main (51c7f2b8e) vs feat/nnx-trainstate-and-training-loop (7c12861a4)| Metric | main 51c7f2b8e | feat/nnx-trainstate-and-training-loop 7c12861a4 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.7100 | 8.7100 | 0 |
| TFLOP/s | 115.399 | 113.847 | -1.552 |
| Tok/s | 17394.3 | 17160.3 | -233.926 |
| Avg s/step | 2.468 | 2.447 | -0.021 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-15 21:38:26.261815: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0415 21:38:26.381171 134120840285312 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 21:39:14.151234 134120840285312 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 21:39:14.151346 134120840285312 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 21:39:14.151381 134120840285312 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0415 21:39:14.151409 134120840285312 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 21:39:14.151496 134120840285312 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 21:39:14.244615 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:39:14.643142 134120840285312 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0415 21:39:14.643448 134120840285312 checkpointing.py:677] Setting up checkpoint logger... I0415 21:39:14.643491 134120840285312 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 21:39:14.643526 134120840285312 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 21:39:14.644192 134120840285312 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x79fada5d34d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 21:39:16.986779 134120840285312 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 21:39:16.986934 134120840285312 checkpoint_manager.py:702] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79f45cb638c0>}, handler_registry=None I0415 21:39:16.987187 134120840285312 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79f45cb638c0>` for item "items" 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`. I0415 21:39:16.987229 134120840285312 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79f45c339f40>` 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`. I0415 21:39:16.987259 134120840285312 composite_checkpoint_handler.py:505] Initialized registry DefaultCheckpointHandlerRegistry({('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79f45cb638c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79f45cb638c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79f45c339f40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79f45c339f40>}). I0415 21:39:16.987544 134120840285312 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 21:39:16.987600 134120840285312 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x79f45c32d3a0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0415 21:39:17.108356 134120840285312 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints I0415 21:39:17.108570 134120840285312 checkpoint_manager.py:921] [process=0][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, 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=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_hns=False, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, 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=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False), root_directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x79f45c339eb0> I0415 21:39:17.108652 134120840285312 checkpointing.py:301] Checkpoint manager created! I0415 21:39:17.188105 134120840285312 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0415 21:39:17.191080 134120840285312 reader.py:262] Creating a tf.data.Dataset reading 8 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0. I0415 21:39:17.243328 134120840285312 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split train, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0415 21:39:17.270649 134120840285312 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:39:17.270719 134120840285312 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:39:18.212322 134120840285312 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 21:39:18.212421 134120840285312 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:39:18.314467 134120840285312 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:39:18.314543 134120840285312 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:39:18.329589 134120840285312 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:39:18.329635 134120840285312 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:39:18.351830 134120840285312 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:39:18.351891 134120840285312 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:39:18.367002 134120840285312 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:39:18.367068 134120840285312 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:39:18.382063 134120840285312 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:39:18.382113 134120840285312 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:39:18.404398 134120840285312 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 21:39:18.404453 134120840285312 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:39:18.423489 134120840285312 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 21:39:18.423539 134120840285312 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0415 21:39:18.610121 134120840285312 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 21:39:18.610214 134120840285312 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0415 21:39:18.660779 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0415 21:39:19.052972 134120840285312 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0415 21:39:19.053114 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0415 21:39:19.053161 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0415 21:39:19.053208 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0415 21:39:19.053249 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0415 21:39:19.053277 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0415 21:39:19.053317 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0415 21:39:19.053356 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0415 21:39:19.053385 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0415 21:39:19.053410 134120840285312 maxtext_utils.py:1620] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0415 21:39:19.053446 134120840285312 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0415 21:39:19.053481 134120840285312 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0415 21:39:19.505837 134120840285312 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:39:19.505989 134120840285312 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0415 21:39:19.520745 134120840285312 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:39:19.520797 134120840285312 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0415 21:39:19.794206 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:39:19.909880 134120840285312 max_utils.py:791] Total memory size: 3.6 GB, Output size: 1.5 GB, Temp size: 2.