_eager': True,
[36m(TaskRunner pid=1369074)[0m 'engine_kwargs': {'sglang': {'attention_backend': None},
[36m(TaskRunner pid=1369074)[0m 'vllm': {'disable_mm_preprocessor_cache': False,
[36m(TaskRunner pid=1369074)[0m 'swap_space': None}},
[36m(TaskRunner pid=1369074)[0m 'free_cache_engine': True,
[36m(TaskRunner pid=1369074)[0m 'gpu_memory_utilization': 0.8,
[36m(TaskRunner pid=1369074)[0m 'ignore_eos': False,
[36m(TaskRunner pid=1369074)[0m 'layered_summon': False,
[36m(TaskRunner pid=1369074)[0m 'load_format': 'dummy_dtensor',
[36m(TaskRunner pid=1369074)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=1369074)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=1369074)[0m 'log_prob_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=1369074)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=1369074)[0m 'max_model_len': None,
[36m(TaskRunner pid=1369074)[0m 'max_num_batched_tokens': 16384,
[36m(TaskRunner pid=1369074)[0m 'max_num_seqs': 2048,
[36m(TaskRunner pid=1369074)[0m 'mode': 'sync',
[36m(TaskRunner pid=1369074)[0m 'multi_stage_wake_up': False,
[36m(TaskRunner pid=1369074)[0m 'multi_turn': {'completion_callback': None,
[36m(TaskRunner pid=1369074)[0m 'enable': False,
[36m(TaskRunner pid=1369074)[0m 'format': 'hermes',
[36m(TaskRunner pid=1369074)[0m 'interaction_config_path': None,
[36m(TaskRunner pid=1369074)[0m 'max_assistant_turns': None,
[36m(TaskRunner pid=1369074)[0m 'max_parallel_calls': 1,
[36m(TaskRunner pid=1369074)[0m 'max_tool_response_length': 256,
[36m(TaskRunner pid=1369074)[0m 'max_user_turns': None,
[36m(TaskRunner pid=1369074)[0m 'tokenization_sanity_check_mode': 'strict',
[36m(TaskRunner pid=1369074)[0m 'tool_config_path': None,
[36m(TaskRunner pid=1369074)[0m 'tool_response_truncate_side': 'middle',
[36m(TaskRunner pid=1369074)[0m 'use_inference_chat_template': False},
[36m(TaskRunner pid=1369074)[0m 'n': 16,
[36m(TaskRunner pid=1369074)[0m 'name': 'vllm',
[36m(TaskRunner pid=1369074)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1369074)[0m 'all_ranks': False,
[36m(TaskRunner pid=1369074)[0m 'discrete': False,
[36m(TaskRunner pid=1369074)[0m 'ranks': []},
[36m(TaskRunner pid=1369074)[0m 'prompt_length': 512,
[36m(TaskRunner pid=1369074)[0m 'response_length': 4096,
[36m(TaskRunner pid=1369074)[0m 'temperature': 1.0,
[36m(TaskRunner pid=1369074)[0m 'tensor_model_parallel_size': 1,
[36m(TaskRunner pid=1369074)[0m 'top_k': -1,
[36m(TaskRunner pid=1369074)[0m 'top_p': 1,
[36m(TaskRunner pid=1369074)[0m 'val_kwargs': {'do_sample': False,
[36m(TaskRunner pid=1369074)[0m 'n': 1,
[36m(TaskRunner pid=1369074)[0m 'temperature': 0,
[36m(TaskRunner pid=1369074)[0m 'top_k': -1,
[36m(TaskRunner pid=1369074)[0m 'top_p': 1.0}}},
[36m(TaskRunner pid=1369074)[0m 'algorithm': {'_target_': 'verl.trainer.config.AlgoConfig',
[36m(TaskRunner pid=1369074)[0m 'adv_estimator': 'grpo',
[36m(TaskRunner pid=1369074)[0m 'gamma': 1.0,
[36m(TaskRunner pid=1369074)[0m 'kl_ctrl': {'_target_': 'verl.trainer.config.KLControlConfig',
[36m(TaskRunner pid=1369074)[0m 'horizon': 10000,
[36m(TaskRunner pid=1369074)[0m 'kl_coef': 0.001,
[36m(TaskRunner pid=1369074)[0m 'target_kl': 0.1,
[36m(TaskRunner pid=1369074)[0m 'type': 'fixed'},
[36m(TaskRunner pid=1369074)[0m 'kl_penalty': 'kl',
[36m(TaskRunner pid=1369074)[0m 'lam': 1.0,
[36m(TaskRunner pid=1369074)[0m 'norm_adv_by_std_in_grpo': True,
