=== Avatar Training (fire-and-forget) ===
Script: train_pi0fast.py
Started: 2026-03-11 05:57:32 UTC

>>> [0/4] Raising file descriptor limit...
  ulimit -n = 65536
>>> [1/4] Setting environment...
  HF_TOKEN set: 37 chars
  WANDB_API_KEY set: 0 chars
>>> [2/4] Logging into HuggingFace...
The token has not been saved to the git credentials helper. Pass `add_to_git_credential=True` in this function directly or `--add-to-git-credential` if using via `hf`CLI if you want to set the git credential as well.
Token is valid (permission: write).
The token `beta-pc` has been saved to /root/.cache/huggingface/stored_tokens
Your token has been saved to /root/.cache/huggingface/token
Login successful.
Note: Environment variable`HF_TOKEN` is set and is the current active token independently from the token you've just configured.
⚠️  Warning: 'huggingface-cli login' is deprecated. Use 'hf auth login' instead.
>>> [3/4] Pre-downloading pretrained model...
  Downloading lerobot/pi0fast-base (default)...

Fetching 6 files:   0%|          | 0/6 [00:00<?, ?it/s]
Fetching 6 files:  67%|██████▋   | 4/6 [00:15<00:07,  3.87s/it]
Fetching 6 files: 100%|██████████| 6/6 [00:15<00:00,  2.58s/it]
  Cached at /root/.cache/huggingface/hub/models--lerobot--pi0fast-base/snapshots/0737e07fac1e63519c0bd38bd869934ecd9c3f86
  GPUs: 4x NVIDIA A100-SXM4-80GB
>>> [4/4] Launching accelerate with 4 GPUs...
  Script: train_pi0fast.py
  Train args: --dataset orange_hook_right_20260310 --steps 10000 --batch-size 8 --lr 5e-5 --save-freq 2000 --push-to-hub --num-workers 8 --task-text pick\ up\ the\ orange\ flat\ package 

The following values were not passed to `accelerate launch` and had defaults used instead:
	`--num_machines` was set to a value of `1`
	`--mixed_precision` was set to a value of `'no'`
	`--dynamo_backend` was set to a value of `'no'`
To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.

============================================================
  π₀-FAST Training
============================================================
  Dataset:     hamidavatar/orange_hook_right_20260310
  Pretrained:  lerobot/pi0fast-base
  Model repo:  hamidavatar/pi0fast-orange-hook-right-b32-20260311-0557
  Steps:       10000
  Batch/GPU:   8  (eff: 32)
  Chunk size:  10
  LR:          5e-05
  Save freq:   2000
  GPUs:        4
============================================================

lerobot_train \
  --dataset.repo_id=hamidavatar/orange_hook_right_20260310 \
  --policy.type=pi0_fast \
  --policy.pretrained_path=lerobot/pi0fast-base \
  --policy.repo_id=hamidavatar/pi0fast-orange-hook-right-b32-20260311-0557 \
  --policy.dtype=bfloat16 \
  --policy.gradient_checkpointing=true \
  --policy.chunk_size=10 \
  --policy.n_action_steps=10 \
  --policy.max_action_tokens=256 \
  --policy.optimizer_lr=5e-05 \
  --policy.optimizer_weight_decay=0.01 \
  --policy.scheduler_warmup_steps=1000 \
  --policy.scheduler_decay_steps=10000 \
  --policy.scheduler_decay_lr=1e-6 \
  --batch_size=8 \
  --steps=10000 \
  --save_freq=2000 \
  --log_freq=100 \
  --save_checkpoint=true \
  --seed=1000 \
  --num_workers=8 \
  --output_dir=outputs/train/pi0fast \
  --policy.push_to_hub=true \
  --policy.private=true \

[Avatar] cudnn.benchmark=True, tf32=True
[Avatar] Checkpoint push enabled → hamidavatar/pi0fast-orange-hook-right-b32-20260311-0557 (with per-step tags)
[Avatar] Dataset task text patched → "pick up the orange flat package"
[Avatar] Installed corrupt-frame retry handler
[Avatar] Checkpoint push enabled → hamidavatar/pi0fast-orange-hook-right-b32-20260311-0557 (with per-step tags)
[Avatar] Dataset task text patched → "pick up the orange flat package"
[Avatar] Installed corrupt-frame retry handler
[Avatar] Checkpoint push enabled → hamidavatar/pi0fast-orange-hook-right-b32-20260311-0557 (with per-step tags)
[Avatar] Dataset task text patched → "pick up the orange flat package"
[Avatar] Installed corrupt-frame retry handler
[Avatar] Checkpoint push enabled → hamidavatar/pi0fast-orange-hook-right-b32-20260311-0557 (with per-step tags)
[Avatar] Dataset task text patched → "pick up the orange flat package"
[Avatar] Installed corrupt-frame retry handler
WARNING:lerobot.configs.policies:Device 'None' is not available. Switching to 'cuda'.
