| subset angkul07/abc-teleop -> /workspace/data/yam7h/teleop: 8 episodes, 2214 frames, 3 cams |
| Uninstalled 1 package in 3ms |
| Installed 40 packages in 437ms |
| 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. |
|
Generating train split: 0 examples [00:00, ? examples/s]
Generating train split: 2214 examples [00:00, 281433.78 examples/s] |
|
Generating train split: 0 examples [00:00, ? examples/s]
Generating train split: 527 examples [00:00, 128788.57 examples/s] |
|
Computing stats: 0%| | 0/62 [00:00<?, ?it/s]
Computing stats: 2%|▏ | 1/62 [00:06<06:46, 6.67s/it]
Computing stats: 21%|██ | 13/62 [00:06<00:18, 2.66it/s]
Computing stats: 42%|████▏ | 26/62 [00:06<00:05, 6.27it/s]
Computing stats: 55%|█████▍ | 34/62 [00:07<00:03, 9.07it/s]
Computing stats: 68%|██████▊ | 42/62 [00:07<00:01, 12.61it/s]
Computing stats: 81%|████████ | 50/62 [00:07<00:00, 16.92it/s]
Computing stats: 94%|█████████▎| 58/62 [00:07<00:00, 21.93it/s]
Computing stats: 100%|██████████| 62/62 [00:07<00:00, 8.27it/s] |
| Writing stats to: /workspace/openpi/assets/pi0_fast_yam7h_ea/yam7h_p50 |
| 09:19:27.165 [I] Running on: dbd491d61013 (2930:train.py:209) |
| INFO:2026-08-07 09:19:27,704:jax._src.xla_bridge:925: Unable to initialize backend 'rocm': module 'jaxlib.xla_extension' has no attribute 'GpuAllocatorConfig' |
| 09:19:27.704 [I] Unable to initialize backend 'rocm': module 'jaxlib.xla_extension' has no attribute 'GpuAllocatorConfig' (2930:xla_bridge.py:925) |
| INFO:2026-08-07 09:19:27,704:jax._src.xla_bridge:925: Unable to initialize backend 'tpu': INTERNAL: Failed to open libtpu.so: libtpu.so: cannot open shared object file: No such file or directory |
| 09:19:27.704 [I] Unable to initialize backend 'tpu': INTERNAL: Failed to open libtpu.so: libtpu.so: cannot open shared object file: No such file or directory (2930:xla_bridge.py:925) |
| 09:19:28.787 [I] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7821e7a0f7d0> (2930:base_pytree_checkpoint_handler.py:334) |
| 09:19:28.787 [I] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7821e7a0f7d0> (2930:base_pytree_checkpoint_handler.py:334) |
| 09:19:28.788 [I] [thread=MainThread] Failed to get flag value for EXPERIMENTAL_ORBAX_USE_DISTRIBUTED_PROCESS_ID. (2930:multihost.py:390) |
| 09:19:28.788 [I] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=None, item_handlers={'assets': <openpi.training.checkpoints.CallbackHandler object at 0x7821eb57a250>, 'train_state': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78209c1bc350>, 'params': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78209c0a1e90>}, handler_registry=None (2930:checkpoint_manager.py:620) |
| 09:19:28.788 [I] Deferred registration for item: "assets". Adding handler `<openpi.training.checkpoints.CallbackHandler object at 0x7821eb57a250>` for item "assets" and save args `<class 'openpi.training.checkpoints.CallbackSave'>` and restore args `<class 'openpi.training.checkpoints.CallbackRestore'>` to `_handler_registry`. (2930:composite_checkpoint_handler.py:234) |
| 09:19:28.788 [I] Deferred registration for item: "train_state". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78209c1bc350>` for item "train_state" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. (2930:composite_checkpoint_handler.py:234) |