[36m(TaskRunner pid=1369074)[0m 'pf_ppo': {'_target_': 'verl.trainer.config.PFPPOConfig',
[36m(TaskRunner pid=1369074)[0m 'reweight_method': 'pow',
[36m(TaskRunner pid=1369074)[0m 'weight_pow': 2.0},
[36m(TaskRunner pid=1369074)[0m 'use_kl_in_reward': False,
[36m(TaskRunner pid=1369074)[0m 'use_pf_ppo': False},
[36m(TaskRunner pid=1369074)[0m 'critic': {'checkpoint': {'load_contents': ['model', 'optimizer', 'extra'],
[36m(TaskRunner pid=1369074)[0m 'save_contents': ['model', 'optimizer', 'extra']},
[36m(TaskRunner pid=1369074)[0m 'cliprange_value': 0.5,
[36m(TaskRunner pid=1369074)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=1369074)[0m 'forward_micro_batch_size': None,
[36m(TaskRunner pid=1369074)[0m 'forward_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=1369074)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=1369074)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=1369074)[0m 'model': {'enable_activation_offload': False,
[36m(TaskRunner pid=1369074)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=1369074)[0m 'external_lib': None,
[36m(TaskRunner pid=1369074)[0m 'fsdp_config': {'forward_prefetch': False,
[36m(TaskRunner pid=1369074)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=1369074)[0m 'offload_policy': False,
[36m(TaskRunner pid=1369074)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=1369074)[0m 'param_offload': False,
[36m(TaskRunner pid=1369074)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=1369074)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=1369074)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=1369074)[0m 'lora_rank': 0,
[36m(TaskRunner pid=1369074)[0m 'override_config': {},
[36m(TaskRunner pid=1369074)[0m 'path': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct',
[36m(TaskRunner pid=1369074)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=1369074)[0m 'tokenizer_path': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct',
[36m(TaskRunner pid=1369074)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=1369074)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=1369074)[0m 'use_shm': False},
[36m(TaskRunner pid=1369074)[0m 'optim': {'lr': 1e-05,
[36m(TaskRunner pid=1369074)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=1369074)[0m 'min_lr_ratio': None,
[36m(TaskRunner pid=1369074)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=1369074)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=1369074)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=1369074)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=1369074)[0m 'ppo_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=1369074)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=1369074)[0m 'ppo_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=1369074)[0m 'ppo_mini_batch_size': 32,
[36m(TaskRunner pid=1369074)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1369074)[0m 'all_ranks': False,
[36m(TaskRunner pid=1369074)[0m 'discrete': False,
[36m(TaskRunner pid=1369074)[0m 'ranks': []},
[36m(TaskRunner pid=1369074)[0m 'rollout_n': 16,
[36m(TaskRunner pid=1369074)[0m 'shuffle': False,
[36m(TaskRunner pid=1369074)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=1369074)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=1369074)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=1369074)[0m 'custom_reward_function': {'name': 'compute_score_batch',