WARNING:lerobot.configs.policies:Device 'None' is not available. Switching to 'cuda'.
WARNING:lerobot.configs.policies:Device 'None' is not available. Switching to 'cuda'.
WARNING:lerobot.configs.policies:Device 'None' is not available. Switching to 'cuda'.
INFO 2026-03-11 05:58:06 ot_train.py:197 {'batch_size': 8,
 'checkpoint_path': None,
 'dataset': {'episodes': None,
             'image_transforms': {'enable': False,
                                  'max_num_transforms': 3,
                                  'random_order': False,
                                  'tfs': {'affine': {'kwargs': {'degrees': [-5.0,
                                                                            5.0],
                                                                'translate': [0.05,
                                                                              0.05]},
                                                     'type': 'RandomAffine',
                                                     'weight': 1.0},
                                          'brightness': {'kwargs': {'brightness': [0.8,
                                                                                   1.2]},
                                                         'type': 'ColorJitter',
                                                         'weight': 1.0},
                                          'contrast': {'kwargs': {'contrast': [0.8,
                                                                               1.2]},
                                                       'type': 'ColorJitter',
                                                       'weight': 1.0},
                                          'hue': {'kwargs': {'hue': [-0.05,
                                                                     0.05]},
                                                  'type': 'ColorJitter',
                                                  'weight': 1.0},
                                          'saturation': {'kwargs': {'saturation': [0.5,
                                                                                   1.5]},
                                                         'type': 'ColorJitter',
                                                         'weight': 1.0},
                                          'sharpness': {'kwargs': {'sharpness': [0.5,
                                                                                 1.5]},
                                                        'type': 'SharpnessJitter',
                                                        'weight': 1.0}}},
             'repo_id': 'hamidavatar/orange_hook_right_20260310',
             'revision': None,
             'root': None,
             'streaming': False,
             'use_imagenet_stats': True,
             'video_backend': 'torchcodec'},
 'env': None,
 'eval': {'batch_size': 50, 'n_episodes': 50, 'use_async_envs': False},
 'eval_freq': 20000,
 'job_name': 'pi0_fast',
 'log_freq': 100,
 'num_workers': 8,
 'optimizer': {'betas': [0.9, 0.95],
               'eps': 1e-08,
               'grad_clip_norm': 1.0,
               'lr': 5e-05,
               'type': 'adamw',
               'weight_decay': 0.01},
 'output_dir': 'outputs/train/pi0fast',