| 09:19:28.788 [I] Deferred registration for item: "params". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78209c0a1e90>` for item "params" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. (2930:composite_checkpoint_handler.py:234) |
| 09:19:28.788 [I] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78209c45ac10>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. (2930:composite_checkpoint_handler.py:234) |
| 09:19:28.788 [I] Initialized registry DefaultCheckpointHandlerRegistry({('assets', <class 'openpi.training.checkpoints.CallbackSave'>): <openpi.training.checkpoints.CallbackHandler object at 0x7821eb57a250>, ('assets', <class 'openpi.training.checkpoints.CallbackRestore'>): <openpi.training.checkpoints.CallbackHandler object at 0x7821eb57a250>, ('train_state', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78209c1bc350>, ('train_state', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78209c1bc350>, ('params', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78209c0a1e90>, ('params', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78209c0a1e90>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78209c45ac10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78209c45ac10>}). (2930:composite_checkpoint_handler.py:502) |
| 09:19:28.788 [I] orbax-checkpoint version: 0.11.13 (2930:abstract_checkpointer.py:35) |
| 09:19:28.788 [I] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x782097ecfe20> timeout: 7200 secs and primary_host=0 for async checkpoint writes (2930:async_checkpointer.py:170) |
| 09:19:28.789 [I] Found 0 checkpoint steps in /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e (2930:checkpoint_manager.py:1701) |
| 09:19:28.789 [I] [process=0][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=4, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=False, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=AsyncOptions(timeout_secs=7200, barrier_sync_fn=None, post_finalization_callback=None, create_directories_asynchronously=True), multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=None, prevent_write_metrics=False), root_directory=/workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x78207c213a50> (2930:checkpoint_manager.py:801) |
| 09:19:30.738 [I] Loaded norm stats from /workspace/openpi/assets/pi0_fast_yam7h_ea/yam7h_p50 (2930:config.py:239) |
| 09:19:30.740 [I] data_config: DataConfig(repo_id='angkul07/abc-teleop', asset_id='yam7h_p50', norm_stats={'state': NormStats(mean=array([ 0.40010831, 1.2446171 , 1.12815332, -0.37802806, -0.16871828, |
| -0.14126156, 0.45964989, -0.41583106, 1.4478097 , 0.84182328, |
| 0.46756488, 0.26250845, -0.26482391, 0.39356148]), std=array([0.95272809, 0.41245013, 0.74113858, 1.02163243, 0.4025619 , |
| 0.42175013, 0.47215483, 1.29791224, 0.44621828, 0.46785772, |
| 0.97214621, 0.64248192, 0.60956204, 0.4571799 ]), q01=array([-1.00500087e+00, 5.90748287e-04, 1.53646450e-01, -1.57070005e+00, |
| -1.39928050e+00, -1.18746808e+00, 2.07537213e-03, -2.54240238e+00, |
| 2.69945458e-01, 2.59891499e-01, -1.11878431e+00, -7.01378860e-01, |
| -1.90642516e+00, 6.52602306e-03]), q99=array([2.64948381, 2.44844322, 2.40492031, 1.57007177, 1.15353093, |
| 1.16859837, 0.99863454, 1.4733806 , 2.2749618 , 1.96118775, |
| 1.57009526, 1.57023446, 0.7669128 , 0.99902859])), 'actions': NormStats(mean=array([ 0.02155279, -0.00579135, -0.00228477, -0.01728834, -0.0166839 , |