[36m(TaskRunner pid=1369074)[0m 'path': '/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py',
[36m(TaskRunner pid=1369074)[0m 'reward_kwargs': {'complex_format_reward_weight': 0.0,
[36m(TaskRunner pid=1369074)[0m 'final_answer_in_samples_reward_weight': 0.0,
[36m(TaskRunner pid=1369074)[0m 'reflection_correctness_reward_weight': 0.0,
[36m(TaskRunner pid=1369074)[0m 'response_or_sample': 'sample',
[36m(TaskRunner pid=1369074)[0m 'reward_max': 10.0,
[36m(TaskRunner pid=1369074)[0m 'reward_min': 0.0,
[36m(TaskRunner pid=1369074)[0m 'sample_correctness_reward_weight': 0.0,
[36m(TaskRunner pid=1369074)[0m 'sample_count_penalty_weight': 0.0,
[36m(TaskRunner pid=1369074)[0m 'similarity_penalty_weight': 0.0,
[36m(TaskRunner pid=1369074)[0m 'simple_format_reward_weight': 0.0,
[36m(TaskRunner pid=1369074)[0m 'transition_penalty_weight': 0.0,
[36m(TaskRunner pid=1369074)[0m 'verdict_correctness_reward_weight': 0.0}},
[36m(TaskRunner pid=1369074)[0m 'data': {'custom_cls': {'name': None, 'path': None},
[36m(TaskRunner pid=1369074)[0m 'dataloader_num_workers': 8,
[36m(TaskRunner pid=1369074)[0m 'filter_overlong_prompts': False,
[36m(TaskRunner pid=1369074)[0m 'filter_overlong_prompts_workers': 1,
[36m(TaskRunner pid=1369074)[0m 'image_key': 'images',
[36m(TaskRunner pid=1369074)[0m 'max_prompt_length': 512,
[36m(TaskRunner pid=1369074)[0m 'max_response_length': 4096,
[36m(TaskRunner pid=1369074)[0m 'prompt_key': 'prompt',
[36m(TaskRunner pid=1369074)[0m 'return_full_prompt': False,
[36m(TaskRunner pid=1369074)[0m 'return_multi_modal_inputs': True,
[36m(TaskRunner pid=1369074)[0m 'return_raw_chat': False,
[36m(TaskRunner pid=1369074)[0m 'return_raw_input_ids': False,
[36m(TaskRunner pid=1369074)[0m 'reward_fn_key': 'data_source',
[36m(TaskRunner pid=1369074)[0m 'sampler': {'class_name': None, 'class_path': None},
[36m(TaskRunner pid=1369074)[0m 'shuffle': True,
[36m(TaskRunner pid=1369074)[0m 'tokenizer': None,
[36m(TaskRunner pid=1369074)[0m 'train_batch_size': 256,
[36m(TaskRunner pid=1369074)[0m 'train_files': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/train.parquet',
[36m(TaskRunner pid=1369074)[0m 'truncation': 'error',
[36m(TaskRunner pid=1369074)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=1369074)[0m 'use_shm': False,
[36m(TaskRunner pid=1369074)[0m 'val_batch_size': None,
[36m(TaskRunner pid=1369074)[0m 'val_files': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/test.parquet',
[36m(TaskRunner pid=1369074)[0m 'validation_shuffle': False,
[36m(TaskRunner pid=1369074)[0m 'video_key': 'videos'},
[36m(TaskRunner pid=1369074)[0m 'ray_init': {'num_cpus': None, 'timeline_json_file': None},
[36m(TaskRunner pid=1369074)[0m 'reward_model': {'enable': False,
[36m(TaskRunner pid=1369074)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=1369074)[0m 'launch_reward_fn_async': True,
[36m(TaskRunner pid=1369074)[0m 'max_length': None,
[36m(TaskRunner pid=1369074)[0m 'micro_batch_size': None,
[36m(TaskRunner pid=1369074)[0m 'micro_batch_size_per_gpu': None,
[36m(TaskRunner pid=1369074)[0m 'model': {'external_lib': None,
[36m(TaskRunner pid=1369074)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=1369074)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=1369074)[0m 'param_offload': False,
