Hugging Face's logo

push-that-thing
/
model

Model card Files
xet
 Community
model
File size: 92,794 Bytes
cbfefc3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
[05-07 17:16:59|INFO|imaginaire/trainer.py:116:__init__] Config:
* model: {'config': {'train_architecture': 'lora', 'lora_rank': 32, 'lora_alpha': 32, 'lora_target_modules': 'q_proj,k_proj,v_proj,output_proj,x_embedder.proj.1,linear_1,linear_2,mlp.layer1,mlp.layer2', 'init_lora_weights': True, 'precision': 'bfloat16', 'input_video_key': 'video', 'input_image_key': 'images', 'loss_reduce': 'mean', 'loss_scale': 100.0, 'adjust_video_noise': True, 'model_manager_config': {'dit_path': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt', 'text_encoder_path': '', '_target_': 'cosmos_predict2.models.video2world_model.Predict2ModelManagerConfig'}, 'pipe_config': {'adjust_video_noise': True, 'conditioner': {'fps': {'output_key': 'fps', 'dropout_rate': 0.0, 'dtype': None, 'input_key': 'fps', '_target_': <class 'cosmos_predict2.conditioner.ReMapkey'>}, 'padding_mask': {'output_key': 'padding_mask', 'dropout_rate': 0.0, 'dtype': None, 'input_key': 'padding_mask', '_target_': <class 'cosmos_predict2.conditioner.ReMapkey'>}, 'text': {'dropout_rate': 0.0, 'input_key': ['obs/language_embedding'], '_target_': <class 'cosmos_predict2.conditioner.TextAttr'>}, 'use_video_condition': {'output_key': 'use_video_condition', 'dropout_rate': 0.0, 'input_key': 'fps', '_target_': <class 'cosmos_predict2.conditioner.BooleanFlag'>}, '_target_': <class 'cosmos_predict2.conditioner.VideoConditioner'>}, 'conditioning_strategy': 'frame_replace', 'min_num_conditional_frames': 1, 'max_num_conditional_frames': 2, 'sigma_conditional': 0.0001, 'net': {'max_img_h': 240, 'max_img_w': 240, 'max_frames': 128, 'in_channels': 16, 'out_channels': 16, 'patch_spatial': 2, 'patch_temporal': 1, 'concat_padding_mask': True, 'model_channels': 2048, 'num_blocks': 28, 'num_heads': 16, 'atten_backend': 'minimal_a2a', 'pos_emb_cls': 'rope3d', 'pos_emb_learnable': True, 'pos_emb_interpolation': 'crop', 'use_adaln_lora': True, 'adaln_lora_dim': 256, 'rope_h_extrapolation_ratio': 3.0, 'rope_w_extrapolation_ratio': 3.0, 'rope_t_extrapolation_ratio': 1.0, 'extra_per_block_abs_pos_emb': False, 'rope_enable_fps_modulation': False, 'sac_config': {'mode': 'predict2_2b_720', 'every_n_blocks': 1, '_target_': 'cosmos_predict2.models.text2image_dit.SACConfig'}, '_target_': <class 'cosmos_predict2.models.video2world_dit.MinimalV1LVGDiT'>}, 'tokenizer': {'chunk_duration': 81, 'load_mean_std': False, 'temporal_window': 16, 'name': 'tokenizer', 'vae_pth': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth', '_target_': <class 'cosmos_predict2.tokenizers.tokenizer.TokenizerInterface'>}, 'guardrail_config': {'checkpoint_dir': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints', 'offload_model_to_cpu': True, 'enabled': False}, 'precision': 'bfloat16', 'rectified_flow_t_scaling_factor': 1.0, 'rectified_flow_loss_weight_uniform': True, 'resize_online': False, 'resolution': '480', 'ema': {'enabled': False, 'rate': 0.1, 'iteration_shift': 0, '_target_': 'cosmos_predict2.configs.defaults.ema.EMAConfig'}, 'sigma_data': 1.0, 'state_ch': 16, 'state_t': 16, 'text_encoder': {'cls': <TextEncoderClass.T5: 't5'>, 't5': {'ckpt_path': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/text_encoder/t5-11b', 'num_tokens': 512, 'embed_dim': 1024}}, 'input_video_key': 'video', 'input_image_key': 'images', 'timestamps': {'nfe': 35, 't_min': 0.002, 't_max': 80.0, 'order': 7.0, 'is_forward': False}}, 'debug_without_randomness': False, 'fsdp_shard_size': 0, 'high_sigma_ratio': 0.05}, '_recursive_': False, '_target_': <class 'cosmos_predict2.models.video2world_model.Predict2Video2WorldModel'>}
* world2action_pipe: None
* optimizer: {'optim_type': 'fusedadam', 'model': None, 'lr': 0.0001778279410038923, 'weight_decay': 0.1, 'betas': [0.9, 0.99], 'eps': 1e-08, 'master_weights': True, 'capturable': True, '_target_': <function get_base_optimizer at 0x7f0a19f36d40>}
* scheduler: {'_target_': <class 'cosmos_predict2.configs.defaults.scheduler.ConstantScheduler'>}
* data_config: None
* video_dataset_train: {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}
* video_dataset_val: {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}
* dataloader_train: {'batch_size': 4, 'shuffle': None, 'sampler': {'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <function get_sampler at 0x7f0a293a84c0>}, 'batch_sampler': None, 'num_workers': 12, 'collate_fn': None, 'pin_memory': True, 'drop_last': True, 'timeout': 0, 'worker_init_fn': None, 'multiprocessing_context': None, 'generator': None, 'prefetch_factor': 8, 'persistent_workers': True, 'pin_memory_device': '', 'in_order': False, 'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <class 'torch.utils.data.dataloader.DataLoader'>}
* dataloader_val: {'batch_size': 1, 'shuffle': None, 'sampler': {'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <function get_sampler at 0x7f0a293a84c0>}, 'batch_sampler': None, 'num_workers': 0, 'collate_fn': None, 'pin_memory': False, 'drop_last': False, 'timeout': 0, 'worker_init_fn': None, 'multiprocessing_context': None, 'generator': None, 'prefetch_factor': None, 'persistent_workers': False, 'pin_memory_device': '', 'in_order': False, 'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <class 'torch.utils.data.dataloader.DataLoader'>}
* job:
   * project: posttraining
   * group: video2world
   * name: v2w_push_lora_rank32_lr1.778e-04_bsz32
* trainer:
   * type: <class 'imaginaire.trainer.ImaginaireTrainer'>
   * callbacks: {'ema': {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.EMAModelCallback'>}, 'progress_bar': {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.ProgressBarCallback'>}, 'low_prec': {'config': None, 'trainer': None, 'update_iter': 1, '_target_': <class 'imaginaire.utils.callback.LowPrecisionCallback'>}, 'iter_speed': {'hit_thres': 5, 'every_n': 1000, '_target_': <class 'cosmos_predict2.callbacks.iter_speed.IterSpeed'>}, 'device_monitor': {'every_n': 1000, 'step_size': 1, 'log_memory_detail': True, '_target_': <class 'cosmos_predict2.callbacks.device_monitor.DeviceMonitor'>}, 'manual_gc': {'warm_up': 5, 'every_n': 5, '_target_': <class 'imaginaire.callbacks.manual_gc.ManualGarbageCollection'>}, 'grad_clip': {'clip_norm': 10.0, 'force_finite': True, 'log_wandb': False, '_target_': <class 'cosmos_predict2.callbacks.grad_clip.GradClip'>}, 'video_eval': {'fuse_lora': True, '_target_': <class 'cosmos_predict2.callbacks.video_eval.VideoEvalCallback'>}}
   * distributed_parallelism: ddp
   * ddp:
      * find_unused_parameters: False
      * static_graph: True
      * broadcast_buffers: True
   * cudnn:
      * deterministic: False
      * benchmark: True
   * seed: 0
   * grad_scaler_args: {'enabled': False}
   * max_iter: 500
   * max_val_iter: None
   * logging_iter: 1000
   * run_validation: False
   * validation_iter: 999999999
   * timeout_period: 999999999
   * memory_format: torch.preserve_format
   * grad_accum_iter: 8
   * profiling:
      * enable_profiling: False
      * enable_memory_snapshot: False
      * profile_freq: 1
      * first_n_rank: 4
      * record_shape: True
      * profile_memory: True
      * with_stack: True
      * with_modules: True
* model_parallel: ModelParallelConfig(tensor_model_parallel_size=1, pipeline_model_parallel_comm_backend=None, pipeline_model_parallel_size=1, virtual_pipeline_model_parallel_size=None, sequence_parallel=False, context_parallel_size=1, hierarchical_context_parallel_sizes=None, expert_model_parallel_size=1, expert_tensor_parallel_size=1, moe_extended_tp=False, perform_initialization=True, use_cpu_initialization=False, fp16=False, bf16=False, params_dtype=torch.float32, timers=None, finalize_model_grads_func=None, grad_scale_func=None, no_sync_func=None, grad_sync_func=None, param_sync_func=None, deterministic_mode=False, enable_autocast=False, autocast_dtype=torch.float32, num_microbatches_with_partial_activation_checkpoints=None, gradient_accumulation_fusion=False, async_tensor_model_parallel_allreduce=False, use_te_rng_tracker=False, tp_comm_overlap=False, tp_comm_bulk_wgrad=True, tp_comm_bulk_dgrad=True, tp_comm_overlap_ag=True, tp_comm_overlap_rs=True, tp_comm_overlap_rs_dgrad=False, tp_comm_split_ag=True, tp_comm_atomic_ag=False, tp_comm_split_rs=True, tp_comm_atomic_rs=False, cross_entropy_loss_fusion=False, cross_entropy_fusion_impl='native', tp_comm_overlap_disable_qkv=False, tp_comm_overlap_disable_fc1=False, tp_comm_bootstrap_backend='nccl', pipeline_dtype=None, variable_seq_lengths=False, overlap_p2p_comm=False, batch_p2p_comm=True, batch_p2p_sync=True, use_ring_exchange_p2p=False, deallocate_pipeline_outputs=False, defer_embedding_wgrad_compute=False, wgrad_deferral_limit=0, pipeline_model_parallel_split_rank=None, overlap_p2p_comm_warmup_flush=False, microbatch_group_size_per_vp_stage=1, cpu_offloading=False, cpu_offloading_num_layers=0, _cpu_offloading_context=None, cpu_offloading_activations=False, cpu_offloading_weights=False, barrier_with_L1_time=True)
* checkpoint:
   * type: {'callbacks': None, '_target_': <class 'cosmos_predict2.checkpointer.Checkpointer'>}
   * dcp_async_mode_enabled: False
   * save_iter: 100
   * load_path: 
   * load_training_state: False
   * only_load_scheduler_state: False
   * strict_resume: True
   * jit:
      * enabled: False
      * input_shape: None
      * device: cuda
      * dtype: bfloat16
      * strict: True
   * verbose: True
   * keys_not_to_resume: []
   * broadcast_via_filesystem: False
   * load_ema_to_reg: False
   * dcp_allow_mismatched_size: False
* defaults: ['_self_', {'data_config': None}, {'video_dataset_train': None}, {'video_dataset_val': None}, {'dataloader_train': None}, {'dataloader_val': None}, {'world2action_pipe': None}, {'optimizer': 'fusedadamw'}, {'scheduler': 'constant'}, {'model': None}, {'callbacks': ['basic']}, {'net': None}, {'ema': None}, {'checkpoint': None}, {'ckpt_type': None}, {'experiment': None}]
[05-07 17:16:59|WARNING|imaginaire/utils/misc.py:127:print_environ_variables] Environment variable TORCH_HOME not set!