 'peft': None,
 'policy': {'action_expert_variant': 'gemma_300m',
            'action_tokenizer_name': 'physical-intelligence/fast',
            'chunk_size': 10,
            'compile_mode': 'max-autotune',
            'compile_model': False,
            'device': 'cuda',
            'dtype': 'bfloat16',
            'empty_cameras': 0,
            'fast_skip_tokens': 128,
            'gradient_checkpointing': True,
            'image_resolution': [224, 224],
            'input_features': {},
            'license': None,
            'max_action_dim': 32,
            'max_action_tokens': 256,
            'max_decoding_steps': 256,
            'max_state_dim': 32,
            'n_action_steps': 10,
            'n_obs_steps': 1,
            'normalization_mapping': {'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>,
                                      'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>,
                                      'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>},
            'optimizer_betas': [0.9, 0.95],
            'optimizer_eps': 1e-08,
            'optimizer_grad_clip_norm': 1.0,
            'optimizer_lr': 5e-05,
            'optimizer_weight_decay': 0.01,
            'output_features': {},
            'paligemma_variant': 'gemma_2b',
            'pretrained_path': 'lerobot/pi0fast-base',
            'private': True,
            'push_to_hub': True,
            'repo_id': 'hamidavatar/pi0fast-orange-hook-right-b32-20260311-0557',
            'rtc_config': None,
            'scheduler_decay_lr': 1e-06,
            'scheduler_decay_steps': 10000,
            'scheduler_warmup_steps': 1000,
            'tags': None,
            'temperature': 0.0,
            'text_tokenizer_name': 'google/paligemma-3b-pt-224',
            'tokenizer_max_length': 200,
            'type': 'pi0_fast',
            'use_amp': False,
            'use_kv_cache': True,
            'use_peft': False,
            'validate_action_token_prefix': True},
 'rabc_epsilon': 1e-06,
 'rabc_head_mode': 'sparse',
 'rabc_kappa': 0.01,
 'rabc_progress_path': None,
 'rename_map': {},
 'resume': False,
 'save_checkpoint': True,
 'save_freq': 2000,
 'scheduler': {'decay_lr': 1e-06,
               'num_decay_steps': 10000,
               'num_warmup_steps': 1000,
               'peak_lr': 5e-05,
               'type': 'cosine_decay_with_warmup'},
 'seed': 1000,
 'steps': 10000,
 'tolerance_s': 0.0001,
 'use_policy_training_preset': True,
 'use_rabc': False,
 'wandb': {'disable_artifact': False,
           'enable': False,
           'entity': None,
           'mode': None,
           'notes': None,
           'project': 'lerobot',
           'run_id': None}}
INFO 2026-03-11 05:58:06 ot_train.py:205 Logs will be saved locally.
INFO 2026-03-11 05:58:06 ot_train.py:217 Creating dataset

Fetching 7 files:   0%|          | 0/7 [00:00<?, ?it/s]
Fetching 7 files:  29%|██▊       | 2/7 [00:00<00:00,  9.38it/s]
Fetching 7 files:  43%|████▎     | 3/7 [00:00<00:01,  3.84it/s]
Fetching 7 files: 100%|██████████| 7/7 [00:00<00:00, 10.16it/s]
INFO 2026-03-11 05:58:06 eo_utils.py:106 Using video codec: libsvtav1

Fetching 523 files:   0%|          | 0/523 [00:00<?, ?it/s]
Fetching 523 files:   0%|          | 1/523 [00:00<00:57,  9.15it/s]