[36m(TaskRunner pid=1369074)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=1369074)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=1369074)[0m 'input_tokenizer': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct',
[36m(TaskRunner pid=1369074)[0m 'path': '~/models/FsfairX-LLaMA3-RM-v0.1',
[36m(TaskRunner pid=1369074)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=1369074)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=1369074)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=1369074)[0m 'use_shm': False},
[36m(TaskRunner pid=1369074)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=1369074)[0m 'all_ranks': False,
[36m(TaskRunner pid=1369074)[0m 'discrete': False,
[36m(TaskRunner pid=1369074)[0m 'ranks': []},
[36m(TaskRunner pid=1369074)[0m 'reward_manager': 'batch',
[36m(TaskRunner pid=1369074)[0m 'sandbox_fusion': {'max_concurrent': 64,
[36m(TaskRunner pid=1369074)[0m 'memory_limit_mb': 1024,
[36m(TaskRunner pid=1369074)[0m 'url': None},
[36m(TaskRunner pid=1369074)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=1369074)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=1369074)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=1369074)[0m 'trainer': {'balance_batch': True,
[36m(TaskRunner pid=1369074)[0m 'controller_nsight_options': {'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=1369074)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=1369074)[0m 'trace': 'cuda,nvtx,cublas,ucx'},
[36m(TaskRunner pid=1369074)[0m 'critic_warmup': 0,
[36m(TaskRunner pid=1369074)[0m 'default_hdfs_dir': None,
[36m(TaskRunner pid=1369074)[0m 'default_local_dir': '/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/checkpoints',
[36m(TaskRunner pid=1369074)[0m 'del_local_ckpt_after_load': False,
[36m(TaskRunner pid=1369074)[0m 'device': 'cuda',
[36m(TaskRunner pid=1369074)[0m 'esi_redundant_time': 0,
[36m(TaskRunner pid=1369074)[0m 'experiment_name': '1022_longcontextrl__0epoch_3args_grpo_rl',
[36m(TaskRunner pid=1369074)[0m 'log_val_generations': 0,
[36m(TaskRunner pid=1369074)[0m 'logger': ['console', 'wandb'],
[36m(TaskRunner pid=1369074)[0m 'max_actor_ckpt_to_keep': None,
[36m(TaskRunner pid=1369074)[0m 'max_critic_ckpt_to_keep': None,
[36m(TaskRunner pid=1369074)[0m 'n_gpus_per_node': 2,
[36m(TaskRunner pid=1369074)[0m 'nnodes': 1,
[36m(TaskRunner pid=1369074)[0m 'profile_steps': None,
[36m(TaskRunner pid=1369074)[0m 'project_name': 'jackrl',
[36m(TaskRunner pid=1369074)[0m 'ray_wait_register_center_timeout': 300,
[36m(TaskRunner pid=1369074)[0m 'resume_from_path': None,
[36m(TaskRunner pid=1369074)[0m 'resume_mode': 'disable',
[36m(TaskRunner pid=1369074)[0m 'rollout_data_dir': None,
[36m(TaskRunner pid=1369074)[0m 'save_freq': 20,
[36m(TaskRunner pid=1369074)[0m 'test_freq': 20,
[36m(TaskRunner pid=1369074)[0m 'total_epochs': 50,
[36m(TaskRunner pid=1369074)[0m 'total_training_steps': None,
[36m(TaskRunner pid=1369074)[0m 'val_before_train': True,
[36m(TaskRunner pid=1369074)[0m 'val_only': False,
[36m(TaskRunner pid=1369074)[0m 'validation_data_dir': None,
[36m(TaskRunner pid=1369074)[0m 'worker_nsight_options': {'capture-range': 'cudaProfilerApi',
[36m(TaskRunner pid=1369074)[0m 'capture-range-end': None,
[36m(TaskRunner pid=1369074)[0m 'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=1369074)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=1369074)[0m 'kill': 'none',