| -0.02572007, 0.45139426, 0.00291322, -0.00218901, -0.00544402, |
| -0.00132601, 0.00219626, 0.00799875, 0.38192591]), std=array([0.51440758, 0.36375082, 0.42378896, 0.55077744, 0.46478823, |
| 0.38022304, 0.47820532, 0.15757751, 0.29755983, 0.29284346, |
| 0.3350372 , 0.10243557, 0.17190453, 0.46640092]), q01=array([-1.66757157e+00, -1.30125664e+00, -1.11085481e+00, -2.89170725e+00, |
| -1.69584218e+00, -1.62805669e+00, -1.22540113e-04, -5.20113296e-01, |
| -9.44693478e-01, -1.21951298e+00, -1.24070471e+00, -3.56004610e-01, |
| -4.99472595e-01, 0.00000000e+00]), q99=array([1.58228415, 0.90835382, 1.4677495 , 1.25829733, 1.45866465, |
| 0.73837124, 0.99979982, 0.67696395, 1.06029136, 0.93962311, |
| 0.98209369, 0.30040325, 0.63307333, 0.9998 ]))}, repack_transforms=Group(inputs=[RepackTransform(structure={'observation/top_image': 'observation.images.top', 'observation/left_wrist_image': 'observation.images.left_wrist', 'observation/right_wrist_image': 'observation.images.right_wrist', 'observation/state': 'observation.state', 'actions': 'action', 'prompt': 'prompt'})], outputs=()), data_transforms=Group(inputs=(YamInputs(model_type=<ModelType.PI0_FAST: 'pi0_fast'>), DeltaActions(mask=(True, True, True, True, True, True, False, True, True, True, True, True, True, False))), outputs=(AbsoluteActions(mask=(True, True, True, True, True, True, False, True, True, True, True, True, True, False)), YamOutputs())), model_transforms=Group(inputs=[InjectDefaultPrompt(prompt=None), ResizeImages(height=224, width=224), TokenizeFASTInputs(tokenizer=<openpi.models.tokenizer.FASTTokenizer object at 0x78209c203750>)], outputs=[ExtractFASTActions(tokenizer=<openpi.models.tokenizer.FASTTokenizer object at 0x7821e99e4b90>, action_horizon=50, action_dim=14)]), train_only_transforms=Group(inputs=[ImageAugment(config=ImageAugmentConfig(brightness=0.15, contrast=0.15, saturation=0.1, hue=0.03, crop_scale=(0.95, 1.0), geometric_keys=('base_0_rgb',)))], outputs=()), use_quantile_norm=True, mixture=(MixtureSource(repo_id='angkul07/abc-teleop', samples_per_batch=32, root='/workspace/data/yam7h/teleop', exclude_episodes=(), holdout_fraction=0.0, holdout_seed=0), MixtureSource(repo_id='Kavin60606/fdstudio-run_e2e', samples_per_batch=32, root='/workspace/data/yam7h/ego', exclude_episodes=(), holdout_fraction=0.0, holdout_seed=0)), action_sequence_keys=('action',), prompt_from_task=True, rlds_data_dir=None, action_space=None, datasets=()) (2930:data_loader.py:521) |
| 09:19:31.025 [I] mixture source angkul07/abc-teleop: 2214 train frames, 32 samples/batch (2930:data_loader.py:308) |
| 09:19:31.025 [I] mixture source Kavin60606/fdstudio-run_e2e: 527 train frames, 32 samples/batch (2930:data_loader.py:308) |
| 09:19:31.025 [I] local_batch_size: 64 (2930:data_loader.py:612) |
| 09:19:31.025 [I] mixture sampler: angkul07/abc-teleop -> 32/64 per batch (50.0% gradient share), 2214 frames, 0.62x its storage share (2930:data_loader.py:664) |
| 09:19:31.025 [I] mixture sampler: Kavin60606/fdstudio-run_e2e -> 32/64 per batch (50.0% gradient share), 527 frames, 2.60x its storage share (2930:data_loader.py:664) |
| 09:20:10.730 [I] Initialized data loader: |
| [0].images['base_0_rgb']: (64, 224, 224, 3)@float32 |
| [0].images['left_wrist_0_rgb']: (64, 224, 224, 3)@float32 |
| [0].images['right_wrist_0_rgb']: (64, 224, 224, 3)@float32 |
| [0].image_masks['base_0_rgb']: (64,)@bool |