Fetching 523 files:   0%|          | 2/523 [00:00<03:30,  2.47it/s]
Fetching 523 files:   4%|▎         | 19/523 [00:00<00:20, 24.63it/s]
Fetching 523 files:   4%|▍         | 23/523 [00:01<00:18, 26.44it/s]
Fetching 523 files:   5%|▌         | 27/523 [00:01<00:23, 21.08it/s]
Fetching 523 files:   6%|▌         | 30/523 [00:01<00:26, 18.38it/s]
Fetching 523 files:   6%|▋         | 33/523 [00:01<00:26, 18.35it/s]
Fetching 523 files:   7%|▋         | 36/523 [00:01<00:24, 19.93it/s]
Fetching 523 files:   8%|▊         | 41/523 [00:02<00:18, 25.54it/s]
Fetching 523 files:   9%|▊         | 45/523 [00:02<00:34, 13.73it/s]
Fetching 523 files:  10%|▉         | 52/523 [00:02<00:23, 20.18it/s]
Fetching 523 files:  11%|█         | 56/523 [00:03<00:28, 16.40it/s]
Fetching 523 files:  11%|█▏        | 60/523 [00:03<00:27, 16.73it/s]
Fetching 523 files:  12%|█▏        | 63/523 [00:03<00:26, 17.48it/s]
Fetching 523 files:  13%|█▎        | 68/523 [00:03<00:20, 22.58it/s]
Fetching 523 files:  14%|█▍        | 72/523 [00:03<00:24, 18.29it/s]
Fetching 523 files:  14%|█▍        | 75/523 [00:04<00:22, 19.98it/s]
Fetching 523 files:  15%|█▍        | 78/523 [00:04<00:25, 17.17it/s]
Fetching 523 files:  15%|█▌        | 81/523 [00:04<00:22, 19.32it/s]
Fetching 523 files:  16%|█▌        | 84/523 [00:04<00:24, 17.80it/s]
Fetching 523 files:  17%|█▋        | 87/523 [00:04<00:23, 18.41it/s]
Fetching 523 files:  18%|█▊        | 92/523 [00:05<00:28, 15.04it/s]
Fetching 523 files:  18%|█▊        | 94/523 [00:05<00:31, 13.58it/s]
Fetching 523 files:  19%|█▉        | 101/523 [00:05<00:24, 17.02it/s]
Fetching 523 files:  20%|█▉        | 104/523 [00:05<00:23, 18.13it/s]
Fetching 523 files:  21%|██        | 108/523 [00:05<00:20, 20.54it/s]
Fetching 523 files:  21%|██▏       | 112/523 [00:06<00:26, 15.48it/s]
Fetching 523 files:  22%|██▏       | 117/523 [00:06<00:20, 20.19it/s]
Fetching 523 files:  23%|██▎       | 120/523 [00:06<00:24, 16.47it/s]
Fetching 523 files:  24%|██▎       | 123/523 [00:06<00:22, 17.91it/s]
Fetching 523 files:  24%|██▍       | 126/523 [00:06<00:21, 18.58it/s]
Fetching 523 files:  25%|██▍       | 129/523 [00:07<00:23, 16.51it/s]
Fetching 523 files:  25%|██▌       | 132/523 [00:07<00:24, 15.94it/s]
Fetching 523 files:  26%|██▌       | 135/523 [00:07<00:21, 18.31it/s]
Fetching 523 files:  26%|██▋       | 138/523 [00:07<00:23, 16.29it/s]
Fetching 523 files:  27%|██▋       | 140/523 [00:07<00:26, 14.57it/s]
Fetching 523 files:  28%|██▊       | 145/523 [00:08<00:22, 16.56it/s]
Fetching 523 files:  29%|██▊       | 150/523 [00:08<00:21, 17.51it/s]
Fetching 523 files:  29%|██▉       | 153/523 [00:08<00:19, 18.90it/s]
Fetching 523 files:  30%|██▉       | 156/523 [00:08<00:18, 19.68it/s]
Fetching 523 files:  30%|███       | 159/523 [00:09<00:26, 13.66it/s]
Fetching 523 files:  31%|███▏      | 164/523 [00:09<00:19, 18.75it/s]
Fetching 523 files:  32%|███▏      | 167/523 [00:09<00:20, 17.69it/s]
Fetching 523 files:  33%|███▎      | 170/523 [00:09<00:20, 17.62it/s]
Fetching 523 files:  33%|███▎      | 173/523 [00:09<00:20, 16.69it/s]
Fetching 523 files:  34%|███▎      | 176/523 [00:09<00:19, 18.04it/s]
Fetching 523 files:  34%|███▍      | 179/523 [00:10<00:24, 14.09it/s]
Fetching 523 files:  35%|███▌      | 184/523 [00:10<00:19, 17.41it/s]