[36m(TaskRunner pid=1369074)[0m 'trace': 'cuda,nvtx,cublas,ucx'}}}
[36m(TaskRunner pid=1369074)[0m Registered source: longmult
[36m(TaskRunner pid=1369074)[0m Registered source: countdown
[36m(TaskRunner pid=1369074)[0m Registered source: gsm8k
[36m(TaskRunner pid=1369074)[0m Registered source: arc
[36m(TaskRunner pid=1369074)[0m Registered source: arc_challenge
[36m(TaskRunner pid=1369074)[0m Registered source: arc_easy
[36m(TaskRunner pid=1369074)[0m Registered source: piqa
[36m(TaskRunner pid=1369074)[0m Registered source: mmlu
[36m(TaskRunner pid=1369074)[0m Registered source: mmlu_pro
[36m(TaskRunner pid=1369074)[0m Registered source: csqa
[36m(TaskRunner pid=1369074)[0m Registered source: social_iqa
[36m(TaskRunner pid=1369074)[0m Registered source: strategy_qa
[36m(TaskRunner pid=1369074)[0m Registered source: winogrande
[36m(TaskRunner pid=1369074)[0m Registered source: bbh
[36m(TaskRunner pid=1369074)[0m Registered source: letter_countdown
[36m(TaskRunner pid=1369074)[0m Registered source: acronym
[36m(TaskRunner pid=1369074)[0m using customized reward function 'compute_score_batch' from '/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=1369074)[0m using customized reward function 'compute_score_batch' from '/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=1369074)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=1369074)[0m
Generating train split: 0 examples [00:00, ? examples/s]
[36m(TaskRunner pid=1369074)[0m dataset len: 1000
[36m(TaskRunner pid=1369074)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=1369074)[0m dataset len: 2450
[36m(TaskRunner pid=1369074)[0m Using critic: False
[36m(TaskRunner pid=1369074)[0m [validate_config] All configuration checks passed successfully!
[36m(TaskRunner pid=1369074)[0m
Generating train split: 1000 examples [00:00, 2851.36 examples/s]
Generating train split: 1000 examples [00:00, 2784.34 examples/s]
[36m(TaskRunner pid=1369074)[0m
Generating train split: 0 examples [00:00, ? examples/s]
Generating train split: 2450 examples [00:00, 108907.17 examples/s]
[36m(TaskRunner pid=1369074)[0m Size of train dataloader: 3, Size of val dataloader: 1
[36m(TaskRunner pid=1369074)[0m Total training steps: 150
[36m(TaskRunner pid=1369074)[0m {'a85bf532ed66bc8e0024be05ed57e829792824a011cf2ef813777492': {'CPU': 127.0,
[36m(TaskRunner pid=1369074)[0m 'GPU': 2.0,
[36m(TaskRunner pid=1369074)[0m 'accelerator_type:A100': 1.0,
[36m(TaskRunner pid=1369074)[0m 'memory': 360676897997.0,
[36m(TaskRunner pid=1369074)[0m 'node:10.32.35.179': 1.0,
[36m(TaskRunner pid=1369074)[0m 'node:__internal_head__': 1.0,
[36m(TaskRunner pid=1369074)[0m 'object_store_memory': 154575813427.0}}
[36m(TaskRunner pid=1369074)[0m ('Resource pool to cls: {<verl.single_controller.ray.base.RayResourcePool '
[36m(TaskRunner pid=1369074)[0m "object at 0x144385c1e110>: {'actor_rollout': "
[36m(TaskRunner pid=1369074)[0m '<verl.single_controller.ray.base.RayClassWithInitArgs object at '
[36m(TaskRunner pid=1369074)[0m '0x144385c1e140>}}')
[36m(TaskRunner pid=1369074)[0m colocated worker base class <class 'verl.single_controller.base.worker.Worker'>
[36m(TaskRunner pid=1369074)[0m DeprecationWarning: `ray.state.available_resources_per_node` is a private attribute and access will be removed in a future Ray version.