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0415 21:39:19.910398 134120840285312 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 21:39:19.910565 134120840285312 metric_logger.py:289] number parameters: 1.104 billion W0415 21:39:22.906751 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:39:23.038348 134120840285312 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0415 21:39:23.157885 134120840285312 checkpointing.py:776] Waited 0.11952090263366699 seconds for step 0 to finish before starting checkpointing. I0415 21:39:23.158352 134120840285312 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:39:23.158507 134120840285312 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0415 21:39:23.158941 134120840285312 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0. I0415 21:39:23.242604 134120840285312 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0415 21:39:23.324005 134007673062976 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0 I0415 21:39:23.337351 134120840285312 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:39:23.337437 134120840285312 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False W0415 21:39:23.556571 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:23.569856 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:23.587553 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:23.599673 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:23.615263 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:23.621389 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:23.625784 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:23.629874 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:39:24.032878 134007662577216 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0/items I0415 21:39:24.185361 134007622731328 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289163946408154, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776289163946408154, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0415 21:39:24.493246 3257722 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 21:39:24.493293 3257722 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 21:39:24.874251 134120840285312 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.537310s I0415 21:39:24.878706 134120840285312 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 7.548 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0415 21:39:24.878770 134120840285312 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.634956s (batch_requests_ready=0.089806s, total_serialization_initiated=1.540791s, others=0.004359s) I0415 21:39:24.878844 134120840285312 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.635770s (all_items=0.000021s, per_item={'items': '0.00002098'}, temp_paths=1.635749) I0415 21:39:24.880134 134007683548736 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:39:24.880228 134120840285312 async_checkpointer.py:561] Finished blocking save. Time taken: 1.721670s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0. I0415 21:39:24.880465 134120840285312 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:39:24.880643 134007559816768 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:39:24.880744 134120840285312 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776289163.1583385, 'wait_for_prev_duration_secs': 4.363059997558594e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776289163.1585274, 'checkpointer_blocking_duration_secs': 1.7218413352966309, 'get_old_steps_start_time': 1776289164.8803945, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1776289163.1582422, 'checkpoint_manager_blocking_duration_secs': 1.7224705219268799} I0415 21:39:24.880880 134120840285312 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0415 21:39:24.880927 134120840285312 max_utils.py:750] Memstats: After params initialized: I0415 21:39:24.880980 134120840285312 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0415 21:39:24.881007 134120840285312 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0415 21:39:24.881027 134120840285312 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0415 21:39:24.881061 134120840285312 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0415 21:39:24.881080 134120840285312 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0415 21:39:24.881101 134120840285312 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0415 21:39:24.881119 134120840285312 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0415 21:39:24.881137 134120840285312 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0415 21:39:24.886868 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:24.892962 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:39:24.897400 3255150 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:39:25.211555 134120840285312 metric_logger.py:185] completed step: 0, seconds: 3.127, TFLOP/s/device: 4.345, Tokens/s/device: 654.887, total_weights: 13328, loss: 10.828 I0415 21:39:25.212570 134120840285312 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/tensorboard/' I0415 21:39:25.327763 134120840285312 metric_logger.py:185] completed step: 1, seconds: 2.171, TFLOP/s/device: 6.258, Tokens/s/device: 943.261, total_weights: 12332, loss: 10.828 I0415 21:39:25.445274 134120840285312 metric_logger.py:185] completed step: 2, seconds: 0.009, TFLOP/s/device: 1435.210, Tokens/s/device: 216330.411, total_weights: 15161, loss: 9.823 I0415 21:39:25.562605 134120840285312 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 115.780, Tokens/s/device: 17451.620, total_weights: 13327, loss: 9.405 I0415 21:39:25.641383 134007641605696 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0/items/array_metadatas/process_0 I0415 21:39:45.003162 134120840285312 metric_logger.py:185] completed step: 4, seconds: 0.118, TFLOP/s/device: 115.441, Tokens/s/device: 17400.466, total_weights: 11939, loss: 9.045 I0415 21:39:45.012256 134120840285312 metric_logger.py:185] completed step: 5, seconds: 0.117, TFLOP/s/device: 115.901, Tokens/s/device: 17469.931, total_weights: 15502, loss: 8.953 I0415 21:39:45.129323 134120840285312 metric_logger.py:185] completed step: 6, seconds: 19.440, TFLOP/s/device: 0.699, Tokens/s/device: 105.350, total_weights: 13864, loss: 8.775 I0415 21:39:45.247099 134120840285312 metric_logger.py:185] completed step: 7, seconds: 0.006, TFLOP/s/device: 2369.573, Tokens/s/device: 357167.771, total_weights: 12988, loss: 8.688 I0415 21:39:45.364578 134120840285312 metric_logger.py:185] completed step: 8, seconds: 0.118, TFLOP/s/device: 115.038, Tokens/s/device: 17339.768, total_weights: 13820, loss: 8.740 I0415 21:39:45.481568 134120840285312 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0415 