Fetching 523 files:  36%|███▌      | 187/523 [00:10<00:18, 18.55it/s]
Fetching 523 files:  36%|███▋      | 190/523 [00:10<00:20, 16.31it/s]
Fetching 523 files:  37%|███▋      | 192/523 [00:11<00:27, 12.01it/s]
Fetching 523 files:  38%|███▊      | 198/523 [00:11<00:18, 17.28it/s]
Fetching 523 files:  38%|███▊      | 201/523 [00:11<00:21, 15.14it/s]
Fetching 523 files:  39%|███▉      | 203/523 [00:11<00:25, 12.39it/s]
Fetching 523 files:  39%|███▉      | 206/523 [00:12<00:22, 13.92it/s]
Fetching 523 files:  40%|████      | 211/523 [00:12<00:19, 15.70it/s]
Fetching 523 files:  41%|████      | 213/523 [00:12<00:19, 16.10it/s]
Fetching 523 files:  41%|████▏     | 216/523 [00:12<00:17, 17.36it/s]
Fetching 523 files:  42%|████▏     | 219/523 [00:12<00:15, 19.46it/s]
Fetching 523 files:  42%|████▏     | 222/523 [00:12<00:15, 19.84it/s]
Fetching 523 files:  43%|████▎     | 225/523 [00:13<00:18, 16.45it/s]
Fetching 523 files:  44%|████▎     | 228/523 [00:13<00:18, 15.78it/s]
Fetching 523 files:  44%|████▍     | 230/523 [00:13<00:22, 13.03it/s]
Fetching 523 files:  45%|████▍     | 234/523 [00:13<00:19, 15.11it/s]
Fetching 523 files:  45%|████▌     | 237/523 [00:13<00:19, 14.87it/s]
Fetching 523 files:  46%|████▌     | 240/523 [00:14<00:16, 17.25it/s]
Fetching 523 files:  46%|████▋     | 243/523 [00:14<00:15, 17.87it/s]
Fetching 523 files:  47%|████▋     | 246/523 [00:14<00:15, 18.21it/s]
Fetching 523 files:  47%|████▋     | 248/523 [00:14<00:17, 15.60it/s]
Fetching 523 files:  48%|████▊     | 251/523 [00:14<00:14, 18.16it/s]
Fetching 523 files:  48%|████▊     | 253/523 [00:14<00:16, 16.75it/s]
Fetching 523 files:  49%|████▉     | 256/523 [00:14<00:15, 17.42it/s]
Fetching 523 files:  49%|████▉     | 258/523 [00:15<00:17, 15.53it/s]
Fetching 523 files:  50%|█████     | 264/523 [00:15<00:13, 19.50it/s]
Fetching 523 files:  51%|█████     | 266/523 [00:15<00:13, 19.30it/s]
Fetching 523 files:  51%|█████     | 268/523 [00:15<00:13, 19.32it/s]
Fetching 523 files:  52%|█████▏    | 272/523 [00:15<00:10, 23.64it/s]
Fetching 523 files:  53%|█████▎    | 275/523 [00:15<00:13, 18.04it/s]
Fetching 523 files:  53%|█████▎    | 278/523 [00:16<00:12, 19.16it/s]
Fetching 523 files:  54%|█████▍    | 282/523 [00:16<00:13, 17.40it/s]
Fetching 523 files:  54%|█████▍    | 284/523 [00:16<00:13, 17.56it/s]
Fetching 523 files:  55%|█████▍    | 287/523 [00:16<00:17, 13.57it/s]
Fetching 523 files:  55%|█████▌    | 290/523 [00:16<00:15, 14.98it/s]
Fetching 523 files:  57%|█████▋    | 296/523 [00:17<00:14, 15.54it/s]
Fetching 523 files:  57%|█████▋    | 300/523 [00:17<00:13, 16.14it/s]
Fetching 523 files:  59%|█████▉    | 308/523 [00:17<00:08, 25.25it/s]
Fetching 523 files:  60%|█████▉    | 312/523 [00:18<00:10, 19.26it/s]
Fetching 523 files:  60%|██████    | 315/523 [00:18<00:10, 19.73it/s]
Fetching 523 files:  61%|██████    | 319/523 [00:18<00:11, 17.23it/s]
Fetching 523 files:  62%|██████▏   | 325/523 [00:18<00:09, 20.29it/s]
Fetching 523 files:  63%|██████▎   | 328/523 [00:18<00:10, 18.78it/s]
Fetching 523 files:  63%|██████▎   | 331/523 [00:18<00:09, 20.48it/s]
Fetching 523 files:  64%|██████▍   | 334/523 [00:19<00:10, 17.90it/s]
Fetching 523 files:  64%|██████▍   | 337/523 [00:19<00:09, 19.03it/s]
Fetching 523 files:  65%|██████▌   | 341/523 [00:19<00:09, 18.20it/s]