[36m(TaskRunner pid=1369074)[0m WARNING:2025-10-22 18:10:08,661:Waiting for register center actor s12IqR_register_center to be ready. Elapsed time: 0 seconds out of 300 seconds.
[DEBUG] Found 0 global_step directories
[36m(WorkerDict pid=1373557)[0m Exception raised in creation task: The actor died because of an error raised in its creation task, [36mray::s12IqRWorkerDict_0:0:WorkerDict.__init__()[39m (pid=1373557, ip=10.32.35.179, actor_id=d37635c71f92e9d924dc517501000000, repr=<verl.single_controller.ray.base.WorkerDict object at 0x14619ba13c10>)
[36m(WorkerDict pid=1373557)[0m File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 779, in __init__
[36m(WorkerDict pid=1373557)[0m super().__init__()
[36m(WorkerDict pid=1373557)[0m File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/worker.py", line 161, in __init__
[36m(WorkerDict pid=1373557)[0m self._setup_env_cuda_visible_devices()
[36m(WorkerDict pid=1373557)[0m File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/worker.py", line 233, in _setup_env_cuda_visible_devices
[36m(WorkerDict pid=1373557)[0m raise ValueError("Please don't set ROCR_VISIBLE_DEVICES when HIP/CUDA_VISIBLE_DEVICES is set.")
[36m(WorkerDict pid=1373557)[0m ValueError: Please don't set ROCR_VISIBLE_DEVICES when HIP/CUDA_VISIBLE_DEVICES is set.
Error executing job with overrides: ['trainer.total_epochs=50', 'actor_rollout_ref.actor.optim.lr=1e-06', 'trainer.save_freq=20', 'trainer.test_freq=20', 'trainer.val_before_train=True', 'algorithm.adv_estimator=grpo', 'actor_rollout_ref.rollout.n=16', 'data.train_batch_size=256', 'actor_rollout_ref.actor.ppo_mini_batch_size=32', 'actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8', 'actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16', 'actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16', 'custom_reward_function.reward_kwargs.response_or_sample=sample', 'custom_reward_function.reward_kwargs.simple_format_reward_weight=0.0', 'custom_reward_function.reward_kwargs.complex_format_reward_weight=0.0', 'custom_reward_function.reward_kwargs.sample_correctness_reward_weight=0.0', 'custom_reward_function.reward_kwargs.verdict_correctness_reward_weight=0.0', 'custom_reward_function.reward_kwargs.reflection_correctness_reward_weight=0.0', 'custom_reward_function.reward_kwargs.final_answer_in_samples_reward_weight=0.0', 'custom_reward_function.reward_kwargs.transition_penalty_weight=0.0', 'custom_reward_function.reward_kwargs.similarity_penalty_weight=0.0', 'custom_reward_function.reward_kwargs.sample_count_penalty_weight=0.0', 'custom_reward_function.reward_kwargs.reward_min=0.0', 'custom_reward_function.reward_kwargs.reward_max=10.0', 'reward_model.reward_manager=batch', 'custom_reward_function.name=compute_score_batch', 'reward_model.launch_reward_fn_async=True', 'actor_rollout_ref.model.enable_gradient_checkpointing=True', 'actor_rollout_ref.model.enable_activation_offload=True', 'actor_rollout_ref.rollout.gpu_memory_utilization=0.8', 'actor_rollout_ref.model.use_remove_padding=True', 'actor_rollout_ref.actor.strategy=fsdp2', 'actor_rollout_ref.actor.fsdp_config.forward_prefetch=True', 'actor_rollout_ref.ref.fsdp_config.forward_prefetch=True', 'reward_model.model.fsdp_config.forward_prefetch=True', 'actor_rollout_ref.rollout.max_num_batched_tokens=16384', 'actor_rollout_ref.rollout.max_num_seqs=2048', 'actor_rollout_ref.rollout.tensor_model_parallel_size=1', 'data.max_prompt_length=512', 'data.max_response_length=4096', 'actor_rollout_ref.model.path=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct', 'actor_rollout_ref.rollout.dtype=bfloat16', 'critic.optim.lr=1e-05', 'critic.model.path=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct', 'critic.ppo_micro_batch_size_per_gpu=1', 'algorithm.kl_ctrl.kl_coef=0.001', 'trainer.logger=[console,wandb]', 'trainer.project_name=jackrl', 'trainer.experiment_name=1022_longcontextrl__0epoch_3args_grpo_rl', 'trainer.resume_mode=disable', 'data.train_files=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/train.parquet', 'data.val_files=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/data/test.parquet', 'custom_reward_function.path=/scratch/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py', 'trainer.default_local_dir=/scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/checkpoints', 'actor_rollout_ref.model.trust_remote_code=True', 'critic.model.trust_remote_code=True', 'trainer.nnodes=1', 'trainer.n_gpus_per_node=2']