21:39:45.482394 134120840285312 checkpointing.py:776] Waited 0.0008561611175537109 seconds for step 9 to finish before starting checkpointing. I0415 21:39:45.482673 134120840285312 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:39:51.818328 134007580788288 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 26.938910s (commit=26.504218s, array_metadata_write=0.434692s) I0415 21:39:51.819319 134007683548736 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 442.225 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 28 seconds) (per-host) I0415 21:39:51.819375 134007683548736 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 26.939090s. I0415 21:39:52.048122 134007683548736 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289163946408154, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0415 21:39:52.232564 134007683548736 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 21:39:52.428255 134007622731328 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289163946408154, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0415 21:39:52.642805 134007683548736 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:39:52.642996 134007683548736 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.552472s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0/items I0415 21:39:52.643675 134007683548736 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0/items I0415 21:39:52.882818 134007683548736 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0 I0415 21:39:53.544860 134007683548736 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0`. I0415 21:39:53.545499 134007683548736 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 30.386961s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/0 I0415 21:39:53.545576 134007683548736 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 28.665293s. I0415 21:39:53.545735 134007559816768 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:39:53.545839 134007559816768 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:39:53.545903 134007559816768 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:39:53.545947 134007559816768 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0415 21:39:53.546162 134120840285312 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0415 21:39:53.546294 134120840285312 checkpoint_manager.py:1441] Waiting for previous save to complete took 8.063631 seconds. If this number is high, consider checkpointing less frequently. I0415 21:39:53.547176 134120840285312 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0415 21:39:53.547483 134120840285312 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9. I0415 21:39:53.715453 134120840285312 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:39:53.715543 134120840285312 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0415 21:39:53.778517 134007559816768 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9 I0415 21:39:54.452150 134007549331008 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9/items I0415 21:40:02.907679 134120840285312 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.192647s I0415 21:40:02.912158 134120840285312 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.331 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0415 21:40:02.912218 134120840285312 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.275005s (batch_requests_ready=0.074952s, total_serialization_initiated=9.195685s, others=0.004368s) I0415 21:40:02.912292 134120840285312 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.275521s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=9.275503) I0415 21:40:02.913866 134007580788288 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:40:02.913965 134120840285312 async_checkpointer.py:561] Finished blocking save. Time taken: 9.366744s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9. I0415 21:40:02.914169 134120840285312 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:40:02.914396 134007790503488 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:40:02.914511 134120840285312 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776289185.482634, 'wait_for_prev_duration_secs': 8.063631296157837, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776289193.547201, 'checkpointer_blocking_duration_secs': 9.366881847381592, 'get_old_steps_start_time': 1776289202.9140995, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1776289185.4825804, 'checkpoint_manager_blocking_duration_secs': 17.431905031204224} I0415 21:40:02.914640 134120840285312 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0415 21:40:02.914674 134120840285312 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:40:03.672115 134019083667008 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9/items/array_metadatas/process_0 I0415 21:40:40.925164 134007549331008 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 38.012245s (commit=37.540962s, array_metadata_write=0.471283s) I0415 21:40:40.926241 134007580788288 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 267.225 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 47 seconds) (per-host) I0415 21:40:40.926295 134007580788288 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.012288s. I0415 21:40:41.355133 134007580788288 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 21:40:41.799171 134007580788288 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:40:41.799840 134007580788288 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.591193s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9/items I0415 21:40:41.800489 134007580788288 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9/items I0415 21:40:42.047018 134007580788288 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9 I0415 21:40:42.717937 134007580788288 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9`. I0415 21:40:42.718577 134007580788288 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 49.171362s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_true/checkpoints/9 I0415 21:40:42.718653 134007580788288 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 39.804646s. I0415 21:40:42.718822 134007790503488 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:40:42.718922 134007790503488 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:40:42.718984 134007790503488 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:40:42.719027 134007790503488 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0415 21:40:42.719424 134120840285312 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0415 21:40:42.719564 134120840285312 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:40:42.720447 134120840285312 metric_logger.py:185] completed step: 9, seconds: 0.118, TFLOP/s/device: 115.399, Tokens/s/device: 17394.259, total_weights: 12300, loss: 8.710 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
7c12861a4
