diff --git a/02c97dc5bf4ec64f0a8a17d3c48a37d0/run.log b/02c97dc5bf4ec64f0a8a17d3c48a37d0/run.log index 99b9d308976fc5badcda9f145198391e5cbdd54c..a6cb70c5ef0ba6ff6eb8c4aeb016899d25f9fdd6 100644 --- a/02c97dc5bf4ec64f0a8a17d3c48a37d0/run.log +++ b/02c97dc5bf4ec64f0a8a17d3c48a37d0/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:42:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:42:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/02c97dc5bf4ec64f0a8a17d3c48a37d0/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a vehicle driving through a calm urban residential area during early morning or late afternoon, with soft, warm lighting suggesting dawn or dusk. The road is lined with parked cars, trees, and occasional traffic cones or barriers, possibly indicating temporary roadwork. Tall apartment buildings are visible in the background, and the sky has a pale purple hue. As the vehicle moves forward, it passes by various street elements like utility poles, residential homes, and sparse traffic (including a few parked or slowly moving cars). No pedestrians or cyclists are visible, and the vehicle maintains a steady pace on the straight road, with no significant turns or stops. The environment feels quiet and orderly, with no notable events like near-misses or lane changes.', '--save_path', '/workspace/video2world_out/02c97dc5bf4ec64f0a8a17d3c48a37d0/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:42:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a vehicle driving through a calm urban residential area during early morning or late afternoon, with soft, warm lighting suggesting dawn or dusk. The road is lined with parked cars, trees, and occasional traffic cones or barriers, possibly indicating temporary roadwork. Tall apartment buildings are visible in the background, and the sky has a pale purple hue. As the vehicle moves forward, it passes by various street elements like utility poles, residential homes, and sparse traffic (including a few parked or slowly moving cars). No pedestrians or cyclists are visible, and the vehicle maintains a steady pace on the straight road, with no significant turns or stops. The environment feels quiet and orderly, with no notable events like near-misses or lane changes.', input_path='/workspace/video2world_out/02c97dc5bf4ec64f0a8a17d3c48a37d0/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/02c97dc5bf4ec64f0a8a17d3c48a37d0/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:42:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:42:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:42:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:42:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/02c97dc5bf4ec64f0a8a17d3c48a37d0/video2world_2B.yaml. -[01-15 17:42:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:42:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:42:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:42:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:45:30|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:08:48|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/02c97dc5bf4ec64f0a8a17d3c48a37d0/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a vehicle driving through a calm urban residential area during early morning or late afternoon, with soft, warm lighting suggesting dawn or dusk. The road is lined with parked cars, trees, and occasional traffic cones or barriers, possibly indicating temporary roadwork. Tall apartment buildings are visible in the background, and the sky has a pale purple hue. As the vehicle moves forward, it passes by various street elements like utility poles, residential homes, and sparse traffic (including a few parked or slowly moving cars). No pedestrians or cyclists are visible, and the vehicle maintains a steady pace on the straight road, with no significant turns or stops. The environment feels quiet and orderly, with no notable events like near-misses or lane changes.', '--save_path', '/workspace/video2world_out_14B/02c97dc5bf4ec64f0a8a17d3c48a37d0/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:08:48|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a vehicle driving through a calm urban residential area during early morning or late afternoon, with soft, warm lighting suggesting dawn or dusk. The road is lined with parked cars, trees, and occasional traffic cones or barriers, possibly indicating temporary roadwork. Tall apartment buildings are visible in the background, and the sky has a pale purple hue. As the vehicle moves forward, it passes by various street elements like utility poles, residential homes, and sparse traffic (including a few parked or slowly moving cars). No pedestrians or cyclists are visible, and the vehicle maintains a steady pace on the straight road, with no significant turns or stops. The environment feels quiet and orderly, with no notable events like near-misses or lane changes.', input_path='/workspace/video2world_out_14B/02c97dc5bf4ec64f0a8a17d3c48a37d0/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/02c97dc5bf4ec64f0a8a17d3c48a37d0/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:08:48|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:08:48|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:08:48|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:08:48|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/02c97dc5bf4ec64f0a8a17d3c48a37d0/video2world_14B.yaml. +[01-15 22:08:48|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:08:48|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:08:50|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:08:50|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:11:10|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/072ae5a4543eee1ed73e95bc93d8c243/run.log b/072ae5a4543eee1ed73e95bc93d8c243/run.log index 6f0d1a431316fff62f3ad1c8745e964799946b39..961355157d555c18ce4e6e9f86a6bf60ca224a01 100644 --- a/072ae5a4543eee1ed73e95bc93d8c243/run.log +++ b/072ae5a4543eee1ed73e95bc93d8c243/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:42:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:42:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/072ae5a4543eee1ed73e95bc93d8c243/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'A vehicle drives through a quiet suburban neighborhood on an overcast day, navigating paved streets lined with houses, trees, and parked cars. The lighting is diffused, indicating cloudy conditions. The car moves straight along residential roads, passing intersections where it stops at a stop sign. No pedestrians, cyclists, or significant traffic events are visible; the scene remains calm with consistent, slow movement. The environment features single-family homes, streetlights, and occasional utility poles, with no notable near-misses or abrupt actions', '--save_path', '/workspace/video2world_out/072ae5a4543eee1ed73e95bc93d8c243/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:42:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='A vehicle drives through a quiet suburban neighborhood on an overcast day, navigating paved streets lined with houses, trees, and parked cars. The lighting is diffused, indicating cloudy conditions. The car moves straight along residential roads, passing intersections where it stops at a stop sign. No pedestrians, cyclists, or significant traffic events are visible; the scene remains calm with consistent, slow movement. The environment features single-family homes, streetlights, and occasional utility poles, with no notable near-misses or abrupt actions', input_path='/workspace/video2world_out/072ae5a4543eee1ed73e95bc93d8c243/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/072ae5a4543eee1ed73e95bc93d8c243/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:42:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:42:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:42:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:42:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/072ae5a4543eee1ed73e95bc93d8c243/video2world_2B.yaml. -[01-15 17:42:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:42:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:42:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:42:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:45:42|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:08:49|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/072ae5a4543eee1ed73e95bc93d8c243/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'A vehicle drives through a quiet suburban neighborhood on an overcast day, navigating paved streets lined with houses, trees, and parked cars. The lighting is diffused, indicating cloudy conditions. The car moves straight along residential roads, passing intersections where it stops at a stop sign. No pedestrians, cyclists, or significant traffic events are visible; the scene remains calm with consistent, slow movement. The environment features single-family homes, streetlights, and occasional utility poles, with no notable near-misses or abrupt actions', '--save_path', '/workspace/video2world_out_14B/072ae5a4543eee1ed73e95bc93d8c243/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:08:49|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='A vehicle drives through a quiet suburban neighborhood on an overcast day, navigating paved streets lined with houses, trees, and parked cars. The lighting is diffused, indicating cloudy conditions. The car moves straight along residential roads, passing intersections where it stops at a stop sign. No pedestrians, cyclists, or significant traffic events are visible; the scene remains calm with consistent, slow movement. The environment features single-family homes, streetlights, and occasional utility poles, with no notable near-misses or abrupt actions', input_path='/workspace/video2world_out_14B/072ae5a4543eee1ed73e95bc93d8c243/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/072ae5a4543eee1ed73e95bc93d8c243/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:08:49|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:08:49|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:08:49|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:08:49|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/072ae5a4543eee1ed73e95bc93d8c243/video2world_14B.yaml. +[01-15 22:08:49|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:08:49|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:08:50|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:08:50|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:11:05|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/0766fe2271f1c36b2a8d73885c913731/run.log b/0766fe2271f1c36b2a8d73885c913731/run.log index bcb87c0f222e7cecd0ec26eedd7802c1096a4a84..f13f47791d22ccf56245b6318271636942c79aa0 100644 --- a/0766fe2271f1c36b2a8d73885c913731/run.log +++ b/0766fe2271f1c36b2a8d73885c913731/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:42:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:42:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/0766fe2271f1c36b2a8d73885c913731/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a car', '--save_path', '/workspace/video2world_out/0766fe2271f1c36b2a8d73885c913731/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:42:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a car', input_path='/workspace/video2world_out/0766fe2271f1c36b2a8d73885c913731/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/0766fe2271f1c36b2a8d73885c913731/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:42:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:42:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:42:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:42:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/0766fe2271f1c36b2a8d73885c913731/video2world_2B.yaml. -[01-15 17:42:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:42:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:42:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:42:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:45:36|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:08:47|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/0766fe2271f1c36b2a8d73885c913731/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a car', '--save_path', '/workspace/video2world_out_14B/0766fe2271f1c36b2a8d73885c913731/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:08:47|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a car', input_path='/workspace/video2world_out_14B/0766fe2271f1c36b2a8d73885c913731/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/0766fe2271f1c36b2a8d73885c913731/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:08:47|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:08:47|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:08:47|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:08:47|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/0766fe2271f1c36b2a8d73885c913731/video2world_14B.yaml. +[01-15 22:08:47|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:08:47|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:08:49|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:08:49|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:11:07|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/0856a3f1d05e3aa70d8284d5c67d5f03/run.log b/0856a3f1d05e3aa70d8284d5c67d5f03/run.log index 820700f6348f1e55e8611421a1186bf866e54039..303b300c835e9eb9a6f8b8cae12d32f77aa5762f 100644 --- a/0856a3f1d05e3aa70d8284d5c67d5f03/run.log +++ b/0856a3f1d05e3aa70d8284d5c67d5f03/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:42:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:42:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/0856a3f1d05e3aa70d8284d5c67d5f03/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'A car drives through a sunny urban city center, navigating multi-lane streets lined with tall buildings, palm trees, and street signs like “1st Ave.” The road is busy with other vehicles, including a white sedan ahead, a white pickup truck, and a silver sedan. Pedestrians walk on sidewalks, and construction barriers appear in later segments. The car stops at traffic lights, then proceeds through intersections, passing storefronts, parking signs, and a “No Turn on Red” sign. Bright midday sunlight illuminates the scene, with clear blue skies overhead. The vehicle maintains a steady pace, avoiding near-misses while maneuvering through the bustling city environment.', '--save_path', '/workspace/video2world_out/0856a3f1d05e3aa70d8284d5c67d5f03/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:42:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='A car drives through a sunny urban city center, navigating multi-lane streets lined with tall buildings, palm trees, and street signs like “1st Ave.” The road is busy with other vehicles, including a white sedan ahead, a white pickup truck, and a silver sedan. Pedestrians walk on sidewalks, and construction barriers appear in later segments. The car stops at traffic lights, then proceeds through intersections, passing storefronts, parking signs, and a “No Turn on Red” sign. Bright midday sunlight illuminates the scene, with clear blue skies overhead. The vehicle maintains a steady pace, avoiding near-misses while maneuvering through the bustling city environment.', input_path='/workspace/video2world_out/0856a3f1d05e3aa70d8284d5c67d5f03/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/0856a3f1d05e3aa70d8284d5c67d5f03/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:42:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:42:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:42:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:42:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/0856a3f1d05e3aa70d8284d5c67d5f03/video2world_2B.yaml. -[01-15 17:42:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:42:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:42:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:42:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:45:42|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:08:47|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/0856a3f1d05e3aa70d8284d5c67d5f03/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'A car drives through a sunny urban city center, navigating multi-lane streets lined with tall buildings, palm trees, and street signs like “1st Ave.” The road is busy with other vehicles, including a white sedan ahead, a white pickup truck, and a silver sedan. Pedestrians walk on sidewalks, and construction barriers appear in later segments. The car stops at traffic lights, then proceeds through intersections, passing storefronts, parking signs, and a “No Turn on Red” sign. Bright midday sunlight illuminates the scene, with clear blue skies overhead. The vehicle maintains a steady pace, avoiding near-misses while maneuvering through the bustling city