Traceback (most recent call last):
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 39, in main
run_ppo(config)
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 69, in run_ppo
ray.get(runner.run.remote(config))
File "/scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/ray/_private/auto_init_hook.py", line 22, in auto_init_wrapper
return fn(*args, **kwargs)
File "/scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/ray/_private/client_mode_hook.py", line 104, in wrapper
return func(*args, **kwargs)
File "/scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/ray/_private/worker.py", line 2858, in get
values, debugger_breakpoint = worker.get_objects(object_refs, timeout=timeout)
File "/scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/ray/_private/worker.py", line 958, in get_objects
raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(ActorDiedError): [36mray::TaskRunner.run()[39m (pid=1369074, ip=10.32.35.179, actor_id=fc3b2ae202b42e6f339d7c1301000000, repr=<main_ppo.TaskRunner object at 0x1443d92e3eb0>)
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 232, in run
trainer.init_workers()
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/trainer/ppo/ray_trainer.py", line 931, in init_workers
self.actor_rollout_wg.init_model()
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 51, in __call__
output = ray.get(output)
ray.exceptions.ActorDiedError: The actor died because of an error raised in its creation task, [36mray::s12IqRWorkerDict_0:0:WorkerDict.__init__()[39m (pid=1373557, ip=10.32.35.179, actor_id=d37635c71f92e9d924dc517501000000, repr=<verl.single_controller.ray.base.WorkerDict object at 0x14619ba13c10>)
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 779, in __init__
super().__init__()
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/worker.py", line 161, in __init__
self._setup_env_cuda_visible_devices()
File "/scratch/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/worker.py", line 233, in _setup_env_cuda_visible_devices
raise ValueError("Please don't set ROCR_VISIBLE_DEVICES when HIP/CUDA_VISIBLE_DEVICES is set.")
ValueError: Please don't set ROCR_VISIBLE_DEVICES when HIP/CUDA_VISIBLE_DEVICES is set.
Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.
[INFO] Extracting model from VeRL checkpoint at /scratch/yl11330/skill-factory/workflow_out/1022_longcontextrl__0epoch_3args_grpo/verl/checkpoints
[ERROR] No global_step directories found
EXTRACT OUT: False
[ERROR] Stage error: RuntimeError: Model extraction failed
| /scratch/yl11330/skill-factory/penv/lib/python3.10/site-packages/huggingface_hub/file_download.py:980: UserWarning: `local_dir_use_symlinks` parameter is deprecated and will be ignored. The process to download files to a local folder has been updated and do not rely on symlinks anymore. You only need to pass a destination folder as`local_dir`.
For more details, check out https://huggingface.co/docs/huggingface_hub/main/en/guides/download#download-files-to-local-folder.
warnings.warn(
Fetching 10 files: 0%| | 0/10 [00:00<?, ?it/s]
Fetching 10 files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 2392.37it/s]
Fetching 10 files: 0%| | 0/10 [00:00<?, ?it/s]
Fetching 10 files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 2557.35it/s]
| 1022_longcontextrl__0epoch_3args_grpo | 60.030928 | true |
README.md exists but content is empty.
- Downloads last month
- 9