· feat_nnx_trainstate_and_training_loop_20260416_033532
· full log
2026-04-16 04:05:21.021889: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0416 04:05:21.139290 131296930094208 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 04:06:15.379107 131296930094208 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 04:06:15.379212 131296930094208 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 04:06:15.379246 131296930094208 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0416 04:06:15.379271 131296930094208 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 04:06:15.379356 131296930094208 train.py:818] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 04:06:15.471616 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 04:06:15.876607 131296930094208 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 04:06:15.876776 131296930094208 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 04:06:15.876948 131296930094208 checkpointing.py:688] Setting up checkpoint logger... I0416 04:06:15.876987 131296930094208 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 04:06:15.877029 131296930094208 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 04:06:15.877488 131296930094208 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x77695bff4d10>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 04:06:18.249780 131296930094208 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 04:06:18.249958 131296930094208 checkpoint_manager.py:702] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7762dd538f50>}, handler_registry=None I0416 04:06:18.250226 131296930094208 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7762dd538f50>` for item "items" 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`. I0416 04:06:18.250268 131296930094208 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7762dd53b020>` 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`. I0416 04:06:18.250296 131296930094208 composite_checkpoint_handler.py:505] Initialized registry DefaultCheckpointHandlerRegistry({('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7762dd538f50>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7762dd538f50>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7762dd53b020>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7762dd53b020>}). I0416 04:06:18.250593 131296930094208 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 04:06:18.250641 131296930094208 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7762dd5460c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 04:06:18.374873 131296930094208 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints I0416 04:06:18.375121 131296930094208 checkpoint_manager.py:921] [process=0][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, 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=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_hns=False, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, 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=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False), root_directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7764a9d98b30> I0416 04:06:18.375201 131296930094208 checkpointing.py:302] Checkpoint manager created! I0416 04:06:18.453564 131296930094208 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 04:06:18.456483 131296930094208 reader.py:262] Creating a tf.data.Dataset reading 8 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0. I0416 04:06:18.509205 131296930094208 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split train, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 04:06:18.539556 131296930094208 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 04:06:18.539625 131296930094208 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 04:06:19.551017 131296930094208 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 04:06:19.551306 131296930094208 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 04:06:19.653517 131296930094208 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 04:06:19.653596 131296930094208 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 04:06:19.668487 131296930094208 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 04:06:19.668534 131296930094208 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 04:06:19.690763 131296930094208 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 04:06:19.690820 131296930094208 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 04:06:19.705863 131296930094208 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 04:06:19.705914 131296930094208 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 04:06:19.720702 131296930094208 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 04:06:19.720752 131296930094208 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 04:06:19.742756 131296930094208 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 04:06:19.742814 131296930094208 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 04:06:19.761474 131296930094208 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 04:06:19.761527 131296930094208 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 04:06:19.962603 131296930094208 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 04:06:19.962704 131296930094208 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0416 04:06:20.014905 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0416 04:06:20.404004 131296930094208 maxtext_utils.py:1790] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 04:06:20.404134 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 04:06:20.404177 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 04:06:20.404221 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 04:06:20.404260 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 04:06:20.404286 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 04:06:20.404325 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0416 04:06:20.404362 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0416 04:06:20.404390 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0416 04:06:20.404416 131296930094208 maxtext_utils.py:1790] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0416 04:06:20.404451 131296930094208 maxtext_utils.py:1790] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 04:06:20.404484 131296930094208 maxtext_utils.py:1790] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 04:06:20.839699 131296930094208 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 04:06:20.839784 131296930094208 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 04:06:20.854381 131296930094208 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 04:06:20.854434 131296930094208 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 04:06:21.149616 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 04:06:21.267143 131296930094208 max_utils.py:791] Total memory size: 3.6 GB, Output size: 1.5 GB, Temp size: 2.