Fetching 523 files:  66%|██████▌   | 343/523 [00:19<00:10, 16.47it/s]
Fetching 523 files:  67%|██████▋   | 349/523 [00:19<00:07, 22.22it/s]
Fetching 523 files:  67%|██████▋   | 352/523 [00:20<00:09, 17.80it/s]
Fetching 523 files:  68%|██████▊   | 354/523 [00:20<00:11, 15.19it/s]
Fetching 523 files:  69%|██████▉   | 361/523 [00:20<00:08, 20.15it/s]
Fetching 523 files:  70%|███████   | 367/523 [00:20<00:06, 24.21it/s]
Fetching 523 files:  71%|███████   | 370/523 [00:20<00:06, 23.26it/s]
Fetching 523 files:  72%|███████▏  | 375/523 [00:21<00:05, 27.91it/s]
Fetching 523 files:  72%|███████▏  | 379/523 [00:21<00:05, 24.14it/s]
Fetching 523 files:  73%|███████▎  | 382/523 [00:21<00:05, 23.92it/s]
Fetching 523 files:  74%|███████▍  | 386/523 [00:21<00:05, 24.25it/s]
Fetching 523 files:  74%|███████▍  | 389/523 [00:21<00:06, 19.27it/s]
Fetching 523 files:  75%|███████▌  | 393/523 [00:21<00:05, 22.38it/s]
Fetching 523 files:  76%|███████▌  | 396/523 [00:22<00:05, 22.64it/s]
Fetching 523 files:  76%|███████▋  | 399/523 [00:22<00:08, 14.83it/s]
Fetching 523 files:  78%|███████▊  | 406/523 [00:22<00:05, 22.48it/s]
Fetching 523 files:  78%|███████▊  | 409/523 [00:22<00:06, 18.35it/s]
Fetching 523 files:  79%|███████▉  | 412/523 [00:23<00:07, 15.17it/s]
Fetching 523 files:  79%|███████▉  | 414/523 [00:23<00:07, 14.24it/s]
Fetching 523 files:  80%|████████  | 420/523 [00:23<00:05, 20.24it/s]
Fetching 523 files:  81%|████████  | 423/523 [00:23<00:05, 19.24it/s]
Fetching 523 files:  81%|████████▏ | 426/523 [00:23<00:05, 18.15it/s]
Fetching 523 files:  82%|████████▏ | 429/523 [00:24<00:07, 12.68it/s]
Fetching 523 files:  83%|████████▎ | 435/523 [00:24<00:04, 18.23it/s]
Fetching 523 files:  84%|████████▎ | 438/523 [00:24<00:04, 17.80it/s]
Fetching 523 files:  84%|████████▍ | 441/523 [00:24<00:05, 14.23it/s]
Fetching 523 files:  85%|████████▍ | 443/523 [00:25<00:05, 14.39it/s]
Fetching 523 files:  86%|████████▌ | 451/523 [00:25<00:03, 22.03it/s]
Fetching 523 files:  87%|████████▋ | 456/523 [00:25<00:02, 26.01it/s]
Fetching 523 files:  88%|████████▊ | 459/523 [00:25<00:04, 15.96it/s]
Fetching 523 files:  89%|████████▊ | 463/523 [00:25<00:03, 18.59it/s]
Fetching 523 files:  89%|████████▉ | 467/523 [00:26<00:02, 20.33it/s]
Fetching 523 files:  90%|████████▉ | 470/523 [00:26<00:03, 15.16it/s]
Fetching 523 files:  91%|█████████ | 476/523 [00:26<00:02, 20.68it/s]
Fetching 523 files:  92%|█████████▏| 479/523 [00:26<00:02, 20.69it/s]
Fetching 523 files:  92%|█████████▏| 483/523 [00:26<00:02, 18.93it/s]
Fetching 523 files:  93%|█████████▎| 487/523 [00:27<00:01, 21.74it/s]
Fetching 523 files:  94%|█████████▍| 493/523 [00:27<00:01, 20.94it/s]
Fetching 523 files:  95%|█████████▌| 499/523 [00:27<00:00, 25.23it/s]
Fetching 523 files:  96%|█████████▌| 502/523 [00:27<00:00, 22.17it/s]
Fetching 523 files:  97%|█████████▋| 506/523 [00:28<00:00, 17.31it/s]
Fetching 523 files:  97%|█████████▋| 509/523 [00:28<00:00, 17.03it/s]
Fetching 523 files:  99%|█████████▉| 517/523 [00:28<00:00, 25.93it/s]
Fetching 523 files: 100%|█████████▉| 521/523 [00:28<00:00, 26.20it/s]
Fetching 523 files: 100%|██████████| 523/523 [00:28<00:00, 18.26it/s]
NCCL version 2.26.2+cuda12.2
INFO 2026-03-11 05:58:36 ot_train.py:235 Creating policy
The PI0Fast model is a direct port of the OpenPI implementation. 
This implementation follows the original OpenPI structure for compatibility. 
Original implementation: https://github.com/Physical-Intelligence/openpi
The PI0Fast model is a direct port of the OpenPI implementation. 
This implementation follows the original OpenPI structure for compatibility. 
Original implementation: https://github.com/Physical-Intelligence/openpi
The PI0Fast model is a direct port of the OpenPI implementation. 
This implementation follows the original OpenPI structure for compatibility. 
Original implementation: https://github.com/Physical-Intelligence/openpi
The PI0Fast model is a direct port of the OpenPI implementation. 
This implementation follows the original OpenPI structure for compatibility. 
Original implementation: https://github.com/Physical-Intelligence/openpi
A new version of the following files was downloaded from https://huggingface.co/physical-intelligence/fast:
- processing_action_tokenizer.py
. Make sure to double-check they do not contain any added malicious code. To avoid downloading new versions of the code file, you can pin a revision.
INFO 2026-03-11 05:58:38 pi0_fast.py:837 Loaded FAST tokenizer for action detokenization
INFO 2026-03-11 05:59:22 pi0_fast.py:329 Enabled gradient checkpointing for PI0FastPytorch model
Loading model from: lerobot/pi0fast-base
✓ Loaded state dict from model.safetensors
WARNING 2026-03-11 05:59:23 pi0_fast.py:991 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
WARNING 2026-03-11 05:59:23 pi0_fast.py:991 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
Loading model from: lerobot/pi0fast-base
✓ Loaded state dict from model.safetensors
Loading model from: lerobot/pi0fast-base
✓ Loaded state dict from model.safetensors
Loading model from: lerobot/pi0fast-base
✓ Loaded state dict from model.safetensors
All keys loaded successfully!
All keys loaded successfully!
All keys loaded successfully!
All keys loaded successfully!
INFO 2026-03-11 05:59:32 ot_train.py:290 Creating optimizer and scheduler
INFO 2026-03-11 05:59:32 ot_train.py:325 Output dir: outputs/train/pi0fast
INFO 2026-03-11 05:59:32 ot_train.py:332 cfg.steps=10000 (10K)
INFO 2026-03-11 05:59:32 ot_train.py:333 dataset.num_frames=75507 (76K)
INFO 2026-03-11 05:59:32 ot_train.py:334 dataset.num_episodes=257
INFO 2026-03-11 05:59:32 ot_train.py:337 Effective batch size: 8 x 4 = 32
INFO 2026-03-11 05:59:32 ot_train.py:338 num_learnable_params=2923335408 (3B)
INFO 2026-03-11 05:59:32 ot_train.py:339 num_total_params=2923335408 (3B)

Training:   0%|          | 0/10000 [00:00<?, ?step/s]INFO 2026-03-11 05:59:32 ot_train.py:403 Start offline training on a fixed dataset, with effective batch size: 32
[rank3]: Traceback (most recent call last):
[rank3]:   File "/tmp/train_pi0fast.py", line 313, in <module>
[rank3]:     main()
[rank3]:   File "/tmp/train_pi0fast.py", line 309, in main
[rank3]:     lerobot_main()
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/lerobot/scripts/lerobot_train.py", line 547, in main
[rank3]:     train()
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/lerobot/configs/parser.py", line 233, in wrapper_inner
[rank3]:     response = fn(cfg, *args, **kwargs)
[rank3]:                ^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/lerobot/scripts/lerobot_train.py", line 409, in train
[rank3]:     batch = next(dl_iter)
[rank3]:             ^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/lerobot/datasets/utils.py", line 911, in cycle
[rank3]:     yield next(iterator)
[rank3]:           ^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/accelerate/data_loader.py", line 577, in __iter__
[rank3]:     current_batch = next(dataloader_iter)
[rank3]:                     ^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/torch/utils/data/dataloader.py", line 733, in __next__
[rank3]:     data = self._next_data()
[rank3]:            ^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/torch/utils/data/dataloader.py", line 1515, in _next_data
[rank3]:     return self._process_data(data, worker_id)
[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/torch/utils/data/dataloader.py", line 1550, in _process_data
[rank3]:     data.reraise()
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/torch/_utils.py", line 750, in reraise
[rank3]:     raise exception
[rank3]: lerobot.datasets.video_utils.FrameTimestampError: Caught FrameTimestampError in DataLoader worker process 0.
[rank3]: Original Traceback (most recent call last):
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/torch/utils/data/_utils/worker.py", line 349, in _worker_loop
[rank3]:     data = fetcher.fetch(index)  # type: ignore[possibly-undefined]
[rank3]:            ^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/torch/utils/data/_utils/fetch.py", line 52, in fetch
[rank3]:     data = [self.dataset[idx] for idx in possibly_batched_index]
[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/torch/utils/data/_utils/fetch.py", line 52, in <listcomp>
[rank3]:     data = [self.dataset[idx] for idx in possibly_batched_index]
[rank3]:             ~~~~~~~~~~~~^^^^^
[rank3]:   File "/tmp/train_pi0fast.py", line 107, in _safe_getitem
[rank3]:     result = _orig_getitem(self, idx)
[rank3]:              ^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/tmp/train_pi0fast.py", line 63, in _getitem_with_text
[rank3]:     item = _orig_getitem(self, idx)
[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/lerobot/datasets/lerobot_dataset.py", line 1101, in __getitem__
[rank3]:     video_frames = self._query_videos(query_timestamps, ep_idx)
[rank3]:                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/lerobot/datasets/lerobot_dataset.py", line 1064, in _query_videos
[rank3]:     frames = decode_video_frames(video_path, shifted_query_ts, self.tolerance_s, self.video_backend)
[rank3]:              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/lerobot/datasets/video_utils.py", line 150, in decode_video_frames
[rank3]:     return decode_video_frames_torchcodec(video_path, timestamps, tolerance_s)
[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]:   File "/usr/local/lib/python3.11/dist-packages/lerobot/datasets/video_utils.py", line 362, in decode_video_frames_torchcodec
[rank3]:     raise FrameTimestampError(
[rank3]: lerobot.datasets.video_utils.FrameTimestampError: One or several query timestamps unexpectedly violate the tolerance (tensor([1.3333]) > tolerance_s=0.0001). It means that the closest frame that can be loaded from the video is too far away in time. This might be due to synchronization issues with timestamps during data collection. To be safe, we advise to ignore this item during training.
[rank3]: queried timestamps: tensor([0.3000])
[rank3]: loaded timestamps: tensor([1.6333])
[rank3]: video: /root/.cache/huggingface/lerobot/hamidavatar/orange_hook_right_20260310/videos/observation.images.wrist/chunk-000/file-209.mp4

W0311 05:59:36.024000 2392 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2460 closing signal SIGTERM
W0311 05:59:36.025000 2392 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2461 closing signal SIGTERM
W0311 05:59:36.025000 2392 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2462 closing signal SIGTERM
E0311 05:59:36.030000 2392 torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 3 (pid: 2463) of binary: /usr/bin/python
Traceback (most recent call last):
  File "/usr/local/bin/accelerate", line 8, in <module>
    sys.exit(main())
             ^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/accelerate/commands/accelerate_cli.py", line 50, in main
    args.func(args)
  File "/usr/local/lib/python3.11/dist-packages/accelerate/commands/launch.py", line 1396, in launch_command
    multi_gpu_launcher(args)
  File "/usr/local/lib/python3.11/dist-packages/accelerate/commands/launch.py", line 1023, in multi_gpu_launcher
    distrib_run.run(args)
  File "/usr/local/lib/python3.11/dist-packages/torch/distributed/run.py", line 883, in run
    elastic_launch(
  File "/usr/local/lib/python3.11/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
/tmp/train_pi0fast.py FAILED
------------------------------------------------------------
Failures:
  <NO_OTHER_FAILURES>
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2026-03-11_05:59:36
  host      : 9fcf7db9d97f
  rank      : 3 (local_rank: 3)
  exitcode  : 1 (pid: 2463)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
============================================================

>>> Training finished with exit code 1
Ended: 2026-03-11 05:59:37 UTC
>>> Saving training log to HuggingFace...