environment.', '--save_path', '/workspace/video2world_out_14B/0856a3f1d05e3aa70d8284d5c67d5f03/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:08:47|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='A car drives through a sunny urban city center, navigating multi-lane streets lined with tall buildings, palm trees, and street signs like “1st Ave.” The road is busy with other vehicles, including a white sedan ahead, a white pickup truck, and a silver sedan. Pedestrians walk on sidewalks, and construction barriers appear in later segments. The car stops at traffic lights, then proceeds through intersections, passing storefronts, parking signs, and a “No Turn on Red” sign. Bright midday sunlight illuminates the scene, with clear blue skies overhead. The vehicle maintains a steady pace, avoiding near-misses while maneuvering through the bustling city environment.', input_path='/workspace/video2world_out_14B/0856a3f1d05e3aa70d8284d5c67d5f03/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/0856a3f1d05e3aa70d8284d5c67d5f03/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:08:47|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:08:47|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:08:47|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:08:47|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/0856a3f1d05e3aa70d8284d5c67d5f03/video2world_14B.yaml. +[01-15 22:08:47|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:08:47|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:08:49|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:08:49|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:11:06|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/08a0e63da36bd3fade24514337eca65c/run.log b/08a0e63da36bd3fade24514337eca65c/run.log index ae78e31a0b3e9e7ccbd92b5be6c25f1287554387..2ffbed860e92c9366cb39264abb795b191b49bfe 100644 --- a/08a0e63da36bd3fade24514337eca65c/run.log +++ b/08a0e63da36bd3fade24514337eca65c/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:42:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:42:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/08a0e63da36bd3fade24514337eca65c/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a nighttime urban driving scene on a multi - lane road. The road is illuminated by streetlights and vehicle headlights, with dark surroundings indicating it is night. Multiple cars, including a white pickup truck, a silver sedan, and a red SUV, are present, moving slowly in what appears to be heavy traffic. Construction workers in reflective gear and signs like “KEEP RIGHT” suggest lane - related traffic control. Vehicles maintain a slow pace, likely due to congestion or roadwork. No pedestrians or cyclists are visible. The road surface shows signs of wear, and the overall environment is characterized by artificial lighting from streetlights and vehicle lights.', '--save_path', '/workspace/video2world_out/08a0e63da36bd3fade24514337eca65c/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:42:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a nighttime urban driving scene on a multi - lane road. The road is illuminated by streetlights and vehicle headlights, with dark surroundings indicating it is night. Multiple cars, including a white pickup truck, a silver sedan, and a red SUV, are present, moving slowly in what appears to be heavy traffic. Construction workers in reflective gear and signs like “KEEP RIGHT” suggest lane - related traffic control. Vehicles maintain a slow pace, likely due to congestion or roadwork. No pedestrians or cyclists are visible. The road surface shows signs of wear, and the overall environment is characterized by artificial lighting from streetlights and vehicle lights.', input_path='/workspace/video2world_out/08a0e63da36bd3fade24514337eca65c/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/08a0e63da36bd3fade24514337eca65c/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:42:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:42:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:42:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:42:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/08a0e63da36bd3fade24514337eca65c/video2world_2B.yaml. -[01-15 17:42:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:42:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:42:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:42:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:45:39|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:08:47|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/08a0e63da36bd3fade24514337eca65c/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a nighttime urban driving scene on a multi - lane road. The road is illuminated by streetlights and vehicle headlights, with dark surroundings indicating it is night. Multiple cars, including a white pickup truck, a silver sedan, and a red SUV, are present, moving slowly in what appears to be heavy traffic. Construction workers in reflective gear and signs like “KEEP RIGHT” suggest lane - related traffic control. Vehicles maintain a slow pace, likely due to congestion or roadwork. No pedestrians or cyclists are visible. The road surface shows signs of wear, and the overall environment is characterized by artificial lighting from streetlights and vehicle lights.', '--save_path', '/workspace/video2world_out_14B/08a0e63da36bd3fade24514337eca65c/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:08:47|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a nighttime urban driving scene on a multi - lane road. The road is illuminated by streetlights and vehicle headlights, with dark surroundings indicating it is night. Multiple cars, including a white pickup truck, a silver sedan, and a red SUV, are present, moving slowly in what appears to be heavy traffic. Construction workers in reflective gear and signs like “KEEP RIGHT” suggest lane - related traffic control. Vehicles maintain a slow pace, likely due to congestion or roadwork. No pedestrians or cyclists are visible. The road surface shows signs of wear, and the overall environment is characterized by artificial lighting from streetlights and vehicle lights.', input_path='/workspace/video2world_out_14B/08a0e63da36bd3fade24514337eca65c/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/08a0e63da36bd3fade24514337eca65c/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:08:47|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:08:47|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:08:47|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:08:47|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/08a0e63da36bd3fade24514337eca65c/video2world_14B.yaml. +[01-15 22:08:47|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:08:47|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:08:50|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:08:50|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:11:07|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/16b4eef5d7a47892e2342e6a83f5929f/run.log b/16b4eef5d7a47892e2342e6a83f5929f/run.log index af44f45069e670c23d2be01a68d2bb83f4d07d6a..e6975b7b6aeb6a62eb53b8f487b279b2e79cb5b1 100644 --- a/16b4eef5d7a47892e2342e6a83f5929f/run.log +++ b/16b4eef5d7a47892e2342e6a83f5929f/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:42:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:42:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/16b4eef5d7a47892e2342e6a83f5929f/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'A vehicle drives through a sunny urban intersection, starting at a red light before proceeding as the light turns green. The road is a multi-lane street with buildings, trees, and utility poles lining the sides. Several cars are visible, including a black SUV turning left and other vehicles moving straight or parked along the curb. The sky is clear and bright, indicating a sunny day. As the vehicle continues forward, it travels past more parked cars, commercial buildings, and dense foliage, maintaining a steady pace on the relatively straight road. No pedestrians or cyclists are seen, and the scene remains calm with no notable stops or lane changes beyond the initial intersection.', '--save_path', '/workspace/video2world_out/16b4eef5d7a47892e2342e6a83f5929f/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:42:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='A vehicle drives through a sunny urban intersection, starting at a red light before proceeding as the light turns green. The road is a multi-lane street with buildings, trees, and utility poles lining the sides. Several cars are visible, including a black SUV turning left and other vehicles moving straight or parked along the curb. The sky is clear and bright, indicating a sunny day. As the vehicle continues forward, it travels past more parked cars, commercial buildings, and dense foliage, maintaining a steady pace on the relatively straight road. No pedestrians or cyclists are seen, and the scene remains calm with no notable stops or lane changes beyond the initial intersection.', input_path='/workspace/video2world_out/16b4eef5d7a47892e2342e6a83f5929f/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/16b4eef5d7a47892e2342e6a83f5929f/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:42:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:42:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:42:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:42:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/16b4eef5d7a47892e2342e6a83f5929f/video2world_2B.yaml. -[01-15 17:42:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:42:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:42:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:42:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:45:40|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:08:49|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/16b4eef5d7a47892e2342e6a83f5929f/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'A vehicle drives through a sunny urban intersection, starting at a red light before proceeding as the light turns green. The road is a multi-lane street with buildings, trees, and utility poles lining the sides. Several cars are visible, including a black SUV turning left and other vehicles moving straight or parked along the curb. The sky is clear and bright, indicating a sunny day. As the vehicle continues forward, it travels past more parked cars, commercial buildings, and dense foliage, maintaining a steady pace on the relatively straight road. No pedestrians or cyclists are seen, and the scene remains calm with no notable stops or lane changes beyond the initial intersection.', '--save_path', '/workspace/video2world_out_14B/16b4eef5d7a47892e2342e6a83f5929f/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:08:49|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='A vehicle drives through a sunny urban intersection, starting at a red light before proceeding as the light turns green. The road is a multi-lane street with buildings, trees, and utility poles lining the sides. Several cars are visible, including a black SUV turning left and other vehicles moving straight or parked along the curb. The sky is clear and bright, indicating a sunny day. As the vehicle continues forward, it travels past more parked cars, commercial buildings, and dense foliage, maintaining a steady pace on the relatively straight road. No pedestrians or cyclists are seen, and the scene remains calm with no notable stops or lane changes beyond the initial intersection.', input_path='/workspace/video2world_out_14B/16b4eef5d7a47892e2342e6a83f5929f/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/16b4eef5d7a47892e2342e6a83f5929f/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:08:49|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:08:49|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:08:49|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:08:49|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/16b4eef5d7a47892e2342e6a83f5929f/video2world_14B.yaml. +[01-15 22:08:49|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:08:49|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:08:50|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:08:50|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:11:06|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/191ed6f62a0f4d1c23107f8f4939f153/run.log b/191ed6f62a0f4d1c23107f8f4939f153/run.log index 1daa1b7ff8e2ca31793784827640d2fa6974b11f..c9f4e71d261a9e9423d1d2c71085354cd36073f4 100644 --- a/191ed6f62a0f4d1c23107f8f4939f153/run.log +++ b/191ed6f62a0f4d1c23107f8f4939f153/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:42:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:42:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/191ed6f62a0f4d1c23107f8f4939f153/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny day in an urban area, with a vehicle driving along a multi - lane street flanked by tall residential and commercial buildings. Several cars, including a black SUV in the center lane, navigate the road, stopping at red traffic lights and proceeding when they turn green. Pedestrians are visible on sidewalks, and trees line the street, adding greenery to the cityscape. Traffic signs, such as no - parking and street name markers, are present. The scene shows typical city driving dynamics, with vehicles maintaining their lanes, stopping at intersections, and moving through the urban environment without any notable incidents like near - misses.', '--save_path', '/workspace/video2world_out/191ed6f62a0f4d1c23107f8f4939f153/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:42:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny day in an urban area, with a vehicle driving along a multi - lane street flanked by tall residential and commercial buildings. Several cars, including a black SUV in the center lane, navigate the road, stopping at red traffic lights and proceeding when they turn green. Pedestrians are visible on sidewalks, and trees line the street, adding greenery to the cityscape. Traffic signs, such as no - parking and street name markers, are present. The scene shows typical city driving dynamics, with vehicles maintaining their lanes, stopping at intersections, and moving through the urban environment without any notable incidents like near - misses.', input_path='/workspace/video2world_out/191ed6f62a0f4d1c23107f8f4939f153/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/191ed6f62a0f4d1c23107f8f4939f153/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:42:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:42:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:42:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:42:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/191ed6f62a0f4d1c23107f8f4939f153/video2world_2B.yaml. -[01-15 17:42:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:42:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:42:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:42:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:45:36|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:08:48|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/191ed6f62a0f4d1c23107f8f4939f153/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny day in an urban area, with a vehicle driving along a multi - lane street flanked by tall residential and commercial buildings. Several cars, including a black SUV in the center lane, navigate the road, stopping at red traffic lights and proceeding when they turn green. Pedestrians are visible on sidewalks, and trees line the street, adding greenery to the cityscape. Traffic signs, such as no - parking and street name markers, are present. The scene shows typical city driving dynamics, with vehicles maintaining their lanes, stopping at intersections, and moving through the urban environment without any notable incidents like near - misses.', '--save_path', '/workspace/video2world_out_14B/191ed6f62a0f4d1c23107f8f4939f153/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:08:48|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny day in an urban area, with a vehicle driving along a multi - lane street flanked by tall residential and commercial buildings. Several cars, including a black SUV in the center lane, navigate the road, stopping at red traffic lights and proceeding when they turn green. Pedestrians are visible on sidewalks, and trees line the street, adding greenery to the cityscape. Traffic signs, such as no - parking and street name markers, are present. The scene shows typical city driving dynamics, with vehicles maintaining their lanes, stopping at intersections, and moving through the urban environment without any notable incidents like near - misses.', input_path='/workspace/video2world_out_14B/191ed6f62a0f4d1c23107f8f4939f153/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/191ed6f62a0f4d1c23107f8f4939f153/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:08:48|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:08:48|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:08:48|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:08:48|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/191ed6f62a0f4d1c23107f8f4939f153/video2world_14B.yaml. +[01-15 