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0416 04:06:21.267611 131296930094208 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 04:06:21.267752 131296930094208 metric_logger.py:289] number parameters: 1.104 billion W0416 04:06:24.322215 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 04:06:24.456904 131296930094208 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 04:06:24.577729 131296930094208 checkpointing.py:798] Waited 0.12081575393676758 seconds for step 0 to finish before starting checkpointing. I0416 04:06:24.578161 131296930094208 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 04:06:24.578312 131296930094208 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 04:06:24.578734 131296930094208 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0. I0416 04:06:24.662196 131296930094208 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 04:06:24.746801 131183711094336 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0 I0416 04:06:24.754655 131296930094208 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 04:06:24.754751 131296930094208 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False W0416 04:06:24.973307 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:24.983936 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:24.994994 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:25.004557 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:25.018360 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:25.025809 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:25.031192 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:25.036015 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 04:06:25.492332 131183700608576 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0/items I0416 04:06:25.633687 131183660762688 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776312385393745863, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776312385393745863, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0416 04:06:25.967787 3659836 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 04:06:25.967835 3659836 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 04:06:26.401703 131296930094208 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.647461s I0416 04:06:26.405987 131296930094208 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 7.081 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 04:06:26.406065 131296930094208 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.742936s (batch_requests_ready=0.087741s, total_serialization_initiated=1.651002s, others=0.004192s) I0416 04:06:26.406138 131296930094208 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.743493s (all_items=0.000031s, per_item={'items': '0.00003076'}, temp_paths=1.743462) I0416 04:06:26.407353 131183597848128 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 04:06:26.407510 131296930094208 async_checkpointer.py:561] Finished blocking save. Time taken: 1.829159s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0. I0416 04:06:26.407705 131296930094208 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 04:06:26.407836 131183721580096 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 04:06:26.407907 131296930094208 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776312384.5781481, 'wait_for_prev_duration_secs': 4.172325134277344e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776312384.578333, 'checkpointer_blocking_duration_secs': 1.8292992115020752, 'get_old_steps_start_time': 1776312386.4076502, 'get_old_steps_duration_secs': 2.0503997802734375e-05, 'checkpoint_manager_blocking_start_time': 1776312384.5780642, 'checkpoint_manager_blocking_duration_secs': 1.829817771911621} I0416 04:06:26.408011 131296930094208 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 04:06:26.408062 131296930094208 max_utils.py:750] Memstats: After params initialized: I0416 04:06:26.408107 131296930094208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0416 04:06:26.408131 131296930094208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 04:06:26.408150 131296930094208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 04:06:26.408168 131296930094208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 04:06:26.408187 131296930094208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 04:06:26.408203 131296930094208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 04:06:26.408223 131296930094208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 04:06:26.408241 131296930094208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 04:06:26.413742 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:26.421415 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 04:06:26.426272 3657134 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 04:06:26.759077 131296930094208 metric_logger.py:185] completed step: 0, seconds: 3.189, TFLOP/s/device: 4.261, Tokens/s/device: 642.278, total_weights: 13328, loss: 10.828 I0416 04:06:26.760128 131296930094208 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/tensorboard/' I0416 04:06:26.874525 131296930094208 metric_logger.py:185] completed step: 1, seconds: 2.300, TFLOP/s/device: 5.908, Tokens/s/device: 890.485, total_weights: 12332, loss: 10.828 I0416 04:06:26.992125 131296930094208 metric_logger.py:185] completed step: 2, seconds: 0.010, TFLOP/s/device: 1313.401, Tokens/s/device: 197970.034, total_weights: 15161, loss: 9.823 I0416 04:06:27.109402 131296930094208 metric_logger.py:185] completed step: 3, seconds: 0.125, TFLOP/s/device: 109.074, Tokens/s/device: 16440.819, total_weights: 13327, loss: 9.405 I0416 04:06:27.383620 131183679637056 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0/items/array_metadatas/process_0 I0416 04:06:46.224901 131296930094208 metric_logger.py:185] completed step: 4, seconds: 0.109, TFLOP/s/device: 124.224, Tokens/s/device: 18724.400, total_weights: 11939, loss: 9.045 I0416 04:06:46.238376 131296930094208 metric_logger.py:185] completed step: 5, seconds: 0.117, TFLOP/s/device: 115.975, Tokens/s/device: 17480.966, total_weights: 15502, loss: 8.953 I0416 04:06:46.354693 131296930094208 metric_logger.py:185] completed step: 6, seconds: 19.119, TFLOP/s/device: 0.711, Tokens/s/device: 107.120, total_weights: 13864, loss: 8.775 I0416 04:06:46.471912 131296930094208 metric_logger.py:185] completed step: 7, seconds: 0.007, TFLOP/s/device: 1837.837, Tokens/s/device: 277018.802, total_weights: 12988, loss: 8.688 I0416 04:06:46.589165 131296930094208 metric_logger.py:185] completed step: 8, seconds: 0.116, TFLOP/s/device: 116.903, Tokens/s/device: 17620.842, total_weights: 13820, loss: 8.740 I0416 04:06:46.706598 131296930094208 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0416 04:06:46.707387 131296930094208 checkpointing.py:798] Waited 0.0008168220520019531 seconds for step 9 to finish before