| [0].image_masks['left_wrist_0_rgb']: (64,)@bool |
| [0].image_masks['right_wrist_0_rgb']: (64,)@bool |
| [0].state: (64, 14)@float32 |
| [0].tokenized_prompt: (64, 300)@int32 |
| [0].tokenized_prompt_mask: (64, 300)@bool |
| [0].token_ar_mask: (64, 300)@int32 |
| [0].token_loss_mask: (64, 300)@bool |
| [1]: (64, 50, 14)@float32 (2930:train.py:248) |
| 09:20:11.506 [I] Downloading gs://openpi-assets/checkpoints/pi0_fast_base/params to /root/.cache/openpi/openpi-assets/checkpoints/pi0_fast_base/params (2930:download.py:86) |
| 09:20:11.507 [W] gsutil not found, falling back to gcsfs. This may fail if GCP credentials are not configured correctly. (2930:download.py:111) |
| 09:20:39.269 [I] Progress on: -/10850359508 rate:- remaining:? elapsed:00:00 postfix:- (2930:tqdm_logging.py:145) |
| 09:20:49.286 [I] Progress on: 1.01GiB/10.1GiB rate:111,380,227.2iB/s remaining:01:27 elapsed:00:10 postfix:- (2930:tqdm_logging.py:145) |
| 09:20:59.301 [I] Progress on: 2.10GiB/10.1GiB rate:115,749,697.7iB/s remaining:01:14 elapsed:00:20 postfix:- (2930:tqdm_logging.py:145) |
| 09:21:09.317 [I] Progress on: 3.18GiB/10.1GiB rate:116,024,092.6iB/s remaining:01:04 elapsed:00:30 postfix:- (2930:tqdm_logging.py:145) |
| 09:21:19.332 [I] Progress on: 4.25GiB/10.1GiB rate:115,362,921.1iB/s remaining:00:54 elapsed:00:40 postfix:- (2930:tqdm_logging.py:145) |
| 09:21:29.422 [I] Progress on: 5.35GiB/10.1GiB rate:116,480,649.7iB/s remaining:00:43 elapsed:00:50 postfix:- (2930:tqdm_logging.py:145) |
| 09:21:39.436 [I] Progress on: 6.41GiB/10.1GiB rate:115,178,786.0iB/s remaining:00:34 elapsed:01:00 postfix:- (2930:tqdm_logging.py:145) |
| 09:21:49.454 [I] Progress on: 7.46GiB/10.1GiB rate:113,047,453.1iB/s remaining:00:25 elapsed:01:10 postfix:- (2930:tqdm_logging.py:145) |
| 09:21:59.471 [I] Progress on: 8.50GiB/10.1GiB rate:110,136,451.6iB/s remaining:00:15 elapsed:01:20 postfix:- (2930:tqdm_logging.py:145) |
| 09:22:09.486 [I] Progress on: 9.50GiB/10.1GiB rate:107,903,191.8iB/s remaining:00:06 elapsed:01:30 postfix:- (2930:tqdm_logging.py:145) |
| 09:22:15.493 [I] Progress on: 10.1GiB/10.1GiB rate:107,677,680.0iB/s remaining:00:00 elapsed:01:36 postfix:- (2930:tqdm_logging.py:145) |
| 09:22:15.494 [I] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7821e7a0f7d0> (2930:base_pytree_checkpoint_handler.py:334) |
| 09:22:15.508 [I] Restoring checkpoint from /root/.cache/openpi/openpi-assets/checkpoints/pi0_fast_base/params. (2930:checkpointer.py:298) |
| 09:22:20.955 [I] [process=0] /jax/checkpoint/read/bytes_per_sec: 2.0 GiB/s (total bytes: 10.9 GiB) (time elapsed: 5 seconds) (per-host) (2930:base_pytree_checkpoint_handler.py:114) |
| 09:22:20.955 [I] Finished restoring checkpoint in 5.45 seconds from /root/.cache/openpi/openpi-assets/checkpoints/pi0_fast_base/params. (2930:checkpointer.py:309) |
| 09:22:28.553 [I] Initialized train state: |
| ['PaliGemma']['img']['Transformer']['encoder_norm']['bias'].value: (1152,)@float32 |