22:08:48|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:08:48|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:08:50|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:08:50|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:11:07|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/1c3d50949247baa54eae139fdddb9c53/run.log b/1c3d50949247baa54eae139fdddb9c53/run.log index a0192bb9993bea877058947768aec39d6d9a1eb2..c72d49ba48d1f0ff127fc2dd296be6c0e0a777ab 100644 --- a/1c3d50949247baa54eae139fdddb9c53/run.log +++ b/1c3d50949247baa54eae139fdddb9c53/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:42:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:42:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/1c3d50949247baa54eae139fdddb9c53/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a multi - lane highway scene during the daytime under partly cloudy skies. Several cars, including sedans and SUVs, travel forward in their respective lanes, with a white sedan often visible in the center lane. The road is flanked by tall trees on the right and open space with scattered vegetation on the left. A green road sign appears briefly on the right side of the highway. There are no pedestrians or cyclists; the only notable activity is the steady flow of traffic with no sudden turns, stops, or lane changes. The lighting is bright, indicating clear daytime conditions, and the overall environment is calm with no near - misses or significant events.', '--save_path', '/workspace/video2world_out/1c3d50949247baa54eae139fdddb9c53/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:42:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a multi - lane highway scene during the daytime under partly cloudy skies. Several cars, including sedans and SUVs, travel forward in their respective lanes, with a white sedan often visible in the center lane. The road is flanked by tall trees on the right and open space with scattered vegetation on the left. A green road sign appears briefly on the right side of the highway. There are no pedestrians or cyclists; the only notable activity is the steady flow of traffic with no sudden turns, stops, or lane changes. The lighting is bright, indicating clear daytime conditions, and the overall environment is calm with no near - misses or significant events.', input_path='/workspace/video2world_out/1c3d50949247baa54eae139fdddb9c53/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/1c3d50949247baa54eae139fdddb9c53/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:42:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:42:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:42:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:42:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/1c3d50949247baa54eae139fdddb9c53/video2world_2B.yaml. -[01-15 17:42:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:42:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:42:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:42:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:45:36|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:08:49|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/1c3d50949247baa54eae139fdddb9c53/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a multi - lane highway scene during the daytime under partly cloudy skies. Several cars, including sedans and SUVs, travel forward in their respective lanes, with a white sedan often visible in the center lane. The road is flanked by tall trees on the right and open space with scattered vegetation on the left. A green road sign appears briefly on the right side of the highway. There are no pedestrians or cyclists; the only notable activity is the steady flow of traffic with no sudden turns, stops, or lane changes. The lighting is bright, indicating clear daytime conditions, and the overall environment is calm with no near - misses or significant events.', '--save_path', '/workspace/video2world_out_14B/1c3d50949247baa54eae139fdddb9c53/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:08:49|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a multi - lane highway scene during the daytime under partly cloudy skies. Several cars, including sedans and SUVs, travel forward in their respective lanes, with a white sedan often visible in the center lane. The road is flanked by tall trees on the right and open space with scattered vegetation on the left. A green road sign appears briefly on the right side of the highway. There are no pedestrians or cyclists; the only notable activity is the steady flow of traffic with no sudden turns, stops, or lane changes. The lighting is bright, indicating clear daytime conditions, and the overall environment is calm with no near - misses or significant events.', input_path='/workspace/video2world_out_14B/1c3d50949247baa54eae139fdddb9c53/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/1c3d50949247baa54eae139fdddb9c53/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:08:49|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:08:49|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:08:49|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:08:49|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/1c3d50949247baa54eae139fdddb9c53/video2world_14B.yaml. +[01-15 22:08:49|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:08:49|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:08:50|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:08:50|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:11:06|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/1dc0dd10757270c34b550f09a383c5f4/run.log b/1dc0dd10757270c34b550f09a383c5f4/run.log index 13aa705e3171c5fccad00797b2c5cfa97d9824d2..263687cc67c36fe23ed4d86422e908a2b94ef9fb 100644 --- a/1dc0dd10757270c34b550f09a383c5f4/run.log +++ b/1dc0dd10757270c34b550f09a383c5f4/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:57:20|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:57:20|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/1dc0dd10757270c34b550f09a383c5f4/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/1dc0dd10757270c34b550f09a383c5f4/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:57:20|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/1dc0dd10757270c34b550f09a383c5f4/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/1dc0dd10757270c34b550f09a383c5f4/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:57:20|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:57:20|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:57:20|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:57:20|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/1dc0dd10757270c34b550f09a383c5f4/video2world_2B.yaml. -[01-15 17:57:20|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:57:20|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:57:21|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:57:21|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:59:39|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:50:11|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/1dc0dd10757270c34b550f09a383c5f4/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/1dc0dd10757270c34b550f09a383c5f4/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:50:11|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/1dc0dd10757270c34b550f09a383c5f4/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/1dc0dd10757270c34b550f09a383c5f4/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:50:11|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:50:11|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:50:11|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:50:11|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/1dc0dd10757270c34b550f09a383c5f4/video2world_14B.yaml. +[01-15 22:50:11|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:50:11|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:50:12|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:50:12|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:53:47|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/2357fa0df9e84b61a1fbe8e6c80efee9/run.log b/2357fa0df9e84b61a1fbe8e6c80efee9/run.log index 965b9a8db5124c5ca072cf0c64e6c9138361578d..afa33af1933c4a9168288a888919ea6e52620133 100644 --- a/2357fa0df9e84b61a1fbe8e6c80efee9/run.log +++ b/2357fa0df9e84b61a1fbe8e6c80efee9/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:57:21|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:57:21|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/2357fa0df9e84b61a1fbe8e6c80efee9/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/2357fa0df9e84b61a1fbe8e6c80efee9/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:57:21|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/2357fa0df9e84b61a1fbe8e6c80efee9/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/2357fa0df9e84b61a1fbe8e6c80efee9/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:57:21|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:57:21|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:57:21|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:57:21|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/2357fa0df9e84b61a1fbe8e6c80efee9/video2world_2B.yaml. -[01-15 17:57:21|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:57:21|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:57:22|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:57:22|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:59:40|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:50:28|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/2357fa0df9e84b61a1fbe8e6c80efee9/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/2357fa0df9e84b61a1fbe8e6c80efee9/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:50:28|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/2357fa0df9e84b61a1fbe8e6c80efee9/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/2357fa0df9e84b61a1fbe8e6c80efee9/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:50:28|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:50:28|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:50:28|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:50:28|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/2357fa0df9e84b61a1fbe8e6c80efee9/video2world_14B.yaml. +[01-15 22:50:28|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:50:28|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:50:29|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:50:29|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:53:48|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/2807b24fc6d3a792ae32f4590ae09698/run.log b/2807b24fc6d3a792ae32f4590ae09698/run.log index e404679961a437dfc76fb870781b3c8e384528d2..c4c2611114de05564147c3fa176f0ba5445e33be 100644 --- a/2807b24fc6d3a792ae32f4590ae09698/run.log +++ b/2807b24fc6d3a792ae32f4590ae09698/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:57:23|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:57:23|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/2807b24fc6d3a792ae32f4590ae09698/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows a car driving through a quiet residential neighborhood on an overcast day. The street is lined with parked cars, palm trees, and multi-story buildings, with a "School Zone" sign visible. As the car moves forward, it passes more parked vehicles, houses, and dense hedges. At an intersection, two pedestrians walk near a stop sign, and a person opens the trunk of a parked SUV. The scene remains calm, with no significant traffic or incidents, and the car continues along the tree-lined street past more residential buildings and parked cars.', '--save_path', '/workspace/video2world_out/2807b24fc6d3a792ae32f4590ae09698/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:57:23|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows a car driving through a quiet residential neighborhood on an overcast day. The street is lined with parked cars, palm trees, and multi-story buildings, with a "School Zone" sign visible. As the car moves forward, it passes more parked vehicles, houses, and dense hedges. At an intersection, two pedestrians walk near a stop sign, and a person opens the trunk of a parked SUV. The scene remains calm, with no significant traffic or incidents, and the car continues along the tree-lined street past more residential buildings and parked cars.', input_path='/workspace/video2world_out/2807b24fc6d3a792ae32f4590ae09698/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/2807b24fc6d3a792ae32f4590ae09698/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:57:23|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:57:23|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:57:23|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:57:23|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/2807b24fc6d3a792ae32f4590ae09698/video2world_2B.yaml. -[01-15 17:57:23|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:57:23|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:57:24|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:57:24|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:59:39|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:50:34|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/2807b24fc6d3a792ae32f4590ae09698/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows a car driving through a quiet residential neighborhood on an overcast day. The street is lined with parked cars, palm trees, and multi-story buildings, with a "School Zone" sign visible. As the car moves forward, it passes more parked vehicles, houses, and dense hedges. At an intersection, two pedestrians walk near a stop sign, and a person opens the trunk of a parked SUV. The scene remains calm, with no significant traffic or incidents, and the car continues along the tree-lined street past more residential buildings and parked cars.', '--save_path', '/workspace/video2world_out_14B/2807b24fc6d3a792ae32f4590ae09698/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:50:34|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows a car driving through a quiet residential neighborhood on an overcast day. The street is lined with parked cars, palm trees, and multi-story buildings, with a "School Zone" sign visible. As the car moves forward, it passes more parked vehicles, houses, and dense hedges. At an intersection, two pedestrians walk near a stop sign, and a person opens the trunk of a parked SUV. The scene remains calm, with no significant traffic or incidents, and the car continues along the tree-lined street past more residential buildings and parked cars.', input_path='/workspace/video2world_out_14B/2807b24fc6d3a792ae32f4590ae09698/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/2807b24fc6d3a792ae32f4590ae09698/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:50:34|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:50:34|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:50:34|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:50:34|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/2807b24fc6d3a792ae32f4590ae09698/video2world_14B.yaml. +[01-15 22:50:34|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:50:34|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:50:35|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:50:35|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:53:45|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/29769d067532845c707ff78a86f55d13/run.log b/29769d067532845c707ff78a86f55d13/run.log index 995d3ecbeb2a670250fd92113d9dcaaddb589002..02b93bbe2fc73d99aabd2a4dd251a1b2d55950de 100644 --- a/29769d067532845c707ff78a86f55d13/run.log +++ b/29769d067532845c707ff78a86f55d13/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:57:33|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:57:33|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/29769d067532845c707ff78a86f55d13/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/29769d067532845c707ff78a86f55d13/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:57:33|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/29769d067532845c707ff78a86f55d13/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/29769d067532845c707ff78a86f55d13/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:57:33|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:57:33|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:57:33|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:57:33|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/29769d067532845c707ff78a86f55d13/video2world_2B.yaml. -[01-15 