starting checkpointing. I0416 04:06:46.707671 131296930094208 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 04:06:46.710021 131183618819648 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 20.303401s (commit=19.853163s, array_metadata_write=0.450238s) I0416 04:06:46.711181 131183597848128 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 573.149 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 22 seconds) (per-host) I0416 04:06:46.711267 131183597848128 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.303710s. I0416 04:06:46.956376 131183597848128 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776312385393745863, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0416 04:06:47.133109 131183597848128 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 04:06:47.330842 131183660762688 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776312385393745863, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0416 04:06:47.531824 131183597848128 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 04:06:47.532479 131183597848128 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.535327s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0/items I0416 04:06:47.533191 131183597848128 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0/items I0416 04:06:47.766753 131183597848128 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0 I0416 04:06:48.422514 131183597848128 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0`. I0416 04:06:48.423134 131183597848128 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 23.844786s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/0 I0416 04:06:48.423222 131183597848128 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 22.015667s. I0416 04:06:48.423354 131183721580096 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 04:06:48.423404 131183721580096 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 04:06:48.423470 131183721580096 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 04:06:48.423517 131183721580096 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 04:06:48.423610 131296930094208 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0416 04:06:48.423666 131296930094208 checkpoint_manager.py:1441] Waiting for previous save to complete took 1.716026 seconds. If this number is high, consider checkpointing less frequently. I0416 04:06:48.424835 131296930094208 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 04:06:48.425083 131296930094208 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9. I0416 04:06:48.591368 131182633158208 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9 I0416 04:06:48.608015 131296930094208 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 04:06:48.608123 131296930094208 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0416 04:06:49.290233 131182580729408 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9/items I0416 04:06:58.649482 131296930094208 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.041845s I0416 04:06:58.650020 131296930094208 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.217 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 10 seconds) (per-host) I0416 04:06:58.650089 131296930094208 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 10.141846s (batch_requests_ready=0.096415s, total_serialization_initiated=10.044991s, others=0.000441s) I0416 04:06:58.650156 131296930094208 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 10.142372s (all_items=0.000025s, per_item={'items': '0.00002503'}, temp_paths=10.142347) I0416 04:06:58.654948 131203759867456 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 04:06:58.655059 131296930094208 async_checkpointer.py:561] Finished blocking save. Time taken: 10.230166s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9. I0416 04:06:58.655273 131296930094208 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 04:06:58.655435 131183811757632 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 04:06:58.655511 131296930094208 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776312406.7076213, 'wait_for_prev_duration_secs': 1.7160255908966064, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776312408.4248571, 'checkpointer_blocking_duration_secs': 10.23031735420227, 'get_old_steps_start_time': 1776312418.655191, 'get_old_steps_duration_secs': 4.315376281738281e-05, 'checkpoint_manager_blocking_start_time': 1776312406.70757, 'checkpoint_manager_blocking_duration_secs': 11.94791054725647} I0416 04:06:58.655627 131296930094208 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0416 04:06:58.655667 131296930094208 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 04:06:59.405991 131183534933568 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9/items/array_metadatas/process_0 I0416 04:07:29.419822 131182350042688 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 30.765383s (commit=30.334443s, array_metadata_write=0.430940s) I0416 04:07:29.421015 131203759867456 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 308.872 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 40 seconds) (per-host) I0416 04:07:29.421154 131203759867456 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 30.766047s. I0416 04:07:29.861371 131203759867456 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 04:07:30.297195 131203759867456 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 04:07:30.297370 131203759867456 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.573215s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9/items I0416 04:07:30.297999 131203759867456 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9/items I0416 04:07:30.524137 131203759867456 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9 I0416 04:07:31.193881 131203759867456 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9`. I0416 04:07:31.194533 131203759867456 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 42.769663s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_10_shardy_true/checkpoints/9 I0416 04:07:31.194639 131203759867456 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 32.539535s. I0416 04:07:31.194761 131183811757632 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 04:07:31.194810 131183811757632 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 04:07:31.194872 131183811757632 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 04:07:31.194916 131183811757632 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 04:07:31.195122 131296930094208 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0416 04:07:31.195730 131296930094208 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 04:07:31.196572 131296930094208 metric_logger.py:185] completed step: 9, seconds: 0.119, TFLOP/s/device: 113.847, Tokens/s/device: 17160.333, total_weights: 12300, loss: 8.710 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops