main (51c7f2b8e) vs feat/nnx-trainstate-and-training-loop (a2a803184)| Metric | main 51c7f2b8e | feat/nnx-trainstate-and-training-loop a2a803184 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.7100 | 8.7100 | 0 |
| TFLOP/s | 115.044 | 114.907 | -0.137 |
| Tok/s | 17340.6 | 17320.0 | -20.678 |
| Avg s/step | 2.444 | 2.384 | -0.06 |
| 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 21:41:09.622767: 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 21:41:09.738668 136039721368704 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 21:42:00.293326 136039721368704 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 21:42:00.293632 136039721368704 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 21:42:00.293689 136039721368704 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 21:42:00.293719 136039721368704 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 21:42:00.293902 136039721368704 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0415 21:42:00.803257 136039721368704 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0415 21:42:00.803637 136039721368704 checkpointing.py:677] Setting up checkpoint logger... I0415 21:42:00.803681 136039721368704 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 21:42:00.803761 136039721368704 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 21:42:00.804641 136039721368704 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 0x7bb9a19bf980>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 21:42:03.185611 136039721368704 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 21:42:03.185835 136039721368704 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 0x7bb3f47d1b80>}, handler_registry=None I0415 21:42:03.186203 136039721368704 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bb3f47d1b80>` 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 21:42:03.186247 136039721368704 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bb32433a120>` 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 21:42:03.186277 136039721368704 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 0x7bb3f47d1b80>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bb3f47d1b80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bb32433a120>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bb32433a120>}). I0415 21:42:03.186578 136039721368704 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 21:42:03.186635 136039721368704 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7bb32432d3a0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0415 21:42:03.307849 136039721368704 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints I0415 21:42:03.308109 136039721368704 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_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7bb324338830> I0415 21:42:03.308197 136039721368704 checkpointing.py:301] Checkpoint manager created! I0415 21:42:03.389778 136039721368704 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0415 21:42:03.392795 136039721368704 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 21:42:03.446468 136039721368704 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 21:42:03.475847 136039721368704 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:42:03.475919 136039721368704 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:42:04.472492 136039721368704 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 21:42:04.472597 136039721368704 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:42:04.575199 136039721368704 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:42:04.575276 136039721368704 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:42:04.590301 136039721368704 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:42:04.590347 136039721368704 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:42:04.612481 136039721368704 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 21:42:04.612539 136039721368704 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:42:04.627548 136039721368704 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 21:42:04.627601 136039721368704 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:42:04.642541 136039721368704 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 21:42:04.642588 136039721368704 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:42:04.664762 136039721368704 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 21:42:04.664819 136039721368704 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:42:04.683921 136039721368704 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 21:42:04.683971 136039721368704 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0415 21:42:04.871835 136039721368704 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 21:42:04.871932 136039721368704 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. [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 21:42:06.253221 136039721368704 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0415 21:42:06.253388 136039721368704 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 21:42:06.253429 136039721368704 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 21:42:06.253473 136039721368704 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 21:42:06.253511 136039721368704 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 21:42:06.253537 136039721368704 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 21:42:06.253575 136039721368704 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 21:42:06.253612 136039721368704 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 21:42:06.253639 136039721368704 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 21:42:06.253664 136039721368704 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 21:42:06.253698 136039721368704 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0415 