17:57:33|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:57:33|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:57:34|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:57:34|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:59:40|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:51:02|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/29769d067532845c707ff78a86f55d13/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/29769d067532845c707ff78a86f55d13/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:51:02|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/29769d067532845c707ff78a86f55d13/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/29769d067532845c707ff78a86f55d13/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:51:02|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:51:02|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:51:02|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:51:02|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/29769d067532845c707ff78a86f55d13/video2world_14B.yaml. +[01-15 22:51:02|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:51:02|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:51:03|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:51:03|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:54:15|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/3353933bf837d334f33afcb50ffcef25/run.log b/3353933bf837d334f33afcb50ffcef25/run.log index 0965ba9136ef901d5ba45c56bb04c24891735216..074633e67b43a8d9f55b5e7be10754466cc85449 100644 --- a/3353933bf837d334f33afcb50ffcef25/run.log +++ b/3353933bf837d334f33afcb50ffcef25/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:57:34|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:57:34|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/3353933bf837d334f33afcb50ffcef25/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows a nighttime urban driving scene on a multi - lane road. The environment is dark, illuminated by streetlights, vehicle headlights, and distant building lights. Construction cones and barriers line parts of the road, indicating ongoing roadwork. Several cars are visible ahead, including a silver SUV in the right lane. The driver’s vehicle moves forward through the street, passing under green traffic lights and navigating past construction zones. No pedestrians or cyclists are seen. The road has clear lane markings, and the scene remains consistent with steady forward motion through the city at night.', '--save_path', '/workspace/video2world_out/3353933bf837d334f33afcb50ffcef25/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:57:34|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows a nighttime urban driving scene on a multi - lane road. The environment is dark, illuminated by streetlights, vehicle headlights, and distant building lights. Construction cones and barriers line parts of the road, indicating ongoing roadwork. Several cars are visible ahead, including a silver SUV in the right lane. The driver’s vehicle moves forward through the street, passing under green traffic lights and navigating past construction zones. No pedestrians or cyclists are seen. The road has clear lane markings, and the scene remains consistent with steady forward motion through the city at night.', input_path='/workspace/video2world_out/3353933bf837d334f33afcb50ffcef25/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/3353933bf837d334f33afcb50ffcef25/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:57:34|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:57:34|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:57:34|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:57:34|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/3353933bf837d334f33afcb50ffcef25/video2world_2B.yaml. -[01-15 17:57:34|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:57:34|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:57:35|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:57:35|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:59:41|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:51:02|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/3353933bf837d334f33afcb50ffcef25/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows a nighttime urban driving scene on a multi - lane road. The environment is dark, illuminated by streetlights, vehicle headlights, and distant building lights. Construction cones and barriers line parts of the road, indicating ongoing roadwork. Several cars are visible ahead, including a silver SUV in the right lane. The driver’s vehicle moves forward through the street, passing under green traffic lights and navigating past construction zones. No pedestrians or cyclists are seen. The road has clear lane markings, and the scene remains consistent with steady forward motion through the city at night.', '--save_path', '/workspace/video2world_out_14B/3353933bf837d334f33afcb50ffcef25/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:51:02|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows a nighttime urban driving scene on a multi - lane road. The environment is dark, illuminated by streetlights, vehicle headlights, and distant building lights. Construction cones and barriers line parts of the road, indicating ongoing roadwork. Several cars are visible ahead, including a silver SUV in the right lane. The driver’s vehicle moves forward through the street, passing under green traffic lights and navigating past construction zones. No pedestrians or cyclists are seen. The road has clear lane markings, and the scene remains consistent with steady forward motion through the city at night.', input_path='/workspace/video2world_out_14B/3353933bf837d334f33afcb50ffcef25/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/3353933bf837d334f33afcb50ffcef25/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:51:02|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:51:02|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:51:02|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:51:02|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/3353933bf837d334f33afcb50ffcef25/video2world_14B.yaml. +[01-15 22:51:02|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:51:02|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:51:03|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:51:03|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:54:15|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/36cbc84552200f0982d853310d68506b/run.log b/36cbc84552200f0982d853310d68506b/run.log index 77490b5792290e94e5fbdccb13b457329b73871f..390a739fe80aea147912d02e1ea87ab31cd924f4 100644 --- a/36cbc84552200f0982d853310d68506b/run.log +++ b/36cbc84552200f0982d853310d68506b/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:57:35|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:57:35|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/36cbc84552200f0982d853310d68506b/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny day in a residential neighborhood as a vehicle drives along a two - way street lined with parked cars and colorful houses. A pedestrian crosses the street at a crosswalk early in the sequence. The road is flanked by power lines, and the clear blue sky indicates bright daylight. As the vehicle moves forward, it passes rows of closely packed homes', '--save_path', '/workspace/video2world_out/36cbc84552200f0982d853310d68506b/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:57:35|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny day in a residential neighborhood as a vehicle drives along a two - way street lined with parked cars and colorful houses. A pedestrian crosses the street at a crosswalk early in the sequence. The road is flanked by power lines, and the clear blue sky indicates bright daylight. As the vehicle moves forward, it passes rows of closely packed homes', input_path='/workspace/video2world_out/36cbc84552200f0982d853310d68506b/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/36cbc84552200f0982d853310d68506b/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:57:35|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:57:35|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:57:35|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:57:35|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/36cbc84552200f0982d853310d68506b/video2world_2B.yaml. -[01-15 17:57:35|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:57:35|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:57:36|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:57:36|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:59:43|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:51:03|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/36cbc84552200f0982d853310d68506b/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny day in a residential neighborhood as a vehicle drives along a two - way street lined with parked cars and colorful houses. A pedestrian crosses the street at a crosswalk early in the sequence. The road is flanked by power lines, and the clear blue sky indicates bright daylight. As the vehicle moves forward, it passes rows of closely packed homes', '--save_path', '/workspace/video2world_out_14B/36cbc84552200f0982d853310d68506b/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:51:03|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny day in a residential neighborhood as a vehicle drives along a two - way street lined with parked cars and colorful houses. A pedestrian crosses the street at a crosswalk early in the sequence. The road is flanked by power lines, and the clear blue sky indicates bright daylight. As the vehicle moves forward, it passes rows of closely packed homes', input_path='/workspace/video2world_out_14B/36cbc84552200f0982d853310d68506b/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/36cbc84552200f0982d853310d68506b/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:51:03|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:51:03|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:51:03|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:51:03|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/36cbc84552200f0982d853310d68506b/video2world_14B.yaml. +[01-15 22:51:03|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:51:03|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:51:03|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:51:03|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:54:16|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/37ba76345836e5566857780c1b14b006/run.log b/37ba76345836e5566857780c1b14b006/run.log index e406ef275792d9416547763784b71409562d533d..59f397357dc1bd9e05b320520ecfc5d625b7d04f 100644 --- a/37ba76345836e5566857780c1b14b006/run.log +++ b/37ba76345836e5566857780c1b14b006/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:57:36|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:57:36|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/37ba76345836e5566857780c1b14b006/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a continuous drive through an urban area on a cloudy day. The road is a multi-lane city street lined with parked cars, palm trees, and tall buildings in the background. As the vehicle moves forward, it passes pedestrians near the sidewalk, various parked and moving cars, and a stop sign (though the driver does not stop). The environment features a mix of residential and commercial structures, with construction barriers visible in some sections. The lighting is bright but diffused, indicating overcast conditions. The drive remains steady, with no significant turns, lane changes, or near-misses, focusing on the consistent flow of urban traffic and scenery.', '--save_path', '/workspace/video2world_out/37ba76345836e5566857780c1b14b006/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:57:36|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a continuous drive through an urban area on a cloudy day. The road is a multi-lane city street lined with parked cars, palm trees, and tall buildings in the background. As the vehicle moves forward, it passes pedestrians near the sidewalk, various parked and moving cars, and a stop sign (though the driver does not stop). The environment features a mix of residential and commercial structures, with construction barriers visible in some sections. The lighting is bright but diffused, indicating overcast conditions. The drive remains steady, with no significant turns, lane changes, or near-misses, focusing on the consistent flow of urban traffic and scenery.', input_path='/workspace/video2world_out/37ba76345836e5566857780c1b14b006/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/37ba76345836e5566857780c1b14b006/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:57:36|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:57:36|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:57:36|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:57:36|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/37ba76345836e5566857780c1b14b006/video2world_2B.yaml. -[01-15 17:57:36|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:57:36|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:57:37|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:57:37|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:59:43|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:51:08|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/37ba76345836e5566857780c1b14b006/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a continuous drive through an urban area on a cloudy day. The road is a multi-lane city street lined with parked cars, palm trees, and tall buildings in the background. As the vehicle moves forward, it passes pedestrians near the sidewalk, various parked and moving cars, and a stop sign (though the driver does not stop). The environment features a mix of residential and commercial structures, with construction barriers visible in some sections. The lighting is bright but diffused, indicating overcast conditions. The drive remains steady, with no significant turns, lane changes, or near-misses, focusing on the consistent flow of urban traffic and scenery.', '--save_path', '/workspace/video2world_out_14B/37ba76345836e5566857780c1b14b006/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:51:08|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a continuous drive through an urban area on a cloudy day. The road is a multi-lane city street lined with parked cars, palm trees, and tall buildings in the background. As the vehicle moves forward, it passes pedestrians near the sidewalk, various parked and moving cars, and a stop sign (though the driver does not stop). The environment features a mix of residential and commercial structures, with construction barriers visible in some sections. The lighting is bright but diffused, indicating overcast conditions. The drive remains steady, with no significant turns, lane changes, or near-misses, focusing on the consistent flow of urban traffic and scenery.', input_path='/workspace/video2world_out_14B/37ba76345836e5566857780c1b14b006/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/37ba76345836e5566857780c1b14b006/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:51:08|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:51:08|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:51:08|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:51:08|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/37ba76345836e5566857780c1b14b006/video2world_14B.yaml. +[01-15 22:51:08|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:51:08|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:51:09|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:51:09|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:55:10|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/382435cbc2681dd696589de26a9d73f7/run.log b/382435cbc2681dd696589de26a9d73f7/run.log index fc1aa2faed6c256d68069d95e82b9227d8594dbb..4e6c9c1f5cc03c86d765125bd1498f2dc83542b9 100644 --- a/382435cbc2681dd696589de26a9d73f7/run.log +++ b/382435cbc2681dd696589de26a9d73f7/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 17:57:37|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 17:57:37|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/382435cbc2681dd696589de26a9d73f7/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny urban driving scene along a tree-lined street with mixed residential and commercial buildings. Cars are parked along both sides, and the road features marked bike lanes. A motorcyclist in a red jacket rides through intersections, occasionally stopping at crosswalks where pedestrians also cross. The environment shifts from areas with bare trees to sections with lush green foliage, indicating a mix of seasons or tree types. Traffic flows steadily, with vehicles maintaining their lanes, and a stop sign is visible near the end. The clear blue sky and bright sunlight suggest a calm, pleasant day.', '--save_path', '/workspace/video2world_out/382435cbc2681dd696589de26a9d73f7/video2world_2B.mp4', '--disable_guardrail'] -[01-15 