| ['PaliGemma']['img']['Transformer']['encoder_norm']['scale'].value: (1152,)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['LayerNorm_0']['bias'].value: (27, 1152)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['LayerNorm_0']['scale'].value: (27, 1152)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['LayerNorm_1']['bias'].value: (27, 1152)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['LayerNorm_1']['scale'].value: (27, 1152)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MlpBlock_0']['Dense_0']['bias'].value: (27, 4304)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MlpBlock_0']['Dense_0']['kernel'].value: (27, 1152, 4304)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MlpBlock_0']['Dense_1']['bias'].value: (27, 1152)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MlpBlock_0']['Dense_1']['kernel'].value: (27, 4304, 1152)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MultiHeadDotProductAttention_0']['key']['bias'].value: (27, 16, 72)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MultiHeadDotProductAttention_0']['key']['kernel'].value: (27, 1152, 16, 72)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MultiHeadDotProductAttention_0']['out']['bias'].value: (27, 1152)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MultiHeadDotProductAttention_0']['out']['kernel'].value: (27, 16, 72, 1152)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MultiHeadDotProductAttention_0']['query']['bias'].value: (27, 16, 72)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MultiHeadDotProductAttention_0']['query']['kernel'].value: (27, 1152, 16, 72)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MultiHeadDotProductAttention_0']['value']['bias'].value: (27, 16, 72)@float32 |
| ['PaliGemma']['img']['Transformer']['encoderblock']['MultiHeadDotProductAttention_0']['value']['kernel'].value: (27, 1152, 16, 72)@float32 |
| ['PaliGemma']['img']['embedding']['bias'].value: (1152,)@float32 |
| ['PaliGemma']['img']['embedding']['kernel'].value: (14, 14, 3, 1152)@float32 |
| ['PaliGemma']['img']['head']['bias'].value: (2048,)@float32 |
| ['PaliGemma']['img']['head']['kernel'].value: (1152, 2048)@float32 |
| ['PaliGemma']['img']['pos_embedding'].value: (1, 256, 1152)@float32 |
| ['PaliGemma']['llm']['embedder']['input_embedding'].value: (257152, 2048)@bfloat16 |
| ['PaliGemma']['llm']['final_norm']['scale'].value: (2048,)@bfloat16 |
| ['PaliGemma']['llm']['layers']['attn']['attn_vec_einsum']['lora_a'].value: (18, 8, 256, 16)@float32 |
| ['PaliGemma']['llm']['layers']['attn']['attn_vec_einsum']['lora_b'].value: (18, 8, 16, 2048)@float32 |
| ['PaliGemma']['llm']['layers']['attn']['attn_vec_einsum']['w'].value: (18, 8, 256, 2048)@bfloat16 |
| ['PaliGemma']['llm']['layers']['attn']['kv_einsum']['lora_a'].value: (18, 2, 1, 2048, 16)@float32 |
| ['PaliGemma']['llm']['layers']['attn']['kv_einsum']['lora_b'].value: (18, 2, 1, 16, 256)@float32 |
| ['PaliGemma']['llm']['layers']['attn']['kv_einsum']['w'].value: (18, 2, 1, 2048, 256)@bfloat16 |
| ['PaliGemma']['llm']['layers']['attn']['q_einsum']['lora_a'].value: (18, 8, 2048, 16)@float32 |
| ['PaliGemma']['llm']['layers']['attn']['q_einsum']['lora_b'].value: (18, 8, 16, 256)@float32 |
| ['PaliGemma']['llm']['layers']['attn']['q_einsum']['w'].value: (18, 8, 2048, 256)@bfloat16 |
| ['PaliGemma']['llm']['layers']['mlp']['gating_einsum'].value: (18, 2, 2048, 16384)@bfloat16 |
| ['PaliGemma']['llm']['layers']['mlp']['gating_einsum_lora_a'].value: (18, 2, 2048, 16)@float32 |
| ['PaliGemma']['llm']['layers']['mlp']['gating_einsum_lora_b'].value: (18, 2, 16, 16384)@float32 |
| ['PaliGemma']['llm']['layers']['mlp']['linear'].value: (18, 16384, 2048)@bfloat16 |
| ['PaliGemma']['llm']['layers']['mlp']['linear_lora_a'].value: (18, 16384, 16)@float32 |
