=== Avatar Training (fire-and-forget) === Script: train_pi0fast.py Started: 2026-03-20 06:12:38 UTC >>> [0/4] Raising file descriptor limit... ulimit -n = 1048576 >>> [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 avatarrobotics/pi0fast-alpha-bot-right-b32-20260315-2001 @ step-90000... Fetching 9 files: 0%| | 0/9 [00:00>> [3.5/4] Pre-downloading dataset (single process)... Dataset key: alpha_bot_right_combined Downloading avatarrobotics/alpha-bot-right-combined... Fetching 12927 files: 0%| | 0/12927 [00:00>> [4/4] Launching accelerate with 2 GPUs... Script: train_pi0fast.py Train args: --dataset alpha_bot_right_combined --pretrained-path avatarrobotics/pi0fast-alpha-bot-right-b32-20260315-2001:step-90000 --steps 310000 --batch-size 2 --lr 5e-5 --chunk-size 10 --save-freq 30000 --warmup-steps 100 --push-to-hub 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: avatarrobotics/alpha-bot-right-combined Pretrained: avatarrobotics/pi0fast-alpha-bot-right-b32-20260315-2001:step-90000 Model repo: avatarrobotics/pi0fast-alpha-bot-right-b4-20260320-0645 Steps: 310000 Batch/GPU: 2 (eff: 4) Chunk size: 10 LR: 5e-05 Save freq: 30000 GPUs: 2 ============================================================ [Avatar] Downloading checkpoint avatarrobotics/pi0fast-alpha-bot-right-b32-20260315-2001 @ step-90000 ... Fetching 9 files: 0%| | 0/9 [00:00 center_crop_resize (no padding waste) [Avatar] Forced video_backend='pyav' (CPU-safe) [Avatar] Dataset pre-downloaded by launcher script [Avatar] Checkpoint push enabled → avatarrobotics/pi0fast-alpha-bot-right-b4-20260320-0645 (with per-step tags) [Avatar] Dataset task text patched (multi-task from tasks.json) [Avatar] Dataset timestamp tolerance set to 2.0s [Avatar] Installed corrupt-frame retry handler [Avatar] Patched resize_with_pad_torch -> center_crop_resize (no padding waste) [Avatar] Forced video_backend='pyav' (CPU-safe) WARNING:lerobot.configs.policies:Device 'None' is not available. Switching to 'cuda'. INFO 2026-03-20 06:45:51 ot_train.py:197 {'batch_size': 2, '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': 'avatarrobotics/alpha-bot-right-combined', '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': 4, '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': , 'STATE': , 'VISUAL': }, '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': '/tmp/pretrained_checkpoint', 'private': True, 'push_to_hub': True, 'repo_id': 'avatarrobotics/pi0fast-alpha-bot-right-b4-20260320-0645', 'rtc_config': None, 'scheduler_decay_lr': 1e-06, 'scheduler_decay_steps': 310000, 'scheduler_warmup_steps': 100, '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': 30000, 'scheduler': {'decay_lr': 1e-06, 'num_decay_steps': 310000, 'num_warmup_steps': 100, 'peak_lr': 5e-05, 'type': 'cosine_decay_with_warmup'}, 'seed': 1000, 'steps': 310000, 'tolerance_s': 2.0, '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 2026-03-20 06:45:51 ot_train.py:205 Logs will be saved locally. INFO 2026-03-20 06:45:51 ot_train.py:217 Creating dataset INFO 2026-03-20 06:45:51 eo_utils.py:106 Using video codec: libsvtav1 WARNING:lerobot.configs.policies:Device 'None' is not available. Switching to 'cuda'. Fetching 12926 files: 0%| | 0/12926 [00:00 tolerance_s=2, retrying with idx=467651 [rank0]:[W320 06:53:43.952835980 reducer.cpp:1430] Warning: find_unused_parameters=True was specified in DDP constructor, but did not find any unused parameters in the forward pass. This flag results in an extra traversal of the autograd graph every iteration, which can adversely affect performance. If your model indeed never has any unused parameters in the forward pass, consider turning this flag off. Note that this warning may be a false positive if your model has flow control causing later iterations to have unused parameters. (function operator()) [rank1]:[W320 06:53:44.691872194 reducer.cpp:1430] Warning: find_unused_parameters=True was specified in DDP constructor, but did not find any unused parameters in the forward pass. This flag results in an extra traversal of the autograd graph every iteration, which can adversely affect performance. If your model indeed never has any unused parameters in the forward pass, consider turning this flag off. Note that this warning may be a false positive if your model has flow control causing later iterations to have unused parameters. (function operator()) [rank0]: Traceback (most recent call last): [rank0]: File "/tmp/train_pi0fast.py", line 486, in [rank0]: main() [rank0]: File "/tmp/train_pi0fast.py", line 482, in main [rank0]: lerobot_main() [rank0]: File "/usr/local/lib/python3.11/dist-packages/lerobot/scripts/lerobot_train.py", line 547, in main [rank0]: train() [rank0]: File "/usr/local/lib/python3.11/dist-packages/lerobot/configs/parser.py", line 233, in wrapper_inner [rank0]: response = fn(cfg, *args, **kwargs) [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^ [rank0]: File "/usr/local/lib/python3.11/dist-packages/lerobot/scripts/lerobot_train.py", line 413, in train [rank0]: train_tracker, output_dict = update_policy( [rank0]: ^^^^^^^^^^^^^^ [rank0]: File "/usr/local/lib/python3.11/dist-packages/lerobot/scripts/lerobot_train.py", line 134, in update_policy [rank0]: optimizer.step() [rank0]: File "/usr/local/lib/python3.11/dist-packages/accelerate/optimizer.py", line 179, in step [rank0]: self.optimizer.step(closure) [rank0]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/lr_scheduler.py", line 124, in wrapper [rank0]: return func.__get__(opt, opt.__class__)(*args, **kwargs) [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ [rank0]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/optimizer.py", line 485, in wrapper [rank0]: out = func(*args, **kwargs) [rank0]: ^^^^^^^^^^^^^^^^^^^^^ [rank0]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/optimizer.py", line 79, in _use_grad [rank0]: ret = func(self, *args, **kwargs) [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^ [rank0]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/adam.py", line 236, in step [rank0]: has_complex = self._init_group( [rank0]: ^^^^^^^^^^^^^^^^^ [rank0]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/adam.py", line 176, in _init_group [rank0]: state["exp_avg"] = torch.zeros_like( [rank0]: ^^^^^^^^^^^^^^^^^ [rank0]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 10.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.72 GiB is allocated by PyTorch, and 202.22 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables) [rank1]: Traceback (most recent call last): [rank1]: File "/tmp/train_pi0fast.py", line 486, in [rank1]: main() [rank1]: File "/tmp/train_pi0fast.py", line 482, in main [rank1]: lerobot_main() [rank1]: File "/usr/local/lib/python3.11/dist-packages/lerobot/scripts/lerobot_train.py", line 547, in main [rank1]: train() [rank1]: File "/usr/local/lib/python3.11/dist-packages/lerobot/configs/parser.py", line 233, in wrapper_inner [rank1]: response = fn(cfg, *args, **kwargs) [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^ [rank1]: File "/usr/local/lib/python3.11/dist-packages/lerobot/scripts/lerobot_train.py", line 413, in train [rank1]: train_tracker, output_dict = update_policy( [rank1]: ^^^^^^^^^^^^^^ [rank1]: File "/usr/local/lib/python3.11/dist-packages/lerobot/scripts/lerobot_train.py", line 134, in update_policy [rank1]: optimizer.step() [rank1]: File "/usr/local/lib/python3.11/dist-packages/accelerate/optimizer.py", line 179, in step [rank1]: self.optimizer.step(closure) [rank1]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/lr_scheduler.py", line 124, in wrapper [rank1]: return func.__get__(opt, opt.__class__)(*args, **kwargs) [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ [rank1]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/optimizer.py", line 485, in wrapper [rank1]: out = func(*args, **kwargs) [rank1]: ^^^^^^^^^^^^^^^^^^^^^ [rank1]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/optimizer.py", line 79, in _use_grad [rank1]: ret = func(self, *args, **kwargs) [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^ [rank1]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/adam.py", line 236, in step [rank1]: has_complex = self._init_group( [rank1]: ^^^^^^^^^^^^^^^^^ [rank1]: File "/usr/local/lib/python3.11/dist-packages/torch/optim/adam.py", line 176, in _init_group [rank1]: state["exp_avg"] = torch.zeros_like( [rank1]: ^^^^^^^^^^^^^^^^^ [rank1]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 64.00 MiB. GPU 1 has a total capacity of 23.53 GiB of which 10.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.72 GiB is allocated by PyTorch, and 202.22 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables) Training: 0%| | 0/310000 [00:23 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: ------------------------------------------------------------ Root Cause (first observed failure): [0]: time : 2026-03-20_06:53:55 host : c743c5135bc4 rank : 0 (local_rank: 0) exitcode : 1 (pid: 15415) error_file: traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html ============================================================ >>> Training finished with exit code 1 Ended: 2026-03-20 06:53:57 UTC >>> Saving training log to HuggingFace...