17:57:37|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny urban driving scene along a tree-lined street with mixed residential and commercial buildings. Cars are parked along both sides, and the road features marked bike lanes. A motorcyclist in a red jacket rides through intersections, occasionally stopping at crosswalks where pedestrians also cross. The environment shifts from areas with bare trees to sections with lush green foliage, indicating a mix of seasons or tree types. Traffic flows steadily, with vehicles maintaining their lanes, and a stop sign is visible near the end. The clear blue sky and bright sunlight suggest a calm, pleasant day.', input_path='/workspace/video2world_out/382435cbc2681dd696589de26a9d73f7/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/382435cbc2681dd696589de26a9d73f7/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 17:57:37|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 17:57:37|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 17:57:37|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 17:57:37|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/382435cbc2681dd696589de26a9d73f7/video2world_2B.yaml. -[01-15 17:57:37|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 17:57:37|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 17:57:38|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:57:38|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 17:59:39|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 22:51:12|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/382435cbc2681dd696589de26a9d73f7/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny urban driving scene along a tree-lined street with mixed residential and commercial buildings. Cars are parked along both sides, and the road features marked bike lanes. A motorcyclist in a red jacket rides through intersections, occasionally stopping at crosswalks where pedestrians also cross. The environment shifts from areas with bare trees to sections with lush green foliage, indicating a mix of seasons or tree types. Traffic flows steadily, with vehicles maintaining their lanes, and a stop sign is visible near the end. The clear blue sky and bright sunlight suggest a calm, pleasant day.', '--save_path', '/workspace/video2world_out_14B/382435cbc2681dd696589de26a9d73f7/video2world_14B.mp4', '--disable_guardrail'] +[01-15 22:51:12|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny urban driving scene along a tree-lined street with mixed residential and commercial buildings. Cars are parked along both sides, and the road features marked bike lanes. A motorcyclist in a red jacket rides through intersections, occasionally stopping at crosswalks where pedestrians also cross. The environment shifts from areas with bare trees to sections with lush green foliage, indicating a mix of seasons or tree types. Traffic flows steadily, with vehicles maintaining their lanes, and a stop sign is visible near the end. The clear blue sky and bright sunlight suggest a calm, pleasant day.', input_path='/workspace/video2world_out_14B/382435cbc2681dd696589de26a9d73f7/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/382435cbc2681dd696589de26a9d73f7/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 22:51:12|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 22:51:12|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 22:51:12|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 22:51:12|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/382435cbc2681dd696589de26a9d73f7/video2world_14B.yaml. +[01-15 22:51:12|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 22:51:12|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 22:51:13|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:51:13|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 22:55:09|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/389d3da3535b3d3320bc004466426775/run.log b/389d3da3535b3d3320bc004466426775/run.log index 6f8695813e8454d3013ca0a4dba6b4ecbf93b25d..ec4077e6dfc1e8fcb4da45fb767cfdf1ec2f7e7f 100644 --- a/389d3da3535b3d3320bc004466426775/run.log +++ b/389d3da3535b3d3320bc004466426775/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:10:11|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:10:11|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/389d3da3535b3d3320bc004466426775/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/389d3da3535b3d3320bc004466426775/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:10:11|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/389d3da3535b3d3320bc004466426775/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/389d3da3535b3d3320bc004466426775/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:10:11|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:10:11|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:10:11|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:10:11|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/389d3da3535b3d3320bc004466426775/video2world_2B.yaml. -[01-15 18:10:11|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:10:11|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:10:12|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:10:12|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:12:31|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 23:32:04|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/389d3da3535b3d3320bc004466426775/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/389d3da3535b3d3320bc004466426775/video2world_14B.mp4', '--disable_guardrail'] +[01-15 23:32:04|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/389d3da3535b3d3320bc004466426775/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/389d3da3535b3d3320bc004466426775/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 23:32:04|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 23:32:04|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 23:32:04|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 23:32:04|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/389d3da3535b3d3320bc004466426775/video2world_14B.yaml. +[01-15 23:32:04|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 23:32:04|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 23:32:05|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:32:05|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:35:55|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/3af2f9c169b9bd2c6029e0585815293f/run.log b/3af2f9c169b9bd2c6029e0585815293f/run.log index 306212422ab9b82655ee3f8c20bd78bfeb214d04..41fe151c5bcf8699544fc218982b22661f9b6d56 100644 --- a/3af2f9c169b9bd2c6029e0585815293f/run.log +++ b/3af2f9c169b9bd2c6029e0585815293f/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:10:13|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:10:13|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/3af2f9c169b9bd2c6029e0585815293f/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/3af2f9c169b9bd2c6029e0585815293f/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:10:13|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/3af2f9c169b9bd2c6029e0585815293f/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/3af2f9c169b9bd2c6029e0585815293f/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:10:13|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:10:13|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:10:13|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:10:13|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/3af2f9c169b9bd2c6029e0585815293f/video2world_2B.yaml. -[01-15 18:10:13|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:10:13|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:10:14|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:10:14|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:12:30|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 23:32:54|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/3af2f9c169b9bd2c6029e0585815293f/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/3af2f9c169b9bd2c6029e0585815293f/video2world_14B.mp4', '--disable_guardrail'] +[01-15 23:32:54|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/3af2f9c169b9bd2c6029e0585815293f/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/3af2f9c169b9bd2c6029e0585815293f/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 23:32:54|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 23:32:54|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 23:32:54|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 23:32:54|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/3af2f9c169b9bd2c6029e0585815293f/video2world_14B.yaml. +[01-15 23:32:54|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 23:32:54|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 23:32:55|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:32:55|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:35:58|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/3e14f2dfc6c268c2755dfc48722b0e0d/run.log b/3e14f2dfc6c268c2755dfc48722b0e0d/run.log index 94f3e186331df70f812791da13938ab4079127ab..70edf6f32462987d2ae993175e92c77c0402b023 100644 --- a/3e14f2dfc6c268c2755dfc48722b0e0d/run.log +++ b/3e14f2dfc6c268c2755dfc48722b0e0d/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:10:16|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:10:16|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/3e14f2dfc6c268c2755dfc48722b0e0d/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a car driving through a residential urban area on an overcast day', '--save_path', '/workspace/video2world_out/3e14f2dfc6c268c2755dfc48722b0e0d/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:10:16|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a car driving through a residential urban area on an overcast day', input_path='/workspace/video2world_out/3e14f2dfc6c268c2755dfc48722b0e0d/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/3e14f2dfc6c268c2755dfc48722b0e0d/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:10:16|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:10:16|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:10:16|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:10:17|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/3e14f2dfc6c268c2755dfc48722b0e0d/video2world_2B.yaml. -[01-15 18:10:17|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:10:17|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:10:18|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:10:18|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:12:31|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 23:32:56|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/3e14f2dfc6c268c2755dfc48722b0e0d/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a car driving through a residential urban area on an overcast day', '--save_path', '/workspace/video2world_out_14B/3e14f2dfc6c268c2755dfc48722b0e0d/video2world_14B.mp4', '--disable_guardrail'] +[01-15 23:32:56|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a car driving through a residential urban area on an overcast day', input_path='/workspace/video2world_out_14B/3e14f2dfc6c268c2755dfc48722b0e0d/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/3e14f2dfc6c268c2755dfc48722b0e0d/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 23:32:56|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 23:32:56|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 23:32:56|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 23:32:56|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/3e14f2dfc6c268c2755dfc48722b0e0d/video2world_14B.yaml. +[01-15 23:32:56|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 23:32:56|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 23:32:57|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:32:57|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:35:58|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/46d9bf6665de9c09dc462804a3be9717/run.log b/46d9bf6665de9c09dc462804a3be9717/run.log index 9ccb347b7a980f4d9ac84051270e1e175668c3dd..84aa2170fe74437a63697cd397ddbe52b7418abd 100644 --- a/46d9bf6665de9c09dc462804a3be9717/run.log +++ b/46d9bf6665de9c09dc462804a3be9717/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:10:26|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:10:26|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/46d9bf6665de9c09dc462804a3be9717/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny urban driving scene on a multi - lane road with double yellow lines. The sky is bright blue with scattered white clouds, indicating clear weather. A white car is consistently ahead in the same lane, while other vehicles are parked along the sides or moving in adjacent lanes. Pedestrians are visible on sidewalks, with some walking and one crossing the street. The environment includes modern buildings, palm trees, and streetlights. The driving vehicle maintains a steady pace, following the white car without significant turns or stops. Notable events include a pedestrian crossing the road and various cars parked or moving along the street. The overall atmosphere is calm and typical of a busy city during the day.', '--save_path', '/workspace/video2world_out/46d9bf6665de9c09dc462804a3be9717/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:10:26|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny urban driving scene on a multi - lane road with double yellow lines. The sky is bright blue with scattered white clouds, indicating clear weather. A white car is consistently ahead in the same lane, while other vehicles are parked along the sides or moving in adjacent lanes. Pedestrians are visible on sidewalks, with some walking and one crossing the street. The environment includes modern buildings, palm trees, and streetlights. The driving vehicle maintains a steady pace, following the white car without significant turns or stops. Notable events include a pedestrian crossing the road and various cars parked or moving along the street. The overall atmosphere is calm and typical of a busy city during the day.', input_path='/workspace/video2world_out/46d9bf6665de9c09dc462804a3be9717/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/46d9bf6665de9c09dc462804a3be9717/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:10:26|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:10:26|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:10:26|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:10:26|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/46d9bf6665de9c09dc462804a3be9717/video2world_2B.yaml. -[01-15 18:10:26|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:10:26|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:10:27|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:10:27|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:12:34|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 23:33:07|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/46d9bf6665de9c09dc462804a3be9717/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny urban driving scene on a multi - lane road with double yellow lines. The sky is bright blue with scattered white clouds, indicating clear weather. A white car is consistently ahead in the same lane, while other vehicles are parked along the sides or moving in adjacent lanes. Pedestrians are visible on sidewalks, with some walking and one crossing the street. The environment includes modern buildings, palm trees, and streetlights. The driving vehicle maintains a steady pace, following the white car without significant turns or stops. Notable events include a pedestrian crossing the road and various cars parked or moving along the street. The overall atmosphere is calm and typical of a busy city during the day.', '--save_path', '/workspace/video2world_out_14B/46d9bf6665de9c09dc462804a3be9717/video2world_14B.mp4', '--disable_guardrail'] +[01-15 23:33:07|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny urban driving scene on a multi - lane road with double yellow lines. The sky is bright blue with scattered white clouds, indicating clear weather. A white car is consistently ahead in the same lane, while other vehicles are parked along the sides or moving in adjacent lanes. Pedestrians are visible on sidewalks, with some walking and one crossing the street. The environment includes modern buildings, palm trees, and streetlights. The driving vehicle maintains a steady pace, following the white car without significant turns or stops. Notable events include a pedestrian crossing the road and various cars parked or moving along the street. The overall atmosphere is calm and typical of a busy city during the day.', input_path='/workspace/video2world_out_14B/46d9bf6665de9c09dc462804a3be9717/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/46d9bf6665de9c09dc462804a3be9717/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 23:33:07|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 23:33:07|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 