[05-07 17:16:59|INFO|imaginaire/utils/misc.py:125:print_environ_variables] Environment variable IMAGINAIRE_OUTPUT_ROOT: /home/ubuntu/checkpoints
[05-07 17:16:59|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0.
[05-07 17:16:59|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback ema: {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.EMAModelCallback'>}
[05-07 17:16:59|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback progress_bar: {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.ProgressBarCallback'>}
[05-07 17:16:59|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback low_prec: {'config': None, 'trainer': None, 'update_iter': 1, '_target_': <class 'imaginaire.utils.callback.LowPrecisionCallback'>}
[05-07 17:16:59|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback iter_speed: {'hit_thres': 5, 'every_n': 1000, '_target_': <class 'cosmos_predict2.callbacks.iter_speed.IterSpeed'>}
[05-07 17:16:59|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback device_monitor: {'every_n': 1000, 'step_size': 1, 'log_memory_detail': True, '_target_': <class 'cosmos_predict2.callbacks.device_monitor.DeviceMonitor'>}
[05-07 17:16:59|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback manual_gc: {'warm_up': 5, 'every_n': 5, '_target_': <class 'imaginaire.callbacks.manual_gc.ManualGarbageCollection'>}
[05-07 17:16:59|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback grad_clip: {'clip_norm': 10.0, 'force_finite': True, 'log_wandb': False, '_target_': <class 'cosmos_predict2.callbacks.grad_clip.GradClip'>}
[05-07 17:16:59|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback video_eval: {'fuse_lora': True, '_target_': <class 'cosmos_predict2.callbacks.video_eval.VideoEvalCallback'>}
[05-07 17:16:59|CRITICAL|cosmos_predict2/models/video2world_model.py:138:__init__] Using mean loss reduce with loss scale 100.0
[05-07 17:16:59|WARNING|cosmos_predict2/pipelines/video2world.py:277:from_config] precision torch.bfloat16
[05-07 17:17:00|INFO|cosmos_predict2/tokenizers/tokenizer.py:687:_video_vae] Loading /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth
[05-07 17:17:00|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:689:_video_vae] Successfully loaded /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth
[05-07 17:17:00|INFO|imaginaire/utils/distributed.py:430:sync_model_states] Synchronizing model states from rank 0 to all ranks in process group [0].
[05-07 17:17:00|INFO|cosmos_predict2/pipelines/video2world.py:335:from_config] Loading DiT from /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt
[05-07 17:17:04|SUCCESS|cosmos_predict2/pipelines/video2world.py:354:from_config] Successfully loaded DiT from /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt
[05-07 17:17:05|INFO|cosmos_predict2/models/video2world_model.py:321:add_lora_to_model] Adding LoRA adapters: rank=32, alpha=32, targets=['q_proj', 'k_proj', 'v_proj', 'output_proj', 'x_embedder.proj.1', 'linear_1', 'linear_2', 'mlp.layer1', 'mlp.layer2']
[05-07 17:17:05|INFO|cosmos_predict2/models/video2world_model.py:345:add_lora_to_model] LoRA injection successful: 46,336,256 trainable parameters out of 2,002,749,696 total (2.314%)
[05-07 17:17:05|INFO|cosmos_predict2/models/video2world_model.py:363:_log_lora_statistics] LoRA parameter breakdown:
[05-07 17:17:05|INFO|cosmos_predict2/models/video2world_model.py:365:_log_lora_statistics]   lora_A: 22,153,472 parameters
[05-07 17:17:05|INFO|cosmos_predict2/models/video2world_model.py:365:_log_lora_statistics]   lora_B: 24,182,784 parameters
[05-07 17:17:05|INFO|cosmos_predict2/models/video2world_model.py:366:_log_lora_statistics]   Total LoRA: 46,336,256 parameters
[05-07 17:17:05|INFO|cosmos_predict2/models/video2world_model.py:185:__init__] Total parameters: 2.00B, Frozen parameters: 1,778,679,808, Trainable parameters: 224,069,888
[05-07 17:17:05|INFO|cosmos_predict2/models/video2world_model.py:202:__init__] FSDP (Fully Sharded Data Parallel) is disabled.
[05-07 17:17:05|INFO|cosmos_predict2/data/dataset_video.py:102:__init__] 204 videos in total
[05-07 17:17:05|INFO|cosmos_predict2/data/dataset_video.py:112:__init__] 204 videos in train.
[05-07 17:17:05|INFO|cosmos_predict2/data/dataset_video.py:102:__init__] 204 videos in total
[05-07 17:17:05|INFO|cosmos_predict2/data/dataset_video.py:112:__init__] 204 videos in train.
[05-07 17:17:05|CRITICAL|cosmos_predict2/utils/optim_instantiate_dtensor.py:49:get_base_optimizer] total num parameters : 224,069,888
[05-07 17:17:05|WARNING|cosmos_predict2/utils/fused_adam_dtensor.py:103:__init__] FusedAdam master_weights: True capturable: True
[05-07 17:17:05|INFO|cosmos_predict2/checkpointer.py:288:load] Training from scratch.
[05-07 17:17:05|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint loading: 0.0013 seconds
[05-07 17:17:05|CRITICAL|imaginaire/trainer.py:178:train] Distributed parallelism mode: ddp
[05-07 17:17:06|INFO|imaginaire/trainer.py:186:train] Starting training...
[05-07 17:17:06|INFO|cosmos_predict2/callbacks/device_monitor.py:92:on_train_start] DeviceMonitor callback: local_dir: /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/DeviceMonitor
[05-07 17:17:09|WARNING|imaginaire/utils/distributed.py:284:ddp_sync_grad] DDP static_graph=True is incompatible with sync_grad(). Performance will be reduced.