| ['PaliGemma']['llm']['layers']['mlp']['linear_lora_b'].value: (18, 16, 2048)@float32 |
| ['PaliGemma']['llm']['layers']['pre_attention_norm']['scale'].value: (18, 2048)@bfloat16 |
| ['PaliGemma']['llm']['layers']['pre_ffw_norm']['scale'].value: (18, 2048)@bfloat16 (2930:train.py:259) |
| 09:22:28.557 [I] Progress on: -/30 rate:- remaining:? elapsed:00:00 postfix:- (2930:tqdm_logging.py:145) |
| 09:22:28.558 [I] Per-step metrics log: /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/train_metrics.log (2930:train.py:296) |
|
09:25:19.900 [I] Progress on: -/30 rate:- remaining:? elapsed:02:51 postfix:- (2930:tqdm_logging.py:145) |
| 09:25:35.386 [I] Progress on: 4.00it/30.0it rate:28.2s/it remaining:12:14 elapsed:03:06 postfix:- (2930:tqdm_logging.py:145) |
| 09:25:49.749 [I] Progress on: 6.00it/30.0it rate:16.1s/it remaining:06:25 elapsed:03:21 postfix:- (2930:tqdm_logging.py:145) |
| 09:26:04.142 [I] Progress on: 8.00it/30.0it rate:11.3s/it remaining:04:07 elapsed:03:35 postfix:- (2930:tqdm_logging.py:145) |
| 09:26:18.500 [I] Progress on: 10.0it/30.0it rate:9.1s/it remaining:03:02 elapsed:03:49 postfix:- (2930:tqdm_logging.py:145) |
| 09:26:32.864 [I] Progress on: 12.0it/30.0it rate:8.1s/it remaining:02:26 elapsed:04:04 postfix:- (2930:tqdm_logging.py:145) |
| 09:26:47.191 [I] Progress on: 14.0it/30.0it rate:7.6s/it remaining:02:02 elapsed:04:18 postfix:- (2930:tqdm_logging.py:145) |
| 09:27:01.551 [I] Progress on: 16.0it/30.0it rate:7.4s/it remaining:01:43 elapsed:04:32 postfix:- (2930:tqdm_logging.py:145) |
| 09:27:15.909 [I] Progress on: 18.0it/30.0it rate:7.3s/it remaining:01:27 elapsed:04:47 postfix:- (2930:tqdm_logging.py:145) |
| 09:27:30.270 [I] Progress on: 20.0it/30.0it rate:7.2s/it remaining:01:12 elapsed:05:01 postfix:- (2930:tqdm_logging.py:145) |
| 09:27:44.639 [I] Progress on: 22.0it/30.0it rate:7.2s/it remaining:00:57 elapsed:05:16 postfix:- (2930:tqdm_logging.py:145) |
| 09:27:58.998 [I] Progress on: 24.0it/30.0it rate:7.2s/it remaining:00:43 elapsed:05:30 postfix:- (2930:tqdm_logging.py:145) |
| 09:28:13.350 [I] Progress on: 26.0it/30.0it rate:7.2s/it remaining:00:28 elapsed:05:44 postfix:- (2930:tqdm_logging.py:145) |
| 09:28:27.712 [I] Progress on: 28.0it/30.0it rate:7.2s/it remaining:00:14 elapsed:05:59 postfix:- (2930:tqdm_logging.py:145) |
| 09:28:42.088 [I] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. (2930:checkpoint_manager.py:1987) |
| 09:28:42.089 [I] [process=0] Saving checkpoint at step 29 (2930:checkpoint_manager.py:1408) |
| 09:28:42.089 [I] [process=0] Started async saving checkpoint to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29. (2930:async_checkpointer.py:439) |
| 09:28:42.089 [I] Using ThreadSafeKeyValueSignalingClient (2930:signaling_client.py:332) |
| 09:28:42.094 [I] Creating tmp directory /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0 (2930:atomicity.py:144) |
| 09:28:42.095 [I] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1786094922094935432, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1786094922094935432, "commit_timestamp_nsecs": null, "custom_metadata": {}} to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/_CHECKPOINT_METADATA (2930:checkpoint.py:186) |
| 09:28:42.097 [I] Creating tmp directory /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/assets.orbax-checkpoint-tmp-1 (2930:atomicity.py:144) |
| 09:28:42.097 [I] Creating tmp directory /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/params.orbax-checkpoint-tmp-2 (2930:atomicity.py:144) |
| 09:28:42.097 [I] Creating tmp directory /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/train_state.orbax-checkpoint-tmp-3 (2930:atomicity.py:144) |
| 09:28:42.105 [I] Transferring arrays to host memory with options: use_replica_parallel=True, enable_pinned_host_transfer=False (2930:replica_slices.py:341) |
| 09:28:50.891 [I] Transferring arrays to host memory with options: use_replica_parallel=True, enable_pinned_host_transfer=False (2930:replica_slices.py:341) |
| 09:28:50.899 [I] Array name: 'params.PaliGemma.llm.layers.mlp.gating_einsum.value', global shape: (18, 2, 2048, 16384), write shape: (18, 2, 2048, 16384), chosen chunk shape: (18, 2, 2048, 8192) (2930:tensorstore_utils.py:408) |
| 09:28:50.911 [I] [process=0][thread=array_type_handler] Wrote 42 array_metadata.ArrayMetadata to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/params.orbax-checkpoint-tmp-2/array_metadatas/process_0 (2930:array_metadata_store.py:198) |
| 09:28:52.484 [I] [process=0] /jax/checkpoint/write/blocking_bytes_per_sec: 622.8 MiB/s (total bytes: 6.3 GiB) (time elapsed: 10 seconds) (per-host) (2930:base_pytree_checkpoint_handler.py:114) |
| 09:28:52.485 [I] [process=0] /jax/checkpoint/write/blocking_bytes_per_sec: 325.0 MiB/s (total bytes: 3.3 GiB) (time elapsed: 10 seconds) (per-host) (2930:base_pytree_checkpoint_handler.py:114) |
| 09:28:52.495 [I] [process=0][thread=async_save] Background save thread started. (2930:async_checkpointer.py:77) |
| 09:28:52.496 [I] Finished blocking save in 10.41 seconds. Continuing to save asynchronously to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29. (2930:async_checkpointer.py:548) |
| 09:28:52.497 [I] [process=0][thread=MainThread][step=29] Starting CheckpointManager Save Finalize thread=save_finalize (2930:checkpoint_manager.py:1453) |
| 09:28:52.497 [I] [process=0][thread=save_finalize] Waiting for background save thread=async_save. (2930:async_checkpointer.py:258) |
| 09:28:52.497 [I] {'step': 29, 'event_type': 'save', 'directory': '/workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1786094922.0887296, 'wait_for_prev_duration_secs': 8.916854858398438e-05, 'checkpointer_blocking_start_time': 1786094922.0890322, 'checkpointer_blocking_duration_secs': 10.408042669296265, 'get_old_steps_start_time': 1786094932.4970958, 'get_old_steps_duration_secs': 3.814697265625e-06, 'checkpoint_manager_blocking_start_time': 1786094922.0886383, 'checkpoint_manager_blocking_duration_secs': 10.408944606781006} (2930:standard_logger.py:34) |
| 09:28:52.498 [I] Progress on: 30.0it/30.0it rate:10.3s/it remaining:00:00 elapsed:06:23 postfix:- (2930:tqdm_logging.py:145) |
| 09:28:52.498 [I] Progress on: 30.0it/30.0it rate:10.3s/it remaining:00:00 elapsed:06:23 postfix:- (2930:tqdm_logging.py:145) |
| 09:28:52.505 [I] [process=0][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/train_state.orbax-checkpoint-tmp-3/array_metadatas/process_0 (2930:array_metadata_store.py:198) |
| 09:28:52.508 [I] Waiting for checkpoint manager to finish (2930:train.py:324) |
| 09:28:52.509 [I] [process=0][thread=MainThread][step=29][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. (2930:checkpoint_manager.py:1998) |
| 09:28:58.520 [I] [process=0] /jax/checkpoint/write/bytes_per_sec: 394.0 MiB/s (total bytes: 6.3 GiB) (time elapsed: 16 seconds) (per-host) (2930:base_pytree_checkpoint_handler.py:114) |
| 09:28:58.525 [I] [process=0] /jax/checkpoint/write/bytes_per_sec: 205.5 MiB/s (total bytes: 3.3 GiB) (time elapsed: 16 seconds) (per-host) (2930:base_pytree_checkpoint_handler.py:114) |
| 09:28:58.525 [I] [process=0][thread=async_save] 5 Handler Commit operations completed. (2930:async_checkpointer.py:87) |
| 09:28:58.526 [I] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1786094922094935432, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/_CHECKPOINT_METADATA (2930:checkpoint.py:226) |
| 09:28:58.526 [I] Renaming /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/assets.orbax-checkpoint-tmp-1 to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/assets (2930:atomicity.py:289) |
| 09:28:58.527 [I] Updated Metadata={'item_handlers': {'assets': 'openpi.training.checkpoints.CallbackHandler', 'params': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler', 'train_state': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1786094922094935432, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/_CHECKPOINT_METADATA (2930:checkpoint.py:245) |
| 09:28:58.528 [I] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. (2930:array_metadata_store.py:362) |
| 09:28:58.534 [I] Renaming /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/train_state.orbax-checkpoint-tmp-3 to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/train_state (2930:atomicity.py:289) |
| 09:28:58.536 [I] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. (2930:array_metadata_store.py:362) |
| 09:28:58.541 [I] Renaming /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/params.orbax-checkpoint-tmp-2 to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0/params (2930:atomicity.py:289) |
| 09:28:58.542 [I] Renaming /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29.orbax-checkpoint-tmp-0 to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29 (2930:atomicity.py:289) |
| 09:28:58.542 [I] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `/workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29`. (2930:atomicity.py:573) |
| 09:28:58.542 [I] Finished asynchronous save (blocking + background) in 16.45 seconds to /workspace/openpi/checkpoints/pi0_fast_yam7h_ea/run_e2e/29 (2930:async_checkpointer.py:412) |
| 09:28:58.542 [I] [process=0][thread=async_save] Background save thread done. (2930:async_checkpointer.py:143) |
| 09:28:58.542 [I] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. (2930:async_checkpointer.py:266) |
| 09:28:58.542 [I] [process=0][thread=save_finalize] No errors found in background save thread=async_save. (2930:async_checkpointer.py:276) |
| 09:28:58.542 [I] [process=0][thread=save_finalize][step=29] CheckpointManager Save Finalize is syncing with other hosts... (2930:checkpoint_manager.py:2095) |
| 09:28:58.542 [I] [process=0][thread=save_finalize][step=29] CheckpointManager Save Finalize is done on all hosts. (2930:checkpoint_manager.py:2110) |
| 09:28:58.543 [I] [process=0][thread=MainThread][step=29][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=29. (2930:checkpoint_manager.py:2010) |
| Step 0: grad_norm=12.8822, loss=11.3075, param_norm=1666.0126 |
|
|