23:33:07|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 23:33:07|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/46d9bf6665de9c09dc462804a3be9717/video2world_14B.yaml. +[01-15 23:33:07|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 23:33:07|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 23:33:08|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:33:08|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:37:01|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/47d66149c4407b0a70d0c926de7fb142/run.log b/47d66149c4407b0a70d0c926de7fb142/run.log index 194857dc18253f5ad4f96805b3bd353e81a62ec9..e47f71c5a836e3b89f6b8dd6c706107ecead3e79 100644 --- a/47d66149c4407b0a70d0c926de7fb142/run.log +++ b/47d66149c4407b0a70d0c926de7fb142/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:10:27|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:10:27|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/47d66149c4407b0a70d0c926de7fb142/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/47d66149c4407b0a70d0c926de7fb142/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:10:27|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/47d66149c4407b0a70d0c926de7fb142/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/47d66149c4407b0a70d0c926de7fb142/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:10:27|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:10:27|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:10:27|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:10:27|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/47d66149c4407b0a70d0c926de7fb142/video2world_2B.yaml. -[01-15 18:10:27|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:10:27|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:10:28|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:10:28|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:12:32|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 23:33:21|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/47d66149c4407b0a70d0c926de7fb142/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/47d66149c4407b0a70d0c926de7fb142/video2world_14B.mp4', '--disable_guardrail'] +[01-15 23:33:21|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/47d66149c4407b0a70d0c926de7fb142/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/47d66149c4407b0a70d0c926de7fb142/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 23:33:21|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 23:33:21|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 23:33:21|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 23:33:21|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/47d66149c4407b0a70d0c926de7fb142/video2world_14B.yaml. +[01-15 23:33:21|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 23:33:21|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 23:33:22|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:33:22|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:37:01|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/599e9d36c5fda7568ad5c37c61052791/run.log b/599e9d36c5fda7568ad5c37c61052791/run.log index b8f6070c24e0c00428e0350bbd55c0994cd42d2f..12d535a48c581811ea697a59a421324862db5786 100644 --- a/599e9d36c5fda7568ad5c37c61052791/run.log +++ b/599e9d36c5fda7568ad5c37c61052791/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:10:29|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:10:29|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/599e9d36c5fda7568ad5c37c61052791/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/599e9d36c5fda7568ad5c37c61052791/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:10:29|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/599e9d36c5fda7568ad5c37c61052791/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/599e9d36c5fda7568ad5c37c61052791/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:10:29|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:10:29|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:10:29|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:10:29|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/599e9d36c5fda7568ad5c37c61052791/video2world_2B.yaml. -[01-15 18:10:29|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:10:29|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:10:30|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:10:30|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:12:31|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 23:33:31|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/599e9d36c5fda7568ad5c37c61052791/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/599e9d36c5fda7568ad5c37c61052791/video2world_14B.mp4', '--disable_guardrail'] +[01-15 23:33:31|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/599e9d36c5fda7568ad5c37c61052791/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/599e9d36c5fda7568ad5c37c61052791/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 23:33:31|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 23:33:31|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 23:33:31|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 23:33:31|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/599e9d36c5fda7568ad5c37c61052791/video2world_14B.yaml. +[01-15 23:33:31|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 23:33:31|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 23:33:32|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:33:32|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:36:59|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/7081b41a970d0ee5585638c76ef686e7/run.log b/7081b41a970d0ee5585638c76ef686e7/run.log index 30e966e232fed25da9cd0312bae817da515ad58f..16d89bed9b5a34d223709798d8bdd65214ba6f64 100644 --- a/7081b41a970d0ee5585638c76ef686e7/run.log +++ b/7081b41a970d0ee5585638c76ef686e7/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:10:28|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:10:28|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/7081b41a970d0ee5585638c76ef686e7/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/7081b41a970d0ee5585638c76ef686e7/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:10:28|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/7081b41a970d0ee5585638c76ef686e7/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/7081b41a970d0ee5585638c76ef686e7/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:10:28|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:10:28|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:10:28|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:10:28|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/7081b41a970d0ee5585638c76ef686e7/video2world_2B.yaml. -[01-15 18:10:28|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:10:28|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:10:29|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:10:29|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:12:31|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 23:33:42|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/7081b41a970d0ee5585638c76ef686e7/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/7081b41a970d0ee5585638c76ef686e7/video2world_14B.mp4', '--disable_guardrail'] +[01-15 23:33:42|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/7081b41a970d0ee5585638c76ef686e7/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/7081b41a970d0ee5585638c76ef686e7/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 23:33:42|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 23:33:42|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 23:33:42|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 23:33:42|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/7081b41a970d0ee5585638c76ef686e7/video2world_14B.yaml. +[01-15 23:33:42|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 23:33:42|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 23:33:43|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:33:43|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:37:00|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/7123a45605c1ca347202c04d645026ba/run.log b/7123a45605c1ca347202c04d645026ba/run.log index 4b7e4d8dd7a913418e25b733b277ab96a57a7bfa..8b6619a7e408738c20764697c04a606e2b83229f 100644 --- a/7123a45605c1ca347202c04d645026ba/run.log +++ b/7123a45605c1ca347202c04d645026ba/run.log @@ -2,30 +2,30 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:10:30|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:10:30|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/7123a45605c1ca347202c04d645026ba/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows', '--save_path', '/workspace/video2world_out/7123a45605c1ca347202c04d645026ba/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:10:30|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows', input_path='/workspace/video2world_out/7123a45605c1ca347202c04d645026ba/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/7123a45605c1ca347202c04d645026ba/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:10:30|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:10:30|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:10:30|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:10:30|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/7123a45605c1ca347202c04d645026ba/video2world_2B.yaml. -[01-15 18:10:30|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:10:30|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:10:31|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:10:31|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:12:31|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-15 23:34:16|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/7123a45605c1ca347202c04d645026ba/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows', '--save_path', '/workspace/video2world_out_14B/7123a45605c1ca347202c04d645026ba/video2world_14B.mp4', '--disable_guardrail'] +[01-15 23:34:16|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows', input_path='/workspace/video2world_out_14B/7123a45605c1ca347202c04d645026ba/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/7123a45605c1ca347202c04d645026ba/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-15 23:34:16|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-15 23:34:16|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-15 23:34:16|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-15 23:34:16|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/7123a45605c1ca347202c04d645026ba/video2world_14B.yaml. +[01-15 23:34:16|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-15 23:34:16|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-15 23:34:17|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:34:17|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-15 23:37:06|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/7124d8f92f3a3c68c0806dc3d00c0143/run.log b/7124d8f92f3a3c68c0806dc3d00c0143/run.log index ba591add3d88e26d8a24524487e5b191743f72e5..14b73c0aeea224205c63cd87600f53c126179ada 100644 --- a/7124d8f92f3a3c68c0806dc3d00c0143/run.log +++ b/7124d8f92f3a3c68c0806dc3d00c0143/run.log @@ -2,30 +2,27 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:23:01|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:23:01|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/7124d8f92f3a3c68c0806dc3d00c0143/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/7124d8f92f3a3c68c0806dc3d00c0143/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:23:01|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/7124d8f92f3a3c68c0806dc3d00c0143/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/7124d8f92f3a3c68c0806dc3d00c0143/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:23:01|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:23:01|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:23:01|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:23:02|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/7124d8f92f3a3c68c0806dc3d00c0143/video2world_2B.yaml. -[01-15 18:23:02|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:23:02|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:23:03|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:23:03|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:25:43|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-16 00:14:12|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/7124d8f92f3a3c68c0806dc3d00c0143/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/7124d8f92f3a3c68c0806dc3d00c0143/video2world_14B.mp4', '--disable_guardrail'] +[01-16 00:14:12|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/7124d8f92f3a3c68c0806dc3d00c0143/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/7124d8f92f3a3c68c0806dc3d00c0143/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-16 00:14:12|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-16 00:14:12|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-16 00:14:12|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-16 00:14:12|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/7124d8f92f3a3c68c0806dc3d00c0143/video2world_14B.yaml. +[01-16 00:14:12|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-16 00:14:12|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-16 00:14:13|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:14:13|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:18:25|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/78ffaf552a9ec88defa802ced636263a/run.log b/78ffaf552a9ec88defa802ced636263a/run.log index 7b2fd745e2298b9c76972f8f0a2f5e591e675099..198db2801e30521b9e8213863cb871ed775fb906 100644 --- a/78ffaf552a9ec88defa802ced636263a/run.log +++ b/78ffaf552a9ec88defa802ced636263a/run.log @@ -2,30 +2,27 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:23:04|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:23:04|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/78ffaf552a9ec88defa802ced636263a/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a vehicle driving through a sunny, residential neighborhood with paved streets, utility poles, and power lines. The road features marked bike lanes, parked cars along both sides, and occasional moving vehicles. The environment includes houses, trees, and clear skies, with no visible pedestrians or cyclists. The vehicle progresses steadily, passing intersections, a "STOP" sign, and a blue truck parked on the side. There are no sharp turns, stops, or near - misses; the scene is calm, with the car maintaining a consistent pace through the neighborhood.', '--save_path', '/workspace/video2world_out/78ffaf552a9ec88defa802ced636263a/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:23:04|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a vehicle driving through a sunny, residential neighborhood with paved streets, utility poles, and power lines. The road features marked bike lanes, parked cars along both sides, and occasional moving vehicles. The environment includes houses, trees, and clear skies, with no visible pedestrians or cyclists. The vehicle progresses steadily, passing intersections, a "STOP" sign, and a blue truck parked on the side. There are no sharp turns, stops, or near - misses; the scene is calm, with the car maintaining a consistent pace through the neighborhood.', input_path='/workspace/video2world_out/78ffaf552a9ec88defa802ced636263a/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/78ffaf552a9ec88defa802ced636263a/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:23:04|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:23:04|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:23:04|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:23:04|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/78ffaf552a9ec88defa802ced636263a/video2world_2B.yaml. -[01-15 18:23:04|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:23:04|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:23:05|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:23:05|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:25:41|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-16 00:15:05|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/78ffaf552a9ec88defa802ced636263a/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a vehicle driving through a sunny, residential neighborhood with paved streets, utility poles, and power lines. The road features marked bike lanes, parked cars along both sides, and occasional moving vehicles. The environment includes houses, trees, and clear skies, with no visible pedestrians or cyclists. The vehicle progresses steadily, passing intersections, a "STOP" sign, and a blue truck parked on the side. There are no sharp turns, stops, or near - misses; the scene is calm, with the car maintaining a consistent pace through the neighborhood.', '--save_path', '/workspace/video2world_out_14B/78ffaf552a9ec88defa802ced636263a/video2world_14B.mp4', '--disable_guardrail'] +[01-16 00:15:05|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a vehicle driving through a sunny, residential neighborhood with paved streets, utility poles, and power lines. The road features marked bike lanes, parked cars along both sides, and occasional moving vehicles. The environment includes houses, trees, and clear skies, with no visible pedestrians or cyclists. The vehicle progresses steadily, passing intersections, a "STOP" sign, and a blue truck parked on the side. There are no sharp turns, stops, or near - misses; the scene is calm, with the car maintaining a consistent pace through the neighborhood.', input_path='/workspace/video2world_out_14B/78ffaf552a9ec88defa802ced636263a/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/78ffaf552a9ec88defa802ced636263a/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-16 00:15:05|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-16 00:15:05|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-16 00:15:05|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-16 00:15:05|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/78ffaf552a9ec88defa802ced636263a/video2world_14B.yaml. +[01-16 00:15:05|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-16 00:15:05|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-16 00:15:06|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:15:06|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:18:27|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/86887df5cd0c0a1610f0a1dd6d18171a/run.log b/86887df5cd0c0a1610f0a1dd6d18171a/run.log index c031bfd2181d0dd83d5a2b0eb5b42ee9cdd3ba48..3328be056720105c93909d0273e2111777556e56 100644 --- a/86887df5cd0c0a1610f0a1dd6d18171a/run.log +++ b/86887df5cd0c0a1610f0a1dd6d18171a/run.log @@ -2,30 +2,27 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:23:06|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:23:06|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/86887df5cd0c0a1610f0a1dd6d18171a/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a nighttime urban driving scene on a multi - lane road illuminated by streetlights. The weather appears clear, with no signs of rain or adverse conditions. As the vehicle moves forward, it passes through a construction zone marked by orange - and - white barriers and traffic cones. Several cars are visible ahead, some with headlights on,', '--save_path', '/workspace/video2world_out/86887df5cd0c0a1610f0a1dd6d18171a/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:23:06|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a nighttime urban driving scene on a multi - lane road illuminated by streetlights. The weather appears clear, with no signs of rain or adverse conditions. As the vehicle moves forward, it passes through a construction zone marked by orange - and - white barriers and traffic cones. Several cars are visible ahead, some with headlights on,', input_path='/workspace/video2world_out/86887df5cd0c0a1610f0a1dd6d18171a/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/86887df5cd0c0a1610f0a1dd6d18171a/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:23:06|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:23:06|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:23:06|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:23:06|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/86887df5cd0c0a1610f0a1dd6d18171a/video2world_2B.yaml. -[01-15 18:23:06|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:23:06|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:23:07|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:23:07|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:25:42|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-16 00:15:12|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/86887df5cd0c0a1610f0a1dd6d18171a/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a nighttime urban driving scene on a multi - lane road illuminated by streetlights. The weather appears clear, with no signs of rain or adverse conditions. As the vehicle moves forward, it passes through a construction zone marked by orange - and - white barriers and traffic cones. Several cars are visible ahead, some with headlights on,', '--save_path', '/workspace/video2world_out_14B/86887df5cd0c0a1610f0a1dd6d18171a/video2world_14B.mp4', '--disable_guardrail'] +[01-16 00:15:12|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a nighttime urban driving scene on a multi - lane road illuminated by streetlights. The weather appears clear, with no signs of rain or adverse conditions. As the vehicle moves forward, it passes through a construction zone marked by orange - and - white barriers and traffic cones. Several cars are visible ahead, some with headlights on,', input_path='/workspace/video2world_out_14B/86887df5cd0c0a1610f0a1dd6d18171a/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/86887df5cd0c0a1610f0a1dd6d18171a/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-16 00:15:12|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-16 00:15:12|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-16 00:15:12|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-16 00:15:12|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/86887df5cd0c0a1610f0a1dd6d18171a/video2world_14B.yaml. +[01-16 00:15:12|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-16 00:15:12|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-16 00:15:13|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:15:13|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:18:27|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/91ecee2ae1306ed10f90e9646fb352c7/run.log b/91ecee2ae1306ed10f90e9646fb352c7/run.log index 17edb39c9037eb101f0a3419687e3a8bcaf41e4b..e9c3a85daef43c0dd408a916bc6a467ece6fa5ea 100644 --- a/91ecee2ae1306ed10f90e9646fb352c7/run.log +++ b/91ecee2ae1306ed10f90e9646fb352c7/run.log @@ -2,30 +2,27 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:23:18|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:23:18|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/91ecee2ae1306ed10f90e9646fb352c7/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows a vehicle driving along a sunny urban street lined with tall palm trees under a bright blue sky with scattered clouds. The road is a multi - lane paved street with clear lane markings', '--save_path', '/workspace/video2world_out/91ecee2ae1306ed10f90e9646fb352c7/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:23:18|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows a vehicle driving along a sunny urban street lined with tall palm trees under a bright blue sky with scattered clouds. The road is a multi - lane paved street with clear lane markings', input_path='/workspace/video2world_out/91ecee2ae1306ed10f90e9646fb352c7/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/91ecee2ae1306ed10f90e9646fb352c7/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:23:18|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:23:18|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:23:18|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:23:18|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/91ecee2ae1306ed10f90e9646fb352c7/video2world_2B.yaml. -[01-15 18:23:18|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:23:18|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:23:19|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:23:19|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:25:41|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-16 00:15:24|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/91ecee2ae1306ed10f90e9646fb352c7/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows a vehicle driving along a sunny urban street lined with tall palm trees under a bright blue sky with scattered clouds. The road is a multi - lane paved street with clear lane markings', '--save_path', '/workspace/video2world_out_14B/91ecee2ae1306ed10f90e9646fb352c7/video2world_14B.mp4', '--disable_guardrail'] +[01-16 00:15:24|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows a vehicle driving along a sunny urban street lined with tall palm trees under a bright blue sky with scattered clouds. The road is a multi - lane paved street with clear lane markings', input_path='/workspace/video2world_out_14B/91ecee2ae1306ed10f90e9646fb352c7/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/91ecee2ae1306ed10f90e9646fb352c7/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-16 00:15:24|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-16 00:15:24|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-16 00:15:24|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-16 00:15:24|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/91ecee2ae1306ed10f90e9646fb352c7/video2world_14B.yaml. +[01-16 00:15:24|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-16 00:15:24|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-16 00:15:25|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:15:25|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:19:01|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/95b95c178892347aeb2e248658946e89/run.log b/95b95c178892347aeb2e248658946e89/run.log index 27dfd55cb504b40ec0bf332051793e7503dc8086..6d73e2fb5859b3f87da04bbcd9db98d39d2840d3 100644 --- a/95b95c178892347aeb2e248658946e89/run.log +++ b/95b95c178892347aeb2e248658946e89/run.log @@ -2,30 +2,27 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:23:19|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:23:19|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/95b95c178892347aeb2e248658946e89/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/95b95c178892347aeb2e248658946e89/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:23:19|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/95b95c178892347aeb2e248658946e89/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/95b95c178892347aeb2e248658946e89/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:23:19|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:23:19|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:23:19|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:23:19|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/95b95c178892347aeb2e248658946e89/video2world_2B.yaml. -[01-15 18:23:19|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:23:19|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:23:20|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:23:20|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:25:42|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-16 00:15:38|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/95b95c178892347aeb2e248658946e89/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/95b95c178892347aeb2e248658946e89/video2world_14B.mp4', '--disable_guardrail'] +[01-16 00:15:38|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/95b95c178892347aeb2e248658946e89/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/95b95c178892347aeb2e248658946e89/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-16 00:15:38|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-16 00:15:38|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-16 00:15:38|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-16 00:15:38|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/95b95c178892347aeb2e248658946e89/video2world_14B.yaml. +[01-16 00:15:38|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-16 00:15:38|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-16 00:15:39|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:15:39|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:18:57|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/9c1ff2db624283f075f2aea01392c737/run.log b/9c1ff2db624283f075f2aea01392c737/run.log index 6a4c2dbc343eea5dcc5e5ee697f29db7e3d28b8f..eae96b47185ef827b0e741586ca6b57547441b01 100644 --- a/9c1ff2db624283f075f2aea01392c737/run.log +++ b/9c1ff2db624283f075f2aea01392c737/run.log @@ -2,30 +2,27 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:23:20|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:23:20|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/9c1ff2db624283f075f2aea01392c737/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out/9c1ff2db624283f075f2aea01392c737/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:23:20|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out/9c1ff2db624283f075f2aea01392c737/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/9c1ff2db624283f075f2aea01392c737/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:23:20|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:23:20|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:23:20|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:23:20|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/9c1ff2db624283f075f2aea01392c737/video2world_2B.yaml. -[01-15 18:23:20|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:23:20|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:23:21|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:23:21|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:25:41|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-16 00:16:05|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/9c1ff2db624283f075f2aea01392c737/input0.jpg', '--num_conditional_frames', '1', '--prompt', '[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', '--save_path', '/workspace/video2world_out_14B/9c1ff2db624283f075f2aea01392c737/video2world_14B.mp4', '--disable_guardrail'] +[01-16 00:16:05|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='[ERROR] RuntimeError: Empty response from model. Hint: {"done": true, "done_reason": "length", "eval_count": 350}', input_path='/workspace/video2world_out_14B/9c1ff2db624283f075f2aea01392c737/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/9c1ff2db624283f075f2aea01392c737/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-16 00:16:05|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-16 00:16:05|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-16 00:16:05|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-16 00:16:05|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/9c1ff2db624283f075f2aea01392c737/video2world_14B.yaml. +[01-16 00:16:05|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-16 00:16:05|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-16 00:16:06|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:16:06|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:18:57|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/9c737484c5fc491d2084fd88cc2deaa7/run.log b/9c737484c5fc491d2084fd88cc2deaa7/run.log index 9d5f3bc219cf798e2136fc5780d5be6b7cb15240..e8d0b671031ad3b092ae7b425a5875807c393527 100644 --- a/9c737484c5fc491d2084fd88cc2deaa7/run.log +++ b/9c737484c5fc491d2084fd88cc2deaa7/run.log @@ -2,30 +2,27 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:23:20|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:23:20|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/9c737484c5fc491d2084fd88cc2deaa7/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny day driving through an urban street lined with multi-story buildings. The road is a multi-lane city street with parked cars along both sides and moving vehicles ahead. A white pickup truck is visible in the foreground, while other cars travel in the same direction. Pedestrians walk on the sidewalks, and a bike rack with numerous bicycles is seen on the right. The scene includes a "20" speed limit sign, indicating a residential or commercial area. The driver proceeds through intersections with green traffic lights, passing various buildings, trees, and streetlights. No significant turns or stops occur; the focus is on continuous forward motion through the cityscape.', '--save_path', '/workspace/video2world_out/9c737484c5fc491d2084fd88cc2deaa7/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:23:20|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny day driving through an urban street lined with multi-story buildings. The road is a multi-lane city street with parked cars along both sides and moving vehicles ahead. A white pickup truck is visible in the foreground, while other cars travel in the same direction. Pedestrians walk on the sidewalks, and a bike rack with numerous bicycles is seen on the right. The scene includes a "20" speed limit sign, indicating a residential or commercial area. The driver proceeds through intersections with green traffic lights, passing various buildings, trees, and streetlights. No significant turns or stops occur; the focus is on continuous forward motion through the cityscape.', input_path='/workspace/video2world_out/9c737484c5fc491d2084fd88cc2deaa7/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/9c737484c5fc491d2084fd88cc2deaa7/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:23:20|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:23:20|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:23:20|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:23:20|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/9c737484c5fc491d2084fd88cc2deaa7/video2world_2B.yaml. -[01-15 18:23:20|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:23:20|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:23:21|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:23:21|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:25:41|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-16 00:16:07|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/9c737484c5fc491d2084fd88cc2deaa7/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video depicts a sunny day driving through an urban street lined with multi-story buildings. The road is a multi-lane city street with parked cars along both sides and moving vehicles ahead. A white pickup truck is visible in the foreground, while other cars travel in the same direction. Pedestrians walk on the sidewalks, and a bike rack with numerous bicycles is seen on the right. The scene includes a "20" speed limit sign, indicating a residential or commercial area. The driver proceeds through intersections with green traffic lights, passing various buildings, trees, and streetlights. No significant turns or stops occur; the focus is on continuous forward motion through the cityscape.', '--save_path', '/workspace/video2world_out_14B/9c737484c5fc491d2084fd88cc2deaa7/video2world_14B.mp4', '--disable_guardrail'] +[01-16 00:16:07|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video depicts a sunny day driving through an urban street lined with multi-story buildings. The road is a multi-lane city street with parked cars along both sides and moving vehicles ahead. A white pickup truck is visible in the foreground, while other cars travel in the same direction. Pedestrians walk on the sidewalks, and a bike rack with numerous bicycles is seen on the right. The scene includes a "20" speed limit sign, indicating a residential or commercial area. The driver proceeds through intersections with green traffic lights, passing various buildings, trees, and streetlights. No significant turns or stops occur; the focus is on continuous forward motion through the cityscape.', input_path='/workspace/video2world_out_14B/9c737484c5fc491d2084fd88cc2deaa7/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/9c737484c5fc491d2084fd88cc2deaa7/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-16 00:16:07|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-16 00:16:07|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-16 00:16:07|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-16 00:16:07|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/9c737484c5fc491d2084fd88cc2deaa7/video2world_14B.yaml. +[01-16 00:16:07|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-16 00:16:07|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-16 00:16:08|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:16:08|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:19:00|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth diff --git a/a06d3015e09c279006007d6b0be0f83f/run.log b/a06d3015e09c279006007d6b0be0f83f/run.log index d1f97aed8e5c2a50af6ef49ad2fb1009bab844da..9a2204d7cc20627acaf8cfb47893453d66ffb605 100644 --- a/a06d3015e09c279006007d6b0be0f83f/run.log +++ b/a06d3015e09c279006007d6b0be0f83f/run.log @@ -2,30 +2,27 @@ fatal: detected dubious ownership in repository at '/workspace' To add an exception for this directory, call: git config --global --add safe.directory /workspace -[01-15 18:23:22|INFO|imaginaire/constants.py:39:print_environment_info] imaginaire.constants: Namespace(checkpoints='checkpoints', text_encoder=) -[01-15 18:23:22|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '2B', '--input_path', '/workspace/video2world_out/a06d3015e09c279006007d6b0be0f83f/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows a car driving through a sunny urban area with clear blue skies. The road is lined with palm trees and a mix of commercial and office buildings, with parked cars along both sides', '--save_path', '/workspace/video2world_out/a06d3015e09c279006007d6b0be0f83f/video2world_2B.mp4', '--disable_guardrail'] -[01-15 18:23:22|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='2B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows a car driving through a sunny urban area with clear blue skies. The road is lined with palm trees and a mix of commercial and office buildings, with parked cars along both sides', input_path='/workspace/video2world_out/a06d3015e09c279006007d6b0be0f83f/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out/a06d3015e09c279006007d6b0be0f83f/video2world_2B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) -[01-15 18:23:22|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/model-720p-16fps.pt -[01-15 18:23:22|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. -[01-15 18:23:22|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled -[01-15 18:23:22|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out/a06d3015e09c279006007d6b0be0f83f/video2world_2B.yaml. -[01-15 18:23:22|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 2B -[01-15 18:23:22|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 -[01-15 18:23:23|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:23:23|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-2B-Video2World/tokenizer/tokenizer.pth -[01-15 18:28:45|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated - Loading checkpoint shards: 0%| | 0/4 [00:00) +[01-16 00:16:15|INFO|imaginaire/constants.py:40:print_environment_info] sys.argv: ['/workspace/examples/video2world.py', '--model_size', '14B', '--input_path', '/workspace/video2world_out_14B/a06d3015e09c279006007d6b0be0f83f/input0.jpg', '--num_conditional_frames', '1', '--prompt', 'The video shows a car driving through a sunny urban area with clear blue skies. The road is lined with palm trees and a mix of commercial and office buildings, with parked cars along both sides', '--save_path', '/workspace/video2world_out_14B/a06d3015e09c279006007d6b0be0f83f/video2world_14B.mp4', '--disable_guardrail'] +[01-16 00:16:15|INFO|imaginaire/constants.py:41:print_environment_info] args: Namespace(model_size='14B', resolution='720', fps=16, dit_path='', load_ema=False, prompt='The video shows a car driving through a sunny urban area with clear blue skies. The road is lined with palm trees and a mix of commercial and office buildings, with parked cars along both sides', input_path='/workspace/video2world_out_14B/a06d3015e09c279006007d6b0be0f83f/input0.jpg', negative_prompt='The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', aspect_ratio='16:9', num_conditional_frames=1, batch_input_json=None, guidance=7, seed=0, save_path='/workspace/video2world_out_14B/a06d3015e09c279006007d6b0be0f83f/video2world_14B.mp4', num_gpus=1, disable_guardrail=True, offload_guardrail=False, disable_prompt_refiner=False, offload_prompt_refiner=False, offload_text_encoder=False, downcast_text_encoder=False, benchmark=False, use_cuda_graphs=False, natten=False) +[01-16 00:16:15|INFO|examples/video2world.py:210:setup_pipeline] Using dit_path: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/model-720p-16fps.pt +[01-16 00:16:15|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0. +[01-16 00:16:15|WARNING|examples/video2world.py:241:setup_pipeline] Guardrail checks are disabled +[01-16 00:16:15|WARNING|imaginaire/lazy_config/lazy.py:441:save_yaml] Config is saved using omegaconf at /workspace/video2world_out_14B/a06d3015e09c279006007d6b0be0f83f/video2world_14B.yaml. +[01-16 00:16:15|INFO|examples/video2world.py:259:setup_pipeline] Initializing Video2WorldPipeline with model size: 14B +[01-16 00:16:15|WARNING|cosmos_predict2/pipelines/video2world.py:292:from_config] precision torch.bfloat16 +[01-16 00:16:16|INFO|cosmos_predict2/tokenizers/tokenizer.py:599:_video_vae] Loading checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:16:16|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:601:_video_vae] Successfully loaded checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth +[01-16 00:21:27|INFO|imaginaire/auxiliary/text_encoder.py:345:__init__] T5 Text encoder model instantiated + Loading checkpoint shards: 0%| | 0/4 [00:00 + fps: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: fps + output_key: fps + padding_mask: + _target_: + dropout_rate: '0.0' + dtype: null + input_key: padding_mask + output_key: padding_mask + text: + _target_: + dropout_rate: '0.2' + input_key: + - t5_text_embeddings + use_video_condition: + _target_: + dropout_rate: '0.0' + input_key: fps + output_key: use_video_condition +conditioning_strategy: frame_replace +ema: + _target_: cosmos_predict2.configs.base.defaults.ema.EMAConfig + enabled: 'False' + iteration_shift: '0' + rate: '0.1' +guardrail_config: + checkpoint_dir: checkpoints + enabled: 'False' + offload_model_to_cpu: 'False' +input_image_key: images +input_video_key: video +max_num_conditional_frames: '2' +min_num_conditional_frames: '1' +net: + _target_: + adaln_lora_dim: '256' + atten_backend: minimal_a2a + concat_padding_mask: 'True' + extra_per_block_abs_pos_emb: 'False' + in_channels: '16' + max_frames: '128' + max_img_h: '240' + max_img_w: '240' + model_channels: '5120' + num_blocks: '36' + num_heads: '40' + out_channels: '16' + patch_spatial: '2' + patch_temporal: '1' + pos_emb_cls: rope3d + pos_emb_interpolation: crop + pos_emb_learnable: 'True' + rope_enable_fps_modulation: 'False' + rope_h_extrapolation_ratio: '2.0' + rope_t_extrapolation_ratio: '0.8333333333333334' + rope_w_extrapolation_ratio: '2.0' + sac_config: + _target_: cosmos_predict2.models.text2image_dit.SACConfig + every_n_blocks: '1' + mode: predict2_14b_720 + use_adaln_lora: 'True' +precision: bfloat16 +prompt_refiner_config: + checkpoint_dir: checkpoints/nvidia/Cosmos-Reason1-7B + enabled: 'True' + offload_model_to_cpu: 'False' +rectified_flow_loss_weight_uniform: 'True' +rectified_flow_t_scaling_factor: '1.0' +resize_online: 'True' +resolution: '720' +sigma_conditional: '0.0001' +sigma_data: '1.0' +state_ch: '16' +state_t: '24' +text_encoder: + cls: TextEncoderClass.T5 + cosmos_reason1: + ckpt_path: checkpoints/nvidia/Cosmos-Reason1-Private/reason1_internal_real.pt + compute_online: 'True' + embed_dim: '100352' + embedding_concat_strategy: full_concat + model_config: + _target_: + model_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenModelConfig + activation_checkpoint: + mode: selective + models: vlm + selective_ac_option: op + add_answer_tag: 'True' + add_cross_attention: 'False' + add_image_start_end_tag: 'False' + add_tile_tag: 'False' + architectures: + - Qwen2_5_VLForConditionalGeneration + attention_dropout: '0.0' + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + aux_loss_coeff: '0.0' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: '151643' + cache_dir: null + checkpoint: + async_mode: disabled + create_seed_checkpoint: false + enable_checkpoint: false + export_dtype: float32 + folder: checkpoint + interval: 500 + interval_type: steps + model_weights_only: false + chunk_size_feed_forward: '0' + ckpt_dir: null + ckpt_path: null + comm: + init_timeout_seconds: 300 + trace_buf_size: 20000 + train_timeout_seconds: 100 + cp_size: null + cross_attention_hidden_size: null + decoder_start_token_id: null + deterministic: 'False' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + encoder_no_repeat_ngram_size: '0' + eos_token_id: '151645' + ep_size: null + experimental: + enable_async_tensor_parallel: false + enable_compiled_autograd: false + pipeline_parallel_degree: 1 + exponential_decay_length_penalty: null + finetuning_task: null + float8: + enable_float8_linear: false + forced_bos_token_id: null + forced_eos_token_id: null + freeze_llm: 'False' + freeze_mm_projector: 'False' + freeze_vision_encoder: 'False' + fsdp_enabled: 'False' + hidden_act: silu + hidden_size: '3584' + id2label: + 0: LABEL_0 + 1: LABEL_1 + image_token_id: '151655' + initializer_range: '0.02' + intermediate_size: '18944' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + loss_per_token: 'True' + max_batch_size: '1' + max_length: '20' + max_position_embeddings: '128000' + max_seq_len: '128000' + max_window_layers: '28' + min_length: '0' + mm_projector: null + model_type: qwen2_5_vl + name_or_path: Qwen/Qwen2.5-VL-7B-Instruct + no_repeat_ngram_size: '0' + num_attention_heads: '28' + num_beam_groups: '1' + num_beams: '1' + num_hidden_layers: '28' + num_key_value_heads: '4' + num_return_sequences: '1' + num_tiles: '1' + optimizer: + early_step_in_backward: false + end_lr: 2.5e-05 + fused: false + init_lr: 1.0e-05 + lr: 0.0003 + lr_multiplier_llm: 1.0 + lr_multiplier_mm_projector: 1.0 + lr_multiplier_vision_encoder: 0.1 + name: AdamW + output_attentions: 'False' + output_hidden_states: 'True' + output_scores: 'False' + pad_token_id: null + precision: bfloat16 + prefix: null + prepend_padding: 'False' + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + rms_norm_eps: 1e-06 + rope_scaling: + mrope_section: + - '16' + - '24' + - '24' + rope_type: default + type: default + rope_theta: '1000000.0' + seed: '0' + sep_token_id: null + sliding_window: '32768' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'False' + tile_tag_type: space_separated + tokenizer_class: null + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + training: + compile: false + context_parallel_degree: 1 + data_parallel_replicate_degree: 1 + data_parallel_shard_degree: -1 + disable_loss_parallel: false + enable_cpu_offload: false + fsdp_reshard_after_forward: default + mixed_precision_param: bfloat16 + mixed_precision_reduce: float32 + steps: 400000 + tensor_parallel_degree: 1 + use_cosine_decay: false + use_linear_decay: true + warmup_steps: 1000 + training_seq_len: '4096' + transformers_version: 4.51.0.dev0 + typical_p: '1.0' + use_bfloat16: 'False' + use_cache: 'False' + use_fsdp2: 'True' + use_return_dict: 'True' + use_rope_from_torchtitan: 'False' + use_sliding_window: 'False' + video_token_id: '151656' + vision_config: + _target_: imaginaire.configs.reason1.model_config_qwen.QwenVisionConfig + add_cross_attention: 'False' + architectures: null + attn_implementation: flash_attention_2 + attn_implementation_autoset: 'True' + bad_words_ids: null + begin_suppress_tokens: null + bos_token_id: null + chunk_size_feed_forward: '0' + cross_attention_hidden_size: null + decoder_start_token_id: null + depth: '32' + diversity_penalty: '0.0' + do_sample: 'False' + early_stopping: 'False' + embed_dim: null + encoder_no_repeat_ngram_size: '0' + eos_token_id: null + exponential_decay_length_penalty: null + finetuning_task: null + forced_bos_token_id: null + forced_eos_token_id: null + fullatt_block_indexes: + - '7' + - '15' + - '23' + - '31' + hidden_act: silu + hidden_size: '1280' + id2label: + 0: LABEL_0 + 1: LABEL_1 + in_channels: '3' + in_chans: '3' + intermediate_size: '3420' + is_decoder: 'False' + is_encoder_decoder: 'False' + label2id: + LABEL_0: '0' + LABEL_1: '1' + length_penalty: '1.0' + max_length: '20' + min_length: '0' + mlp_ratio: null + model_type: qwen2_5_vl + name_or_path: '' + no_repeat_ngram_size: '0' + num_beam_groups: '1' + num_beams: '1' + num_heads: '16' + num_return_sequences: '1' + out_hidden_size: '3584' + output_attentions: 'False' + output_hidden_states: 'False' + output_scores: 'False' + pad_token_id: null + patch_size: '14' + prefix: null + problem_type: null + pruned_heads: _Nothing.NOTHING + remove_invalid_values: 'False' + repetition_penalty: '1.0' + return_dict: 'True' + return_dict_in_generate: 'False' + sep_token_id: null + spatial_merge_size: '2' + spatial_patch_size: '14' + suppress_tokens: null + task_specific_params: null + temperature: '1.0' + temporal_patch_size: '2' + tf_legacy_loss: 'False' + tie_encoder_decoder: 'False' + tie_word_embeddings: 'True' + tokenizer_class: null + tokens_per_second: '2' + top_k: '50' + top_p: '1.0' + torch_dtype: bfloat16 + torchscript: 'False' + typical_p: '1.0' + use_bfloat16: 'False' + window_size: '112' + vision_encoder: openai/clip-vit-base-patch32 + vision_encoder_config: + depth_init: true + dim: 1024 + ffn_dim_multiplier: null + head_dim: null + hidden_act: null + hidden_dim: 4096 + image_size: 1024 + image_token_id: null + multiple_of: null + n_heads: 16 + n_kv_heads: null + n_layers: 24 + norm_eps: 1.0e-05 + norm_type: rmsnorm + num_channels: 3 + patch_size: 16 + proj_bias: null + qkv_bias: null + rope_theta: 10000.0 + use_cache: false + use_rope_from_torchtitan: false + vision_encoder_in_channels: '3' + vision_end_token_id: '151653' + vision_start_token_id: '151652' + vision_token_id: '151654' + vocab_size: '152064' + z_loss_coeff: '0.0' + tokenizer: + _target_: + cache_dir: checkpoints/nvidia/Cosmos-Reason1-Private/tokenizer + tokenizer_type: Qwen/Qwen2.5-VL-7B-Instruct + n_layers_per_group: '5' + num_tokens: '512' + t5: + ckpt_path: checkpoints/google-t5/t5-11b + embed_dim: '1024' + num_tokens: '512' +timestamps: + is_forward: 'False' + nfe: '35' + order: '7.0' + t_max: '80.0' + t_min: '0.002' +tokenizer: + _target_: + chunk_duration: '81' + load_mean_std: 'False' + name: tokenizer + temporal_window: '16' + vae_pth: checkpoints/nvidia/Cosmos-Predict2-14B-Video2World/tokenizer/tokenizer.pth