[05-07 17:18:29|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 1: Hit counter: 1/5 | Loss: 4.0617 | Time: 89.90s
[05-07 17:19:00|INFO|imaginaire/trainer.py:116:__init__] Config:
* model: {'config': {'train_architecture': 'lora', 'lora_rank': 32, 'lora_alpha': 32, 'lora_target_modules': 'q_proj,k_proj,v_proj,output_proj,x_embedder.proj.1,linear_1,linear_2,mlp.layer1,mlp.layer2', 'init_lora_weights': True, 'precision': 'bfloat16', 'input_video_key': 'video', 'input_image_key': 'images', 'loss_reduce': 'mean', 'loss_scale': 100.0, 'adjust_video_noise': True, 'model_manager_config': {'dit_path': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt', 'text_encoder_path': '', '_target_': 'cosmos_predict2.models.video2world_model.Predict2ModelManagerConfig'}, 'pipe_config': {'adjust_video_noise': True, 'conditioner': {'fps': {'output_key': 'fps', 'dropout_rate': 0.0, 'dtype': None, 'input_key': 'fps', '_target_': <class 'cosmos_predict2.conditioner.ReMapkey'>}, 'padding_mask': {'output_key': 'padding_mask', 'dropout_rate': 0.0, 'dtype': None, 'input_key': 'padding_mask', '_target_': <class 'cosmos_predict2.conditioner.ReMapkey'>}, 'text': {'dropout_rate': 0.0, 'input_key': ['obs/language_embedding'], '_target_': <class 'cosmos_predict2.conditioner.TextAttr'>}, 'use_video_condition': {'output_key': 'use_video_condition', 'dropout_rate': 0.0, 'input_key': 'fps', '_target_': <class 'cosmos_predict2.conditioner.BooleanFlag'>}, '_target_': <class 'cosmos_predict2.conditioner.VideoConditioner'>}, 'conditioning_strategy': 'frame_replace', 'min_num_conditional_frames': 1, 'max_num_conditional_frames': 2, 'sigma_conditional': 0.0001, 'net': {'max_img_h': 240, 'max_img_w': 240, 'max_frames': 128, 'in_channels': 16, 'out_channels': 16, 'patch_spatial': 2, 'patch_temporal': 1, 'concat_padding_mask': True, 'model_channels': 2048, 'num_blocks': 28, 'num_heads': 16, 'atten_backend': 'minimal_a2a', 'pos_emb_cls': 'rope3d', 'pos_emb_learnable': True, 'pos_emb_interpolation': 'crop', 'use_adaln_lora': True, 'adaln_lora_dim': 256, 'rope_h_extrapolation_ratio': 3.0, 'rope_w_extrapolation_ratio': 3.0, 'rope_t_extrapolation_ratio': 1.0, 'extra_per_block_abs_pos_emb': False, 'rope_enable_fps_modulation': False, 'sac_config': {'mode': 'predict2_2b_720', 'every_n_blocks': 1, '_target_': 'cosmos_predict2.models.text2image_dit.SACConfig'}, '_target_': <class 'cosmos_predict2.models.video2world_dit.MinimalV1LVGDiT'>}, 'tokenizer': {'chunk_duration': 81, 'load_mean_std': False, 'temporal_window': 16, 'name': 'tokenizer', 'vae_pth': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth', '_target_': <class 'cosmos_predict2.tokenizers.tokenizer.TokenizerInterface'>}, 'guardrail_config': {'checkpoint_dir': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints', 'offload_model_to_cpu': True, 'enabled': False}, 'precision': 'bfloat16', 'rectified_flow_t_scaling_factor': 1.0, 'rectified_flow_loss_weight_uniform': True, 'resize_online': False, 'resolution': '480', 'ema': {'enabled': False, 'rate': 0.1, 'iteration_shift': 0, '_target_': 'cosmos_predict2.configs.defaults.ema.EMAConfig'}, 'sigma_data': 1.0, 'state_ch': 16, 'state_t': 16, 'text_encoder': {'cls': <TextEncoderClass.T5: 't5'>, 't5': {'ckpt_path': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/text_encoder/t5-11b', 'num_tokens': 512, 'embed_dim': 1024}}, 'input_video_key': 'video', 'input_image_key': 'images', 'timestamps': {'nfe': 35, 't_min': 0.002, 't_max': 80.0, 'order': 7.0, 'is_forward': False}}, 'debug_without_randomness': False, 'fsdp_shard_size': 0, 'high_sigma_ratio': 0.05}, '_recursive_': False, '_target_': <class 'cosmos_predict2.models.video2world_model.Predict2Video2WorldModel'>}
* world2action_pipe: None
* optimizer: {'optim_type': 'fusedadam', 'model': None, 'lr': 0.0001778279410038923, 'weight_decay': 0.1, 'betas': [0.9, 0.99], 'eps': 1e-08, 'master_weights': True, 'capturable': True, '_target_': <function get_base_optimizer at 0x734ac793ab90>}
* scheduler: {'_target_': <class 'cosmos_predict2.configs.defaults.scheduler.ConstantScheduler'>}
* data_config: None
* video_dataset_train: {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}
* video_dataset_val: {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}
* dataloader_train: {'batch_size': 2, 'shuffle': None, 'sampler': {'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <function get_sampler at 0x734ad73ac310>}, 'batch_sampler': None, 'num_workers': 12, 'collate_fn': None, 'pin_memory': True, 'drop_last': True, 'timeout': 0, 'worker_init_fn': None, 'multiprocessing_context': None, 'generator': None, 'prefetch_factor': 8, 'persistent_workers': True, 'pin_memory_device': '', 'in_order': False, 'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <class 'torch.utils.data.dataloader.DataLoader'>}
* dataloader_val: {'batch_size': 1, 'shuffle': None, 'sampler': {'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <function get_sampler at 0x734ad73ac310>}, 'batch_sampler': None, 'num_workers': 0, 'collate_fn': None, 'pin_memory': False, 'drop_last': False, 'timeout': 0, 'worker_init_fn': None, 'multiprocessing_context': None, 'generator': None, 'prefetch_factor': None, 'persistent_workers': False, 'pin_memory_device': '', 'in_order': False, 'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <class 'torch.utils.data.dataloader.DataLoader'>}
* job:
   * project: posttraining
   * group: video2world
   * name: v2w_push_lora_rank32_lr1.778e-04_bsz32
* trainer:
   * type: <class 'imaginaire.trainer.ImaginaireTrainer'>
   * callbacks: {'ema': {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.EMAModelCallback'>}, 'progress_bar': {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.ProgressBarCallback'>}, 'low_prec': {'config': None, 'trainer': None, 'update_iter': 1, '_target_': <class 'imaginaire.utils.callback.LowPrecisionCallback'>}, 'iter_speed': {'hit_thres': 5, 'every_n': 1000, '_target_': <class 'cosmos_predict2.callbacks.iter_speed.IterSpeed'>}, 'device_monitor': {'every_n': 1000, 'step_size': 1, 'log_memory_detail': True, '_target_': <class 'cosmos_predict2.callbacks.device_monitor.DeviceMonitor'>}, 'manual_gc': {'warm_up': 5, 'every_n': 5, '_target_': <class 'imaginaire.callbacks.manual_gc.ManualGarbageCollection'>}, 'grad_clip': {'clip_norm': 10.0, 'force_finite': True, 'log_wandb': False, '_target_': <class 'cosmos_predict2.callbacks.grad_clip.GradClip'>}, 'video_eval': {'fuse_lora': True, '_target_': <class 'cosmos_predict2.callbacks.video_eval.VideoEvalCallback'>}}
   * distributed_parallelism: ddp
   * ddp:
      * find_unused_parameters: False
      * static_graph: True
      * broadcast_buffers: True
   * cudnn:
      * deterministic: False
      * benchmark: True
   * seed: 0
   * grad_scaler_args: {'enabled': False}
   * max_iter: 500
   * max_val_iter: None
   * logging_iter: 1000
   * run_validation: False
   * validation_iter: 999999999
   * timeout_period: 999999999
   * memory_format: torch.preserve_format
   * grad_accum_iter: 4
   * profiling:
      * enable_profiling: False
      * enable_memory_snapshot: False
      * profile_freq: 1
      * first_n_rank: 4
      * record_shape: True
      * profile_memory: True
      * with_stack: True
      * with_modules: True
* model_parallel: ModelParallelConfig(tensor_model_parallel_size=1, pipeline_model_parallel_comm_backend=None, pipeline_model_parallel_size=1, virtual_pipeline_model_parallel_size=None, sequence_parallel=False, context_parallel_size=1, hierarchical_context_parallel_sizes=None, expert_model_parallel_size=1, expert_tensor_parallel_size=1, moe_extended_tp=False, perform_initialization=True, use_cpu_initialization=False, fp16=False, bf16=False, params_dtype=torch.float32, timers=None, finalize_model_grads_func=None, grad_scale_func=None, no_sync_func=None, grad_sync_func=None, param_sync_func=None, deterministic_mode=False, enable_autocast=False, autocast_dtype=torch.float32, num_microbatches_with_partial_activation_checkpoints=None, gradient_accumulation_fusion=False, async_tensor_model_parallel_allreduce=False, use_te_rng_tracker=False, tp_comm_overlap=False, tp_comm_bulk_wgrad=True, tp_comm_bulk_dgrad=True, tp_comm_overlap_ag=True, tp_comm_overlap_rs=True, tp_comm_overlap_rs_dgrad=False, tp_comm_split_ag=True, tp_comm_atomic_ag=False, tp_comm_split_rs=True, tp_comm_atomic_rs=False, cross_entropy_loss_fusion=False, cross_entropy_fusion_impl='native', tp_comm_overlap_disable_qkv=False, tp_comm_overlap_disable_fc1=False, tp_comm_bootstrap_backend='nccl', pipeline_dtype=None, variable_seq_lengths=False, overlap_p2p_comm=False, batch_p2p_comm=True, batch_p2p_sync=True, use_ring_exchange_p2p=False, deallocate_pipeline_outputs=False, defer_embedding_wgrad_compute=False, wgrad_deferral_limit=0, pipeline_model_parallel_split_rank=None, overlap_p2p_comm_warmup_flush=False, microbatch_group_size_per_vp_stage=1, cpu_offloading=False, cpu_offloading_num_layers=0, _cpu_offloading_context=None, cpu_offloading_activations=False, cpu_offloading_weights=False, barrier_with_L1_time=True)
* checkpoint:
   * type: {'callbacks': None, '_target_': <class 'cosmos_predict2.checkpointer.Checkpointer'>}
   * dcp_async_mode_enabled: False
   * save_iter: 100
   * load_path: 
   * load_training_state: False
   * only_load_scheduler_state: False
   * strict_resume: True
   * jit:
      * enabled: False
      * input_shape: None
      * device: cuda
      * dtype: bfloat16
      * strict: True
   * verbose: True
   * keys_not_to_resume: []
   * broadcast_via_filesystem: False
   * load_ema_to_reg: False
   * dcp_allow_mismatched_size: False
* defaults: ['_self_', {'data_config': None}, {'video_dataset_train': None}, {'video_dataset_val': None}, {'dataloader_train': None}, {'dataloader_val': None}, {'world2action_pipe': None}, {'optimizer': 'fusedadamw'}, {'scheduler': 'constant'}, {'model': None}, {'callbacks': ['basic']}, {'net': None}, {'ema': None}, {'checkpoint': None}, {'ckpt_type': None}, {'experiment': None}]
[05-07 17:19:00|WARNING|imaginaire/utils/misc.py:127:print_environ_variables] Environment variable TORCH_HOME not set!
[05-07 17:19:00|INFO|imaginaire/utils/misc.py:125:print_environ_variables] Environment variable IMAGINAIRE_OUTPUT_ROOT: /home/ubuntu/checkpoints
[05-07 17:19:00|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0.
[05-07 17:19:00|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback ema: {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.EMAModelCallback'>}
[05-07 17:19:00|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback progress_bar: {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.ProgressBarCallback'>}
[05-07 17:19:00|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback low_prec: {'config': None, 'trainer': None, 'update_iter': 1, '_target_': <class 'imaginaire.utils.callback.LowPrecisionCallback'>}
[05-07 17:19:00|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback iter_speed: {'hit_thres': 5, 'every_n': 1000, '_target_': <class 'cosmos_predict2.callbacks.iter_speed.IterSpeed'>}
[05-07 17:19:00|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback device_monitor: {'every_n': 1000, 'step_size': 1, 'log_memory_detail': True, '_target_': <class 'cosmos_predict2.callbacks.device_monitor.DeviceMonitor'>}
[05-07 17:19:00|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback manual_gc: {'warm_up': 5, 'every_n': 5, '_target_': <class 'imaginaire.callbacks.manual_gc.ManualGarbageCollection'>}
[05-07 17:19:00|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback grad_clip: {'clip_norm': 10.0, 'force_finite': True, 'log_wandb': False, '_target_': <class 'cosmos_predict2.callbacks.grad_clip.GradClip'>}
[05-07 17:19:00|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback video_eval: {'fuse_lora': True, '_target_': <class 'cosmos_predict2.callbacks.video_eval.VideoEvalCallback'>}
[05-07 17:19:00|CRITICAL|cosmos_predict2/models/video2world_model.py:138:__init__] Using mean loss reduce with loss scale 100.0
[05-07 17:19:00|WARNING|cosmos_predict2/pipelines/video2world.py:277:from_config] precision torch.bfloat16
[05-07 17:19:00|INFO|cosmos_predict2/tokenizers/tokenizer.py:687:_video_vae] Loading /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth
[05-07 17:19:00|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:689:_video_vae] Successfully loaded /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth
[05-07 17:19:00|INFO|imaginaire/utils/distributed.py:430:sync_model_states] Synchronizing model states from rank 0 to all ranks in process group [0].
[05-07 17:19:00|INFO|cosmos_predict2/pipelines/video2world.py:335:from_config] Loading DiT from /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt
[05-07 17:19:05|SUCCESS|cosmos_predict2/pipelines/video2world.py:354:from_config] Successfully loaded DiT from /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt
[05-07 17:19:05|INFO|cosmos_predict2/models/video2world_model.py:321:add_lora_to_model] Adding LoRA adapters: rank=32, alpha=32, targets=['q_proj', 'k_proj', 'v_proj', 'output_proj', 'x_embedder.proj.1', 'linear_1', 'linear_2', 'mlp.layer1', 'mlp.layer2']
[05-07 17:19:06|INFO|cosmos_predict2/models/video2world_model.py:345:add_lora_to_model] LoRA injection successful: 46,336,256 trainable parameters out of 2,002,749,696 total (2.314%)
[05-07 17:19:06|INFO|cosmos_predict2/models/video2world_model.py:363:_log_lora_statistics] LoRA parameter breakdown:
[05-07 17:19:06|INFO|cosmos_predict2/models/video2world_model.py:365:_log_lora_statistics]   lora_A: 22,153,472 parameters
[05-07 17:19:06|INFO|cosmos_predict2/models/video2world_model.py:365:_log_lora_statistics]   lora_B: 24,182,784 parameters
[05-07 17:19:06|INFO|cosmos_predict2/models/video2world_model.py:366:_log_lora_statistics]   Total LoRA: 46,336,256 parameters
[05-07 17:19:06|INFO|cosmos_predict2/models/video2world_model.py:185:__init__] Total parameters: 2.00B, Frozen parameters: 1,778,679,808, Trainable parameters: 224,069,888
[05-07 17:19:06|INFO|cosmos_predict2/models/video2world_model.py:202:__init__] FSDP (Fully Sharded Data Parallel) is disabled.
[05-07 17:19:06|INFO|cosmos_predict2/data/dataset_video.py:102:__init__] 204 videos in total
[05-07 17:19:06|INFO|cosmos_predict2/data/dataset_video.py:112:__init__] 204 videos in train.
[05-07 17:19:06|INFO|cosmos_predict2/data/dataset_video.py:102:__init__] 204 videos in total
[05-07 17:19:06|INFO|cosmos_predict2/data/dataset_video.py:112:__init__] 204 videos in train.
[05-07 17:19:06|CRITICAL|cosmos_predict2/utils/optim_instantiate_dtensor.py:49:get_base_optimizer] total num parameters : 224,069,888
[05-07 17:19:06|WARNING|cosmos_predict2/utils/fused_adam_dtensor.py:103:__init__] FusedAdam master_weights: True capturable: True
[05-07 17:19:06|INFO|cosmos_predict2/checkpointer.py:288:load] Training from scratch.
[05-07 17:19:06|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint loading: 0.0014 seconds
[05-07 17:19:06|CRITICAL|imaginaire/trainer.py:178:train] Distributed parallelism mode: ddp
[05-07 17:19:06|INFO|imaginaire/trainer.py:186:train] Starting training...
[05-07 17:19:06|INFO|cosmos_predict2/callbacks/device_monitor.py:92:on_train_start] DeviceMonitor callback: local_dir: /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/DeviceMonitor
[05-07 17:19:08|WARNING|imaginaire/utils/distributed.py:284:ddp_sync_grad] DDP static_graph=True is incompatible with sync_grad(). Performance will be reduced.
[05-07 17:19:35|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 1: Hit counter: 1/5 | Loss: 4.5430 | Time: 35.34s
[05-07 17:19:52|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 2: Hit counter: 2/5 | Loss: 7.1425 | Time: 17.27s
[05-07 17:20:09|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 3: Hit counter: 3/5 | Loss: 4.8634 | Time: 17.24s
[05-07 17:20:27|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 4: Hit counter: 4/5 | Loss: 4.9483 | Time: 17.32s
[05-07 17:21:34|INFO|imaginaire/trainer.py:116:__init__] Config:
* model: {'config': {'train_architecture': 'lora', 'lora_rank': 32, 'lora_alpha': 32, 'lora_target_modules': 'q_proj,k_proj,v_proj,output_proj,x_embedder.proj.1,linear_1,linear_2,mlp.layer1,mlp.layer2', 'init_lora_weights': True, 'precision': 'bfloat16', 'input_video_key': 'video', 'input_image_key': 'images', 'loss_reduce': 'mean', 'loss_scale': 100.0, 'adjust_video_noise': True, 'model_manager_config': {'dit_path': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt', 'text_encoder_path': '', '_target_': 'cosmos_predict2.models.video2world_model.Predict2ModelManagerConfig'}, 'pipe_config': {'adjust_video_noise': True, 'conditioner': {'fps': {'output_key': 'fps', 'dropout_rate': 0.0, 'dtype': None, 'input_key': 'fps', '_target_': <class 'cosmos_predict2.conditioner.ReMapkey'>}, 'padding_mask': {'output_key': 'padding_mask', 'dropout_rate': 0.0, 'dtype': None, 'input_key': 'padding_mask', '_target_': <class 'cosmos_predict2.conditioner.ReMapkey'>}, 'text': {'dropout_rate': 0.0, 'input_key': ['obs/language_embedding'], '_target_': <class 'cosmos_predict2.conditioner.TextAttr'>}, 'use_video_condition': {'output_key': 'use_video_condition', 'dropout_rate': 0.0, 'input_key': 'fps', '_target_': <class 'cosmos_predict2.conditioner.BooleanFlag'>}, '_target_': <class 'cosmos_predict2.conditioner.VideoConditioner'>}, 'conditioning_strategy': 'frame_replace', 'min_num_conditional_frames': 1, 'max_num_conditional_frames': 2, 'sigma_conditional': 0.0001, 'net': {'max_img_h': 240, 'max_img_w': 240, 'max_frames': 128, 'in_channels': 16, 'out_channels': 16, 'patch_spatial': 2, 'patch_temporal': 1, 'concat_padding_mask': True, 'model_channels': 2048, 'num_blocks': 28, 'num_heads': 16, 'atten_backend': 'minimal_a2a', 'pos_emb_cls': 'rope3d', 'pos_emb_learnable': True, 'pos_emb_interpolation': 'crop', 'use_adaln_lora': True, 'adaln_lora_dim': 256, 'rope_h_extrapolation_ratio': 3.0, 'rope_w_extrapolation_ratio': 3.0, 'rope_t_extrapolation_ratio': 1.0, 'extra_per_block_abs_pos_emb': False, 'rope_enable_fps_modulation': False, 'sac_config': {'mode': 'predict2_2b_720', 'every_n_blocks': 1, '_target_': 'cosmos_predict2.models.text2image_dit.SACConfig'}, '_target_': <class 'cosmos_predict2.models.video2world_dit.MinimalV1LVGDiT'>}, 'tokenizer': {'chunk_duration': 81, 'load_mean_std': False, 'temporal_window': 16, 'name': 'tokenizer', 'vae_pth': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth', '_target_': <class 'cosmos_predict2.tokenizers.tokenizer.TokenizerInterface'>}, 'guardrail_config': {'checkpoint_dir': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints', 'offload_model_to_cpu': True, 'enabled': False}, 'precision': 'bfloat16', 'rectified_flow_t_scaling_factor': 1.0, 'rectified_flow_loss_weight_uniform': True, 'resize_online': False, 'resolution': '480', 'ema': {'enabled': False, 'rate': 0.1, 'iteration_shift': 0, '_target_': 'cosmos_predict2.configs.defaults.ema.EMAConfig'}, 'sigma_data': 1.0, 'state_ch': 16, 'state_t': 16, 'text_encoder': {'cls': <TextEncoderClass.T5: 't5'>, 't5': {'ckpt_path': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/text_encoder/t5-11b', 'num_tokens': 512, 'embed_dim': 1024}}, 'input_video_key': 'video', 'input_image_key': 'images', 'timestamps': {'nfe': 35, 't_min': 0.002, 't_max': 80.0, 'order': 7.0, 'is_forward': False}}, 'debug_without_randomness': False, 'fsdp_shard_size': 0, 'high_sigma_ratio': 0.05}, '_recursive_': False, '_target_': <class 'cosmos_predict2.models.video2world_model.Predict2Video2WorldModel'>}
* world2action_pipe: None
* optimizer: {'optim_type': 'fusedadam', 'model': None, 'lr': 4.445e-05, 'weight_decay': 0.1, 'betas': [0.9, 0.99], 'eps': 1e-08, 'master_weights': True, 'capturable': True, '_target_': <function get_base_optimizer at 0x745843b1ab90>}
* scheduler: {'_target_': <class 'cosmos_predict2.configs.defaults.scheduler.ConstantScheduler'>}
* data_config: None
* video_dataset_train: {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}
* video_dataset_val: {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}
* dataloader_train: {'batch_size': 2, 'shuffle': None, 'sampler': {'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <function get_sampler at 0x745852f4c310>}, 'batch_sampler': None, 'num_workers': 12, 'collate_fn': None, 'pin_memory': True, 'drop_last': True, 'timeout': 0, 'worker_init_fn': None, 'multiprocessing_context': None, 'generator': None, 'prefetch_factor': 8, 'persistent_workers': True, 'pin_memory_device': '', 'in_order': False, 'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <class 'torch.utils.data.dataloader.DataLoader'>}
* dataloader_val: {'batch_size': 1, 'shuffle': None, 'sampler': {'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <function get_sampler at 0x745852f4c310>}, 'batch_sampler': None, 'num_workers': 0, 'collate_fn': None, 'pin_memory': False, 'drop_last': False, 'timeout': 0, 'worker_init_fn': None, 'multiprocessing_context': None, 'generator': None, 'prefetch_factor': None, 'persistent_workers': False, 'pin_memory_device': '', 'in_order': False, 'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <class 'torch.utils.data.dataloader.DataLoader'>}
* job:
   * project: posttraining
   * group: video2world
   * name: v2w_push_lora_rank32_lr1.778e-04_bsz32
* trainer:
   * type: <class 'imaginaire.trainer.ImaginaireTrainer'>
   * callbacks: {'ema': {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.EMAModelCallback'>}, 'progress_bar': {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.ProgressBarCallback'>}, 'low_prec': {'config': None, 'trainer': None, 'update_iter': 1, '_target_': <class 'imaginaire.utils.callback.LowPrecisionCallback'>}, 'iter_speed': {'hit_thres': 5, 'every_n': 1000, '_target_': <class 'cosmos_predict2.callbacks.iter_speed.IterSpeed'>}, 'device_monitor': {'every_n': 1000, 'step_size': 1, 'log_memory_detail': True, '_target_': <class 'cosmos_predict2.callbacks.device_monitor.DeviceMonitor'>}, 'manual_gc': {'warm_up': 5, 'every_n': 5, '_target_': <class 'imaginaire.callbacks.manual_gc.ManualGarbageCollection'>}, 'grad_clip': {'clip_norm': 10.0, 'force_finite': True, 'log_wandb': False, '_target_': <class 'cosmos_predict2.callbacks.grad_clip.GradClip'>}, 'video_eval': {'fuse_lora': True, '_target_': <class 'cosmos_predict2.callbacks.video_eval.VideoEvalCallback'>}}
   * distributed_parallelism: ddp
   * ddp:
      * find_unused_parameters: False
      * static_graph: True
      * broadcast_buffers: True
   * cudnn:
      * deterministic: False
      * benchmark: True
   * seed: 0
   * grad_scaler_args: {'enabled': False}
   * max_iter: 500
   * max_val_iter: None
   * logging_iter: 1000
   * run_validation: False
   * validation_iter: 999999999
   * timeout_period: 999999999
   * memory_format: torch.preserve_format
   * grad_accum_iter: 4
   * profiling:
      * enable_profiling: False
      * enable_memory_snapshot: False
      * profile_freq: 1
      * first_n_rank: 4
      * record_shape: True
      * profile_memory: True
      * with_stack: True
      * with_modules: True
* model_parallel: ModelParallelConfig(tensor_model_parallel_size=1, pipeline_model_parallel_comm_backend=None, pipeline_model_parallel_size=1, virtual_pipeline_model_parallel_size=None, sequence_parallel=False, context_parallel_size=1, hierarchical_context_parallel_sizes=None, expert_model_parallel_size=1, expert_tensor_parallel_size=1, moe_extended_tp=False, perform_initialization=True, use_cpu_initialization=False, fp16=False, bf16=False, params_dtype=torch.float32, timers=None, finalize_model_grads_func=None, grad_scale_func=None, no_sync_func=None, grad_sync_func=None, param_sync_func=None, deterministic_mode=False, enable_autocast=False, autocast_dtype=torch.float32, num_microbatches_with_partial_activation_checkpoints=None, gradient_accumulation_fusion=False, async_tensor_model_parallel_allreduce=False, use_te_rng_tracker=False, tp_comm_overlap=False, tp_comm_bulk_wgrad=True, tp_comm_bulk_dgrad=True, tp_comm_overlap_ag=True, tp_comm_overlap_rs=True, tp_comm_overlap_rs_dgrad=False, tp_comm_split_ag=True, tp_comm_atomic_ag=False, tp_comm_split_rs=True, tp_comm_atomic_rs=False, cross_entropy_loss_fusion=False, cross_entropy_fusion_impl='native', tp_comm_overlap_disable_qkv=False, tp_comm_overlap_disable_fc1=False, tp_comm_bootstrap_backend='nccl', pipeline_dtype=None, variable_seq_lengths=False, overlap_p2p_comm=False, batch_p2p_comm=True, batch_p2p_sync=True, use_ring_exchange_p2p=False, deallocate_pipeline_outputs=False, defer_embedding_wgrad_compute=False, wgrad_deferral_limit=0, pipeline_model_parallel_split_rank=None, overlap_p2p_comm_warmup_flush=False, microbatch_group_size_per_vp_stage=1, cpu_offloading=False, cpu_offloading_num_layers=0, _cpu_offloading_context=None, cpu_offloading_activations=False, cpu_offloading_weights=False, barrier_with_L1_time=True)
* checkpoint:
   * type: {'callbacks': None, '_target_': <class 'cosmos_predict2.checkpointer.Checkpointer'>}
   * dcp_async_mode_enabled: False
   * save_iter: 100
   * load_path: 
   * load_training_state: False
   * only_load_scheduler_state: False
   * strict_resume: True
   * jit:
      * enabled: False
      * input_shape: None
      * device: cuda
      * dtype: bfloat16
      * strict: True
   * verbose: True
   * keys_not_to_resume: []
   * broadcast_via_filesystem: False
   * load_ema_to_reg: False
   * dcp_allow_mismatched_size: False
* defaults: ['_self_', {'data_config': None}, {'video_dataset_train': None}, {'video_dataset_val': None}, {'dataloader_train': None}, {'dataloader_val': None}, {'world2action_pipe': None}, {'optimizer': 'fusedadamw'}, {'scheduler': 'constant'}, {'model': None}, {'callbacks': ['basic']}, {'net': None}, {'ema': None}, {'checkpoint': None}, {'ckpt_type': None}, {'experiment': None}]
[05-07 17:21:34|WARNING|imaginaire/utils/misc.py:127:print_environ_variables] Environment variable TORCH_HOME not set!
[05-07 17:21:34|INFO|imaginaire/utils/misc.py:125:print_environ_variables] Environment variable IMAGINAIRE_OUTPUT_ROOT: /home/ubuntu/checkpoints
[05-07 17:21:34|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0.
[05-07 17:21:34|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback ema: {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.EMAModelCallback'>}
[05-07 17:21:34|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback progress_bar: {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.ProgressBarCallback'>}
[05-07 17:21:34|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback low_prec: {'config': None, 'trainer': None, 'update_iter': 1, '_target_': <class 'imaginaire.utils.callback.LowPrecisionCallback'>}
[05-07 17:21:34|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback iter_speed: {'hit_thres': 5, 'every_n': 1000, '_target_': <class 'cosmos_predict2.callbacks.iter_speed.IterSpeed'>}
[05-07 17:21:34|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback device_monitor: {'every_n': 1000, 'step_size': 1, 'log_memory_detail': True, '_target_': <class 'cosmos_predict2.callbacks.device_monitor.DeviceMonitor'>}
[05-07 17:21:34|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback manual_gc: {'warm_up': 5, 'every_n': 5, '_target_': <class 'imaginaire.callbacks.manual_gc.ManualGarbageCollection'>}
[05-07 17:21:34|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback grad_clip: {'clip_norm': 10.0, 'force_finite': True, 'log_wandb': False, '_target_': <class 'cosmos_predict2.callbacks.grad_clip.GradClip'>}
[05-07 17:21:34|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback video_eval: {'fuse_lora': True, '_target_': <class 'cosmos_predict2.callbacks.video_eval.VideoEvalCallback'>}
[05-07 17:21:34|CRITICAL|cosmos_predict2/models/video2world_model.py:138:__init__] Using mean loss reduce with loss scale 100.0
[05-07 17:21:34|WARNING|cosmos_predict2/pipelines/video2world.py:277:from_config] precision torch.bfloat16
[05-07 17:21:34|INFO|cosmos_predict2/tokenizers/tokenizer.py:687:_video_vae] Loading /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth
[05-07 17:21:34|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:689:_video_vae] Successfully loaded /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth
[05-07 17:21:34|INFO|imaginaire/utils/distributed.py:430:sync_model_states] Synchronizing model states from rank 0 to all ranks in process group [0].
[05-07 17:21:34|INFO|cosmos_predict2/pipelines/video2world.py:335:from_config] Loading DiT from /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt
[05-07 17:21:39|SUCCESS|cosmos_predict2/pipelines/video2world.py:354:from_config] Successfully loaded DiT from /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt
[05-07 17:21:40|INFO|cosmos_predict2/models/video2world_model.py:321:add_lora_to_model] Adding LoRA adapters: rank=32, alpha=32, targets=['q_proj', 'k_proj', 'v_proj', 'output_proj', 'x_embedder.proj.1', 'linear_1', 'linear_2', 'mlp.layer1', 'mlp.layer2']
[05-07 17:21:40|INFO|cosmos_predict2/models/video2world_model.py:345:add_lora_to_model] LoRA injection successful: 46,336,256 trainable parameters out of 2,002,749,696 total (2.314%)
[05-07 17:21:40|INFO|cosmos_predict2/models/video2world_model.py:363:_log_lora_statistics] LoRA parameter breakdown:
[05-07 17:21:40|INFO|cosmos_predict2/models/video2world_model.py:365:_log_lora_statistics]   lora_A: 22,153,472 parameters
[05-07 17:21:40|INFO|cosmos_predict2/models/video2world_model.py:365:_log_lora_statistics]   lora_B: 24,182,784 parameters
[05-07 17:21:40|INFO|cosmos_predict2/models/video2world_model.py:366:_log_lora_statistics]   Total LoRA: 46,336,256 parameters
[05-07 17:21:40|INFO|cosmos_predict2/models/video2world_model.py:185:__init__] Total parameters: 2.00B, Frozen parameters: 1,778,679,808, Trainable parameters: 224,069,888
[05-07 17:21:40|INFO|cosmos_predict2/models/video2world_model.py:202:__init__] FSDP (Fully Sharded Data Parallel) is disabled.
[05-07 17:21:40|INFO|cosmos_predict2/data/dataset_video.py:102:__init__] 204 videos in total
[05-07 17:21:40|INFO|cosmos_predict2/data/dataset_video.py:112:__init__] 204 videos in train.
[05-07 17:21:40|INFO|cosmos_predict2/data/dataset_video.py:102:__init__] 204 videos in total
[05-07 17:21:40|INFO|cosmos_predict2/data/dataset_video.py:112:__init__] 204 videos in train.
[05-07 17:21:40|CRITICAL|cosmos_predict2/utils/optim_instantiate_dtensor.py:49:get_base_optimizer] total num parameters : 224,069,888
[05-07 17:21:40|WARNING|cosmos_predict2/utils/fused_adam_dtensor.py:103:__init__] FusedAdam master_weights: True capturable: True
[05-07 17:21:40|INFO|cosmos_predict2/checkpointer.py:288:load] Training from scratch.
[05-07 17:21:40|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint loading: 0.0014 seconds
[05-07 17:21:40|CRITICAL|imaginaire/trainer.py:178:train] Distributed parallelism mode: ddp
[05-07 17:21:40|INFO|imaginaire/trainer.py:186:train] Starting training...
[05-07 17:21:40|INFO|cosmos_predict2/callbacks/device_monitor.py:92:on_train_start] DeviceMonitor callback: local_dir: /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/DeviceMonitor
[05-07 17:21:42|WARNING|imaginaire/utils/distributed.py:284:ddp_sync_grad] DDP static_graph=True is incompatible with sync_grad(). Performance will be reduced.
[05-07 17:22:09|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 1: Hit counter: 1/5 | Loss: 4.1693 | Time: 35.39s
[05-07 17:22:26|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 2: Hit counter: 2/5 | Loss: 6.1576 | Time: 17.27s
[05-07 17:22:44|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 3: Hit counter: 3/5 | Loss: 4.6536 | Time: 17.24s
[05-07 17:23:01|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 4: Hit counter: 4/5 | Loss: 3.3145 | Time: 17.34s
[05-07 17:23:18|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 5: Hit counter: 5/5 | Loss: 4.0082 | Time: 17.26s
[05-07 17:29:03|CRITICAL|imaginaire/callbacks/manual_gc.py:48:every_n_impl] Garbage collection disabled
[05-07 17:29:17|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000025.pt
[05-07 17:29:26|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 11.6441 seconds
[05-07 17:29:27|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000025.pt
[05-07 17:29:29|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 3.5409 seconds
[05-07 17:29:29|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000025.pt
[05-07 17:29:29|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0019 seconds
[05-07 17:29:29|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000025.pt
[05-07 17:29:29|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0020 seconds
[05-07 17:36:54|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000051.pt
[05-07 17:37:02|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 11.2918 seconds
[05-07 17:37:04|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000051.pt
[05-07 17:37:06|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 3.2635 seconds
[05-07 17:37:06|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000051.pt
[05-07 17:37:06|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0018 seconds
[05-07 17:37:06|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000051.pt
[05-07 17:37:06|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0022 seconds
[05-07 17:44:30|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000076.pt
[05-07 17:44:37|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 9.5314 seconds
[05-07 17:44:38|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000076.pt
[05-07 17:44:40|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 3.4604 seconds
[05-07 17:44:40|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000076.pt
[05-07 17:44:40|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0019 seconds
[05-07 17:44:40|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000076.pt
[05-07 17:44:40|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0021 seconds
[05-07 17:51:31|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000100.pt
[05-07 17:51:38|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 10.0176 seconds
[05-07 17:51:39|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000100.pt
[05-07 17:51:42|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 3.3693 seconds
[05-07 17:51:42|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000100.pt
[05-07 17:51:42|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0019 seconds
[05-07 17:51:42|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000100.pt
[05-07 17:51:42|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0020 seconds
[05-07 18:13:39|INFO|imaginaire/trainer.py:116:__init__] Config:
* model: {'config': {'train_architecture': 'lora', 'lora_rank': 32, 'lora_alpha': 32, 'lora_target_modules': 'q_proj,k_proj,v_proj,output_proj,x_embedder.proj.1,linear_1,linear_2,mlp.layer1,mlp.layer2', 'init_lora_weights': True, 'precision': 'bfloat16', 'input_video_key': 'video', 'input_image_key': 'images', 'loss_reduce': 'mean', 'loss_scale': 100.0, 'adjust_video_noise': True, 'model_manager_config': {'dit_path': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt', 'text_encoder_path': '', '_target_': 'cosmos_predict2.models.video2world_model.Predict2ModelManagerConfig'}, 'pipe_config': {'adjust_video_noise': True, 'conditioner': {'fps': {'output_key': 'fps', 'dropout_rate': 0.0, 'dtype': None, 'input_key': 'fps', '_target_': <class 'cosmos_predict2.conditioner.ReMapkey'>}, 'padding_mask': {'output_key': 'padding_mask', 'dropout_rate': 0.0, 'dtype': None, 'input_key': 'padding_mask', '_target_': <class 'cosmos_predict2.conditioner.ReMapkey'>}, 'text': {'dropout_rate': 0.0, 'input_key': ['obs/language_embedding'], '_target_': <class 'cosmos_predict2.conditioner.TextAttr'>}, 'use_video_condition': {'output_key': 'use_video_condition', 'dropout_rate': 0.0, 'input_key': 'fps', '_target_': <class 'cosmos_predict2.conditioner.BooleanFlag'>}, '_target_': <class 'cosmos_predict2.conditioner.VideoConditioner'>}, 'conditioning_strategy': 'frame_replace', 'min_num_conditional_frames': 1, 'max_num_conditional_frames': 2, 'sigma_conditional': 0.0001, 'net': {'max_img_h': 240, 'max_img_w': 240, 'max_frames': 128, 'in_channels': 16, 'out_channels': 16, 'patch_spatial': 2, 'patch_temporal': 1, 'concat_padding_mask': True, 'model_channels': 2048, 'num_blocks': 28, 'num_heads': 16, 'atten_backend': 'minimal_a2a', 'pos_emb_cls': 'rope3d', 'pos_emb_learnable': True, 'pos_emb_interpolation': 'crop', 'use_adaln_lora': True, 'adaln_lora_dim': 256, 'rope_h_extrapolation_ratio': 3.0, 'rope_w_extrapolation_ratio': 3.0, 'rope_t_extrapolation_ratio': 1.0, 'extra_per_block_abs_pos_emb': False, 'rope_enable_fps_modulation': False, 'sac_config': {'mode': 'predict2_2b_720', 'every_n_blocks': 1, '_target_': 'cosmos_predict2.models.text2image_dit.SACConfig'}, '_target_': <class 'cosmos_predict2.models.video2world_dit.MinimalV1LVGDiT'>}, 'tokenizer': {'chunk_duration': 81, 'load_mean_std': False, 'temporal_window': 16, 'name': 'tokenizer', 'vae_pth': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth', '_target_': <class 'cosmos_predict2.tokenizers.tokenizer.TokenizerInterface'>}, 'guardrail_config': {'checkpoint_dir': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints', 'offload_model_to_cpu': True, 'enabled': False}, 'precision': 'bfloat16', 'rectified_flow_t_scaling_factor': 1.0, 'rectified_flow_loss_weight_uniform': True, 'resize_online': False, 'resolution': '480', 'ema': {'enabled': False, 'rate': 0.1, 'iteration_shift': 0, '_target_': 'cosmos_predict2.configs.defaults.ema.EMAConfig'}, 'sigma_data': 1.0, 'state_ch': 16, 'state_t': 16, 'text_encoder': {'cls': <TextEncoderClass.T5: 't5'>, 't5': {'ckpt_path': '/home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/text_encoder/t5-11b', 'num_tokens': 512, 'embed_dim': 1024}}, 'input_video_key': 'video', 'input_image_key': 'images', 'timestamps': {'nfe': 35, 't_min': 0.002, 't_max': 80.0, 'order': 7.0, 'is_forward': False}}, 'debug_without_randomness': False, 'fsdp_shard_size': 0, 'high_sigma_ratio': 0.05}, '_recursive_': False, '_target_': <class 'cosmos_predict2.models.video2world_model.Predict2Video2WorldModel'>}
* world2action_pipe: None
* optimizer: {'optim_type': 'fusedadam', 'model': None, 'lr': 4.445e-05, 'weight_decay': 0.1, 'betas': [0.9, 0.99], 'eps': 1e-08, 'master_weights': True, 'capturable': True, '_target_': <function get_base_optimizer at 0x79dcb0bc2b90>}
* scheduler: {'_target_': <class 'cosmos_predict2.configs.defaults.scheduler.ConstantScheduler'>}
* data_config: None
* video_dataset_train: {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}
* video_dataset_val: {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}
* dataloader_train: {'batch_size': 2, 'shuffle': None, 'sampler': {'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <function get_sampler at 0x79dcbfd48310>}, 'batch_sampler': None, 'num_workers': 12, 'collate_fn': None, 'pin_memory': True, 'drop_last': True, 'timeout': 0, 'worker_init_fn': None, 'multiprocessing_context': None, 'generator': None, 'prefetch_factor': 8, 'persistent_workers': True, 'pin_memory_device': '', 'in_order': False, 'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': False, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <class 'torch.utils.data.dataloader.DataLoader'>}
* dataloader_val: {'batch_size': 1, 'shuffle': None, 'sampler': {'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <function get_sampler at 0x79dcbfd48310>}, 'batch_sampler': None, 'num_workers': 0, 'collate_fn': None, 'pin_memory': False, 'drop_last': False, 'timeout': 0, 'worker_init_fn': None, 'multiprocessing_context': None, 'generator': None, 'prefetch_factor': None, 'persistent_workers': False, 'pin_memory_device': '', 'in_order': False, 'dataset': {'include_only_with_substrings': None, 'exclude_with_substring': None, 'data_fps': 30.0, 'obs_history': 5, 'is_multi_img': False, 'val_ratio': 0.0, 'dataset_dir': '/home/ubuntu/pdt-mimic/data/push-mimic', 'num_frames': 61, 'video_size': [480, 640], 'is_val': True, '_target_': <class 'cosmos_predict2.data.dataset_video.Dataset'>}, '_target_': <class 'torch.utils.data.dataloader.DataLoader'>}
* job:
   * project: posttraining
   * group: video2world
   * name: v2w_push_lora_rank32_lr1.778e-04_bsz32
* trainer:
   * type: <class 'imaginaire.trainer.ImaginaireTrainer'>
   * callbacks: {'ema': {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.EMAModelCallback'>}, 'progress_bar': {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.ProgressBarCallback'>}, 'low_prec': {'config': None, 'trainer': None, 'update_iter': 1, '_target_': <class 'imaginaire.utils.callback.LowPrecisionCallback'>}, 'iter_speed': {'hit_thres': 5, 'every_n': 1000, '_target_': <class 'cosmos_predict2.callbacks.iter_speed.IterSpeed'>}, 'device_monitor': {'every_n': 1000, 'step_size': 1, 'log_memory_detail': True, '_target_': <class 'cosmos_predict2.callbacks.device_monitor.DeviceMonitor'>}, 'manual_gc': {'warm_up': 5, 'every_n': 5, '_target_': <class 'imaginaire.callbacks.manual_gc.ManualGarbageCollection'>}, 'grad_clip': {'clip_norm': 10.0, 'force_finite': True, 'log_wandb': False, '_target_': <class 'cosmos_predict2.callbacks.grad_clip.GradClip'>}, 'video_eval': {'fuse_lora': True, '_target_': <class 'cosmos_predict2.callbacks.video_eval.VideoEvalCallback'>}}
   * distributed_parallelism: ddp
   * ddp:
      * find_unused_parameters: False
      * static_graph: True
      * broadcast_buffers: True
   * cudnn:
      * deterministic: False
      * benchmark: True
   * seed: 0
   * grad_scaler_args: {'enabled': False}
   * max_iter: 500
   * max_val_iter: None
   * logging_iter: 1000
   * run_validation: False
   * validation_iter: 999999999
   * timeout_period: 999999999
   * memory_format: torch.preserve_format
   * grad_accum_iter: 4
   * profiling:
      * enable_profiling: False
      * enable_memory_snapshot: False
      * profile_freq: 1
      * first_n_rank: 4
      * record_shape: True
      * profile_memory: True
      * with_stack: True
      * with_modules: True
* model_parallel: ModelParallelConfig(tensor_model_parallel_size=1, pipeline_model_parallel_comm_backend=None, pipeline_model_parallel_size=1, virtual_pipeline_model_parallel_size=None, sequence_parallel=False, context_parallel_size=1, hierarchical_context_parallel_sizes=None, expert_model_parallel_size=1, expert_tensor_parallel_size=1, moe_extended_tp=False, perform_initialization=True, use_cpu_initialization=False, fp16=False, bf16=False, params_dtype=torch.float32, timers=None, finalize_model_grads_func=None, grad_scale_func=None, no_sync_func=None, grad_sync_func=None, param_sync_func=None, deterministic_mode=False, enable_autocast=False, autocast_dtype=torch.float32, num_microbatches_with_partial_activation_checkpoints=None, gradient_accumulation_fusion=False, async_tensor_model_parallel_allreduce=False, use_te_rng_tracker=False, tp_comm_overlap=False, tp_comm_bulk_wgrad=True, tp_comm_bulk_dgrad=True, tp_comm_overlap_ag=True, tp_comm_overlap_rs=True, tp_comm_overlap_rs_dgrad=False, tp_comm_split_ag=True, tp_comm_atomic_ag=False, tp_comm_split_rs=True, tp_comm_atomic_rs=False, cross_entropy_loss_fusion=False, cross_entropy_fusion_impl='native', tp_comm_overlap_disable_qkv=False, tp_comm_overlap_disable_fc1=False, tp_comm_bootstrap_backend='nccl', pipeline_dtype=None, variable_seq_lengths=False, overlap_p2p_comm=False, batch_p2p_comm=True, batch_p2p_sync=True, use_ring_exchange_p2p=False, deallocate_pipeline_outputs=False, defer_embedding_wgrad_compute=False, wgrad_deferral_limit=0, pipeline_model_parallel_split_rank=None, overlap_p2p_comm_warmup_flush=False, microbatch_group_size_per_vp_stage=1, cpu_offloading=False, cpu_offloading_num_layers=0, _cpu_offloading_context=None, cpu_offloading_activations=False, cpu_offloading_weights=False, barrier_with_L1_time=True)
* checkpoint:
   * type: {'callbacks': None, '_target_': <class 'cosmos_predict2.checkpointer.Checkpointer'>}
   * dcp_async_mode_enabled: False
   * save_iter: 100
   * load_path: 
   * load_training_state: False
   * only_load_scheduler_state: False
   * strict_resume: True
   * jit:
      * enabled: False
      * input_shape: None
      * device: cuda
      * dtype: bfloat16
      * strict: True
   * verbose: True
   * keys_not_to_resume: []
   * broadcast_via_filesystem: False
   * load_ema_to_reg: False
   * dcp_allow_mismatched_size: False
* defaults: ['_self_', {'data_config': None}, {'video_dataset_train': None}, {'video_dataset_val': None}, {'dataloader_train': None}, {'dataloader_val': None}, {'world2action_pipe': None}, {'optimizer': 'fusedadamw'}, {'scheduler': 'constant'}, {'model': None}, {'callbacks': ['basic']}, {'net': None}, {'ema': None}, {'checkpoint': None}, {'ckpt_type': None}, {'experiment': None}]
[05-07 18:13:39|WARNING|imaginaire/utils/misc.py:127:print_environ_variables] Environment variable TORCH_HOME not set!
[05-07 18:13:39|INFO|imaginaire/utils/misc.py:125:print_environ_variables] Environment variable IMAGINAIRE_OUTPUT_ROOT: /home/ubuntu/checkpoints
[05-07 18:13:39|INFO|imaginaire/utils/misc.py:139:set_random_seed] Using random seed 0.
[05-07 18:13:39|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback ema: {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.EMAModelCallback'>}
[05-07 18:13:39|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback progress_bar: {'config': None, 'trainer': None, '_target_': <class 'imaginaire.utils.callback.ProgressBarCallback'>}
[05-07 18:13:39|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback low_prec: {'config': None, 'trainer': None, 'update_iter': 1, '_target_': <class 'imaginaire.utils.callback.LowPrecisionCallback'>}
[05-07 18:13:39|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback iter_speed: {'hit_thres': 5, 'every_n': 1000, '_target_': <class 'cosmos_predict2.callbacks.iter_speed.IterSpeed'>}
[05-07 18:13:39|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback device_monitor: {'every_n': 1000, 'step_size': 1, 'log_memory_detail': True, '_target_': <class 'cosmos_predict2.callbacks.device_monitor.DeviceMonitor'>}
[05-07 18:13:39|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback manual_gc: {'warm_up': 5, 'every_n': 5, '_target_': <class 'imaginaire.callbacks.manual_gc.ManualGarbageCollection'>}
[05-07 18:13:39|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback grad_clip: {'clip_norm': 10.0, 'force_finite': True, 'log_wandb': False, '_target_': <class 'cosmos_predict2.callbacks.grad_clip.GradClip'>}
[05-07 18:13:39|CRITICAL|imaginaire/utils/callback.py:77:__init__] Instantiating callback video_eval: {'fuse_lora': True, '_target_': <class 'cosmos_predict2.callbacks.video_eval.VideoEvalCallback'>}
[05-07 18:13:39|CRITICAL|cosmos_predict2/models/video2world_model.py:138:__init__] Using mean loss reduce with loss scale 100.0
[05-07 18:13:39|WARNING|cosmos_predict2/pipelines/video2world.py:277:from_config] precision torch.bfloat16
[05-07 18:13:40|INFO|cosmos_predict2/tokenizers/tokenizer.py:687:_video_vae] Loading /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth
[05-07 18:13:40|SUCCESS|cosmos_predict2/tokenizers/tokenizer.py:689:_video_vae] Successfully loaded /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//tokenizer/tokenizer.pth
[05-07 18:13:40|INFO|imaginaire/utils/distributed.py:430:sync_model_states] Synchronizing model states from rank 0 to all ranks in process group [0].
[05-07 18:13:40|INFO|cosmos_predict2/pipelines/video2world.py:335:from_config] Loading DiT from /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt
[05-07 18:13:45|SUCCESS|cosmos_predict2/pipelines/video2world.py:354:from_config] Successfully loaded DiT from /home/ubuntu/pdt-mimic/mimic-video/model/checkpoints/video_backbone//v2w_pretrained_cosmos.pt
[05-07 18:13:45|INFO|cosmos_predict2/models/video2world_model.py:321:add_lora_to_model] Adding LoRA adapters: rank=32, alpha=32, targets=['q_proj', 'k_proj', 'v_proj', 'output_proj', 'x_embedder.proj.1', 'linear_1', 'linear_2', 'mlp.layer1', 'mlp.layer2']
[05-07 18:13:46|INFO|cosmos_predict2/models/video2world_model.py:345:add_lora_to_model] LoRA injection successful: 46,336,256 trainable parameters out of 2,002,749,696 total (2.314%)
[05-07 18:13:46|INFO|cosmos_predict2/models/video2world_model.py:363:_log_lora_statistics] LoRA parameter breakdown:
[05-07 18:13:46|INFO|cosmos_predict2/models/video2world_model.py:365:_log_lora_statistics]   lora_A: 22,153,472 parameters
[05-07 18:13:46|INFO|cosmos_predict2/models/video2world_model.py:365:_log_lora_statistics]   lora_B: 24,182,784 parameters
[05-07 18:13:46|INFO|cosmos_predict2/models/video2world_model.py:366:_log_lora_statistics]   Total LoRA: 46,336,256 parameters
[05-07 18:13:46|INFO|cosmos_predict2/models/video2world_model.py:185:__init__] Total parameters: 2.00B, Frozen parameters: 1,778,679,808, Trainable parameters: 224,069,888
[05-07 18:13:46|INFO|cosmos_predict2/models/video2world_model.py:202:__init__] FSDP (Fully Sharded Data Parallel) is disabled.
[05-07 18:13:46|INFO|cosmos_predict2/data/dataset_video.py:102:__init__] 204 videos in total
[05-07 18:13:46|INFO|cosmos_predict2/data/dataset_video.py:112:__init__] 204 videos in train.
[05-07 18:13:46|INFO|cosmos_predict2/data/dataset_video.py:102:__init__] 204 videos in total
[05-07 18:13:46|INFO|cosmos_predict2/data/dataset_video.py:112:__init__] 204 videos in train.
[05-07 18:13:46|CRITICAL|cosmos_predict2/utils/optim_instantiate_dtensor.py:49:get_base_optimizer] total num parameters : 224,069,888
[05-07 18:13:46|WARNING|cosmos_predict2/utils/fused_adam_dtensor.py:103:__init__] FusedAdam master_weights: True capturable: True
[05-07 18:13:46|INFO|cosmos_predict2/checkpointer.py:206:load] Loading checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000100.pt
[05-07 18:13:48|SUCCESS|cosmos_predict2/checkpointer.py:208:load] Complete loading checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000100.pt
[05-07 18:13:48|INFO|cosmos_predict2/checkpointer.py:206:load] Loading checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000100.pt
[05-07 18:13:49|SUCCESS|cosmos_predict2/checkpointer.py:208:load] Complete loading checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000100.pt
[05-07 18:13:49|INFO|cosmos_predict2/checkpointer.py:206:load] Loading checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000100.pt
[05-07 18:13:49|SUCCESS|cosmos_predict2/checkpointer.py:208:load] Complete loading checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000100.pt
[05-07 18:13:49|INFO|cosmos_predict2/checkpointer.py:206:load] Loading checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000100.pt
[05-07 18:13:49|SUCCESS|cosmos_predict2/checkpointer.py:208:load] Complete loading checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000100.pt
[05-07 18:13:49|INFO|cosmos_predict2/checkpointer.py:212:load] - Loading the model...
[05-07 18:13:49|INFO|cosmos_predict2/checkpointer.py:260:load] - Loading the scheduler...
[05-07 18:13:49|INFO|cosmos_predict2/checkpointer.py:267:load] - Loading the optimizer...
[05-07 18:13:49|INFO|cosmos_predict2/checkpointer.py:280:load] - Loading the gradient scaler...
[05-07 18:13:49|SUCCESS|cosmos_predict2/checkpointer.py:282:load] Done with loading the checkpoint (iteration 100).
[05-07 18:13:49|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint loading: 3.4351 seconds
[05-07 18:13:49|CRITICAL|imaginaire/trainer.py:178:train] Distributed parallelism mode: ddp
[05-07 18:13:50|INFO|imaginaire/trainer.py:186:train] Starting training...
[05-07 18:13:50|INFO|cosmos_predict2/callbacks/device_monitor.py:92:on_train_start] DeviceMonitor callback: local_dir: /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/DeviceMonitor
[05-07 18:13:51|WARNING|imaginaire/utils/distributed.py:284:ddp_sync_grad] DDP static_graph=True is incompatible with sync_grad(). Performance will be reduced.
[05-07 18:14:13|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 101: Hit counter: 1/5 | Loss: 4.8886 | Time: 33.45s
[05-07 18:14:30|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 102: Hit counter: 2/5 | Loss: 5.8927 | Time: 17.23s
[05-07 18:14:49|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 103: Hit counter: 3/5 | Loss: 4.2789 | Time: 19.49s
[05-07 18:15:07|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 104: Hit counter: 4/5 | Loss: 3.6165 | Time: 17.32s
[05-07 18:15:24|INFO|cosmos_predict2/callbacks/iter_speed.py:51:on_training_step_end] Iteration 105: Hit counter: 5/5 | Loss: 4.6682 | Time: 17.27s
[05-07 18:21:09|CRITICAL|imaginaire/callbacks/manual_gc.py:48:every_n_impl] Garbage collection disabled
[05-07 18:42:54|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000200.pt
[05-07 18:43:03|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 12.1174 seconds
[05-07 18:43:04|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000200.pt
[05-07 18:43:07|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 3.6517 seconds
[05-07 18:43:07|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000200.pt
[05-07 18:43:07|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0020 seconds
[05-07 18:43:07|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000200.pt
[05-07 18:43:07|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0020 seconds
[05-07 19:12:02|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000300.pt
[05-07 19:12:09|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 9.9800 seconds
[05-07 19:12:10|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000300.pt
[05-07 19:12:12|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 3.4690 seconds
[05-07 19:12:12|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000300.pt
[05-07 19:12:12|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0021 seconds
[05-07 19:12:12|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000300.pt
[05-07 19:12:12|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0020 seconds
[05-07 19:41:07|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000400.pt
[05-07 19:41:14|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 10.2649 seconds
[05-07 19:41:16|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000400.pt
[05-07 19:41:18|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 3.5255 seconds
[05-07 19:41:18|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000400.pt
[05-07 19:41:18|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0020 seconds
[05-07 19:41:18|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000400.pt
[05-07 19:41:18|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0023 seconds
[05-07 20:10:12|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/model/iter_000000500.pt
[05-07 20:10:19|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 9.8569 seconds
[05-07 20:10:20|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/optim/iter_000000500.pt
[05-07 20:10:22|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 3.6050 seconds
[05-07 20:10:22|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/scheduler/iter_000000500.pt
[05-07 20:10:22|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0020 seconds
[05-07 20:10:22|SUCCESS|cosmos_predict2/checkpointer.py:139:_save_worker_local] Saved checkpoint (local): /home/ubuntu/checkpoints/posttraining/video2world/v2w_push_lora_rank32_lr1.778e-04_bsz32/checkpoints/trainer/iter_000000500.pt
[05-07 20:10:22|INFO|imaginaire/utils/misc.py:266:wrapper] Time spent on checkpoint saving (local): 0.0021 seconds
[05-07 20:10:22|SUCCESS|imaginaire/trainer.py:288:train] Done with training.