21:42:06.253732 136039721368704 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0415 21:42:06.686072 136039721368704 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:42:06.686151 136039721368704 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0415 21:42:06.700444 136039721368704 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:42:06.700492 136039721368704 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0415 21:42:06.959914 3258423 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:42:07.071355 136039721368704 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 21:42:07.071813 136039721368704 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 21:42:07.071959 136039721368704 metric_logger.py:289] number parameters: 1.104 billion W0415 21:42:10.094393 3258423 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:42:10.221867 136039721368704 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0415 21:42:10.343204 136039721368704 checkpointing.py:776] Waited 0.12132096290588379 seconds for step 0 to finish before starting checkpointing. I0415 21:42:10.343669 136039721368704 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:42:10.343827 136039721368704 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0415 21:42:10.344347 136039721368704 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_false/checkpoints/0. I0415 21:42:10.444584 136039721368704 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0415 21:42:10.536906 136039721368704 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:42:10.537000 136039721368704 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 21:42:10.542665 135924979598912 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0 I0415 21:42:11.307525 135924969113152 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0/items I0415 21:42:11.432813 135924920878656 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289331176930324, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776289331176930324, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0/_CHECKPOINT_METADATA I0415 21:42:11.497538 136039721368704 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.961399s I0415 21:42:11.497999 136039721368704 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 11.726 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0415 21:42:11.498063 136039721368704 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.052487s (batch_requests_ready=0.087122s, total_serialization_initiated=0.965019s, others=0.000346s) I0415 21:42:11.498122 136039721368704 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.053052s (all_items=0.000022s, per_item={'items': '0.00002217'}, temp_paths=1.053030) I0415 21:42:11.498910 135924899907136 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:42:11.499009 136039721368704 async_checkpointer.py:561] Finished blocking save. Time taken: 1.155144s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0. I0415 21:42:11.499199 136039721368704 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:42:11.499377 135924990084672 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:42:11.499481 136039721368704 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776289330.3436546, 'wait_for_prev_duration_secs': 4.220008850097656e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776289330.3438473, 'checkpointer_blocking_duration_secs': 1.1552817821502686, 'get_old_steps_start_time': 1776289331.4991467, 'get_old_steps_duration_secs': 1.9550323486328125e-05, 'checkpoint_manager_blocking_start_time': 1776289330.3435602, 'checkpoint_manager_blocking_duration_secs': 1.1558918952941895} I0415 21:42:11.499587 136039721368704 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0415 21:42:11.499624 136039721368704 max_utils.py:750] Memstats: After params initialized: I0415 21:42:11.499707 136039721368704 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0415 21:42:11.499733 136039721368704 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0415 21:42:11.499753 136039721368704 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0415 21:42:11.499772 136039721368704 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0415 21:42:11.499792 136039721368704 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0415 21:42:11.499810 136039721368704 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0415 21:42:11.499828 136039721368704 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0415 21:42:11.499845 136039721368704 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0415 21:42:11.886571 3261490 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 21:42:11.886640 3261490 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 21:42:11.916396 136039721368704 metric_logger.py:185] completed step: 0, seconds: 3.149, TFLOP/s/device: 4.314, Tokens/s/device: 650.278, total_weights: 13328, loss: 10.828 I0415 21:42:11.917402 136039721368704 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/tensorboard/' I0415 21:42:12.018100 136039721368704 metric_logger.py:185] completed step: 1, seconds: 1.678, TFLOP/s/device: 8.095, Tokens/s/device: 1220.181, total_weights: 12332, loss: 10.828 I0415 21:42:12.136241 136039721368704 metric_logger.py:185] completed step: 2, seconds: 0.024, TFLOP/s/device: 564.086, Tokens/s/device: 85025.117, total_weights: 15161, loss: 9.823 I0415 21:42:12.254014 136039721368704 metric_logger.py:185] completed step: 3, seconds: 0.102, TFLOP/s/device: 133.049, Tokens/s/device: 20054.641, total_weights: 13327, loss: 9.405 I0415 21:42:12.601204 135924979598912 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_false/checkpoints/0/items/array_metadatas/process_0 I0415 21:42:31.682191 135924958627392 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 19.792724s (commit=19.361887s, array_metadata_write=0.430837s) I0415 21:42:31.683404 135924899907136 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 595.015 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 21 seconds) (per-host) I0415 21:42:31.683524 135924899907136 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.184452s. I0415 21:42:31.904834 135924899907136 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289331176930324, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0/_CHECKPOINT_METADATA I0415 21:42:31.972490 136039721368704 metric_logger.py:185] completed step: 4, seconds: 0.118, TFLOP/s/device: 114.988, Tokens/s/device: 17332.284, total_weights: 11939, loss: 9.045 I0415 21:42:31.981738 136039721368704 metric_logger.py:185] completed step: 5, seconds: 0.118, TFLOP/s/device: 115.480, Tokens/s/device: 17406.381, total_weights: 15502, loss: 8.953 I0415 21:42:32.093379 135924899907136 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 21:42:32.098398 136039721368704 metric_logger.py:185] completed step: 6, seconds: 19.718, TFLOP/s/device: 0.689, Tokens/s/device: 103.862, total_weights: 13864, loss: 8.775 I0415 21:42:32.216174 136039721368704 metric_logger.py:185] completed step: 7, seconds: 0.006, TFLOP/s/device: 2304.466, Tokens/s/device: 347354.138, total_weights: 12988, loss: 8.688 I0415 21:42:32.283105 135924920878656 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289331176930324, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0/_CHECKPOINT_METADATA I0415 21:42:32.333658 136039721368704 metric_logger.py:185] completed step: 8, seconds: 0.117, TFLOP/s/device: 115.817, Tokens/s/device: 17457.273, total_weights: 13820, loss: 8.740 I0415 21:42:32.451081 136039721368704 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0415 21:42:32.451872 136039721368704 checkpointing.py:776] Waited 0.0008378028869628906 seconds for step 9 to finish before starting checkpointing. I0415 21:42:32.452158 136039721368704 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:42:32.526726 135924899907136 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:42:32.527306 135924899907136 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.580979s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0/items I0415 21:42:32.527971 135924899907136 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0/items I0415 21:42:32.756483 135924899907136 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0 I0415 21:42:33.428232 135924899907136 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_false/checkpoints/0`. I0415 21:42:33.428875 135924899907136 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 23.085012s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/0 I0415 21:42:33.428954 135924899907136 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 21.929889s. I0415 21:42:33.429146 135924990084672 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:42:33.429256 135924990084672 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:42:33.429321 135924990084672 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:42:33.429374 135924990084672 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0415 21:42:33.429473 136039721368704 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. I0415 21:42:33.429705 136039721368704 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0415 21:42:33.430003 136039721368704 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_false/checkpoints/9. I0415 21:42:33.597255 136039721368704 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:42:33.597347 136039721368704 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 21:42:33.600870 135924899907136 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/9 I0415 21:42:34.288077 135924757300800 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/9/items I0415 21:42:37.565797 136039721368704 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 3.968920s I0415 21:42:37.569848 136039721368704 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 3.041 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 4 seconds) (per-host) I0415 21:42:37.569911 136039721368704 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 4.058314s (batch_requests_ready=0.082190s, total_serialization_initiated=3.972176s, others=0.003949s) I0415 21:42:37.569980 136039721368704 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 4.058816s (all_items=0.000025s, per_item={'items': '0.00002527'}, temp_paths=4.058790) I0415 21:42:37.572186 135924990084672 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:42:37.572292 136039721368704 async_checkpointer.py:561] Finished blocking save. Time taken: 4.142539s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/9. I0415 21:42:37.572506 136039721368704 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:42:37.572678 135924704872000 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:42:37.572777 136039721368704 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776289352.4521124, 'wait_for_prev_duration_secs': 0.9774034023284912, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776289353.4297311, 'checkpointer_blocking_duration_secs': 4.142669200897217, 'get_old_steps_start_time': 1776289357.572419, 'get_old_steps_duration_secs': 4.38690185546875e-05, 'checkpoint_manager_blocking_start_time': 1776289352.45207, 'checkpoint_manager_blocking_duration_secs': 5.1206817626953125} I0415 21:42:37.572907 136039721368704 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0415 21:42:37.572941 136039721368704 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:42:38.287317 135924117669440 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_false/checkpoints/9/items/array_metadatas/process_0 I0415 21:43:07.509412 135923289294400 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 29.938845s (commit=29.507184s, array_metadata_write=0.431662s) I0415 21:43:07.510511 135924990084672 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 371.683 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 33 seconds) (per-host) I0415 21:43:07.510617 135924990084672 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 29.938282s. I0415 21:43:07.954742 135924990084672 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 21:43:08.396914 135924990084672 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:43:08.397669 135924990084672 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.587327s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/9/items I0415 21:43:08.398324 135924990084672 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/9/items I0415 21:43:08.639738 135924990084672 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/9 I0415 21:43:09.300949 135924990084672 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_false/checkpoints/9`. I0415 21:43:09.301583 135924990084672 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 35.871837s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_10_shardy_false/checkpoints/9 I0415 21:43:09.301658 135924990084672 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 31.729324s. I0415 21:43:09.301819 135924704872000 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:43:09.301916 135924704872000 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:43:09.301975 135924704872000 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:43:09.302023 135924704872000 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0415 21:43:09.302175 136039721368704 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 21:43:09.302295 136039721368704 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:43:09.302814 136039721368704 metric_logger.py:185] completed step: 9, seconds: 0.118, TFLOP/s/device: 115.044, Tokens/s/device: 17340.649, 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
a2a803184
· feat_nnx_trainstate_and_training_loop_20260416_004836
· full log
2026-04-16 01:19:17.086717: 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 01:19:17.204298 128444442295424 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 01:20:09.926919 128444442295424 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 01:20:09.927033 128444442295424 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 01:20:09.927086 128444442295424 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 01:20:09.927111 128444442295424 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 01:20:09.927194 128444442295424 train.py:812] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0416 01:20:10.450918 128444442295424 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 01:20:10.451126 128444442295424 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 01:20:10.451304 128444442295424 checkpointing.py:688] Setting up checkpoint logger... I0416 01:20:10.451344 128444442295424 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 01:20:10.451385 128444442295424 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 01:20:10.451822 128444442295424 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 0x74d1373fd070>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 01:20:12.830772 128444442295424 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 01:20:12.830967 128444442295424 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 0x74cc31dadb20>}, handler_registry=None I0416 01:20:12.831243 128444442295424 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x74cc31dadb20>` 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 01:20:12.831283 128444442295424 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x74cabab46030>` 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 01:20:12.831310 128444442295424 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 0x74cc31dadb20>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x74cc31dadb20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x74cabab46030>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x74cabab46030>}). I0416 01:20:12.831607 128444442295424 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 01:20:12.831658 128444442295424 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x74caba7420c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 01:20:12.954627 128444442295424 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints I0416 01:20:12.954875 128444442295424 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_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x74cabb9a7770> I0416 01:20:12.954955 128444442295424 checkpointing.py:302] Checkpoint manager created! I0416 01:20:13.035483 128444442295424 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 01:20:13.038794 128444442295424 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 01:20:13.093346 128444442295424 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 01:20:13.123651 128444442295424 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 01:20:13.123720 128444442295424 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 01:20:14.216354 128444442295424 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 01:20:14.216543 128444442295424 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 01:20:14.320747 128444442295424 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 01:20:14.320834 128444442295424 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 01:20:14.335735 128444442295424 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 01:20:14.335788 128444442295424 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 01:20:14.358371 128444442295424 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 01:20:14.358433 128444442295424 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 01:20:14.373389 128444442295424 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 01:20:14.373440 128444442295424 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 01:20:14.388291 128444442295424 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 01:20:14.388341 128444442295424 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 01:20:14.410717 128444442295424 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 01:20:14.410787 128444442295424 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 01:20:14.429708 128444442295424 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 01:20:14.429772 128444442295424 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 01:20:14.626240 128444442295424 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 01:20:14.626343 128444442295424 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. [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 01:20:15.997709 128444442295424 maxtext_utils.py:1790] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 01:20:15.997824 128444442295424 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 01:20:15.997865 128444442295424 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 01:20:15.997910 128444442295424 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 01:20:15.997951 128444442295424 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 01:20:15.997983 128444442295424 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 01:20:15.998021 128444442295424 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 01:20:15.998075 128444442295424 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 01:20:15.998104 128444442295424 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 01:20:15.998131 128444442295424 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 01:20:15.998166 128444442295424 maxtext_utils.py:1790] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 01:20:15.998199 128444442295424 maxtext_utils.py:1790] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 01:20:16.441731 128444442295424 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 01:20:16.441825 128444442295424 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 01:20:16.456449 128444442295424 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 01:20:16.456502 128444442295424 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 01:20:16.738357 3479836 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 01:20:16.851775 128444442295424 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 01:20:16.852367 128444442295424 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 01:20:16.852512 128444442295424 metric_logger.py:289] number parameters: 1.104 billion W0416 01:20:19.964880 3479836 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 01:20:20.090032 128444442295424 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 01:20:20.212133 128444442295424 checkpointing.py:798] Waited 0.12208056449890137 seconds for step 0 to finish before starting checkpointing. I0416 01:20:20.212601 128444442295424 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 01:20:20.212759 128444442295424 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 01:20:20.213300 128444442295424 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0. I0416 01:20:20.305360 128444442295424 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 01:20:20.401782 128444442295424 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 01:20:20.401880 128444442295424 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 01:20:20.407843 128329644508736 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0 I0416 01:20:21.202125 128329634022976 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0/items I0416 01:20:21.362589 128329577399872 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776302421118627689, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776302421118627689, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0/_CHECKPOINT_METADATA I0416 01:20:21.673408 3482891 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 01:20:21.673470 3482891 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 01:20:21.675777 128444442295424 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.274810s I0416 01:20:21.677090 128444442295424 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 9.007 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 01:20:21.677170 128444442295424 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.370368s (batch_requests_ready=0.090623s, total_serialization_initiated=1.278846s, others=0.000899s) I0416 01:20:21.677259 128444442295424 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.371330s (all_items=0.000023s, per_item={'items': '0.00002337'}, temp_paths=1.371307) I0416 01:20:21.679563 128329654994496 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 01:20:21.679849 128444442295424 async_checkpointer.py:561] Finished blocking save. Time taken: 1.467044s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0. I0416 01:20:21.680038 128444442295424 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 01:20:21.680308 128329493513792 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 01:20:21.680733 128444442295424 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776302420.2125883, 'wait_for_prev_duration_secs': 4.2438507080078125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776302420.2127805, 'checkpointer_blocking_duration_secs': 1.4671547412872314, 'get_old_steps_start_time': 1776302421.6799638, 'get_old_steps_duration_secs': 3.075599670410156e-05, 'checkpoint_manager_blocking_start_time': 1776302420.2124984, 'checkpoint_manager_blocking_duration_secs': 1.4681942462921143} I0416 01:20:21.680848 128444442295424 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 01:20:21.680900 128444442295424 max_utils.py:750] Memstats: After params initialized: I0416 01:20:21.680954 128444442295424 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0416 01:20:21.680983 128444442295424 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 01:20:21.681006 128444442295424 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 01:20:21.681027 128444442295424 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 01:20:21.681062 128444442295424 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 01:20:21.681084 128444442295424 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 01:20:21.681104 128444442295424 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 01:20:21.681125 128444442295424 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0416 01:20:22.114993 128444442295424 metric_logger.py:185] completed step: 0, seconds: 3.237, TFLOP/s/device: 4.197, Tokens/s/device: 632.679, total_weights: 13328, loss: 10.828 I0416 01:20:22.116066 128444442295424 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/tensorboard/' I0416 01:20:22.230578 128444442295424 metric_logger.py:185] completed step: 1, seconds: 2.023, TFLOP/s/device: 6.718, Tokens/s/device: 1012.583, total_weights: 12332, loss: 10.828 I0416 01:20:22.348180 128444442295424 metric_logger.py:185] completed step: 2, seconds: 0.011, TFLOP/s/device: 1244.813, Tokens/s/device: 187631.699, total_weights: 15161, loss: 9.823 I0416 01:20:22.454266 128329535456832 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_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0/items/array_metadatas/process_0 I0416 01:20:22.465852 128444442295424 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.473, Tokens/s/device: 17556.041, total_weights: 13327, loss: 9.405 I0416 01:20:41.292744 128444442295424 metric_logger.py:185] completed step: 4, seconds: 0.119, TFLOP/s/device: 113.881, Tokens/s/device: 17165.368, total_weights: 11939, loss: 9.045 I0416 01:20:41.301276 128444442295424 metric_logger.py:185] completed step: 5, seconds: 0.116, TFLOP/s/device: 117.268, Tokens/s/device: 17675.896, total_weights: 15502, loss: 8.953 I0416 01:20:41.418757 128444442295424 metric_logger.py:185] completed step: 6, seconds: 18.825, TFLOP/s/device: 0.722, Tokens/s/device: 108.792, total_weights: 13864, loss: 8.775 I0416 01:20:41.537233 128444442295424 metric_logger.py:185] completed step: 7, seconds: 0.006, TFLOP/s/device: 2282.784, Tokens/s/device: 344086.022, total_weights: 12988, loss: 8.688 I0416 01:20:41.654947 128444442295424 metric_logger.py:185] completed step: 8, seconds: 0.119, TFLOP/s/device: 114.497, Tokens/s/device: 17258.233, total_weights: 13820, loss: 8.740 I0416 01:20:41.771968 128444442295424 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0416 01:20:41.772822 128444442295424 checkpointing.py:798] Waited 0.0008895397186279297 seconds for step 9 to finish before starting checkpointing. I0416 01:20:41.773114 128444442295424 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 01:20:42.880339 128329503999552 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 21.202044s (commit=20.787848s, array_metadata_write=0.414195s) I0416 01:20:42.881888 128329654994496 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 559.770 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 22 seconds) (per-host) I0416 01:20:42.882059 128329654994496 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 21.202246s. I0416 01:20:43.112029 128329654994496 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776302421118627689, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0/_CHECKPOINT_METADATA I0416 01:20:43.326318 128329654994496 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 01:20:43.515252 128329577399872 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776302421118627689, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0/_CHECKPOINT_METADATA I0416 01:20:43.697994 128329654994496 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 01:20:43.698924 128329654994496 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.546195s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0/items I0416 01:20:43.699730 128329654994496 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0/items I0416 01:20:43.943128 128329654994496 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0 I0416 01:20:44.644778 128329654994496 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_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0`. I0416 01:20:44.645567 128329654994496 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 24.432774s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/0 I0416 01:20:44.645682 128329654994496 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 22.965883s. I0416 01:20:44.645954 128329493513792 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 01:20:44.646109 128329493513792 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 01:20:44.646198 128329493513792 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 01:20:44.646259 128329493513792 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 01:20:44.646486 128444442295424 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 01:20:44.646647 128444442295424 checkpoint_manager.py:1441] Waiting for previous save to complete took 2.873529 seconds. If this number is high, consider checkpointing less frequently. I0416 01:20:44.647818 128444442295424 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 01:20:44.648186 128444442295424 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9. I0416 01:20:44.828560 128329493513792 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9 I0416 01:20:44.836682 128444442295424 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 01:20:44.836792 128444442295424 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 01:20:45.520325 128328533018176 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9/items I0416 01:20:55.445669 128444442295424 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.609476s I0416 01:20:55.450625 128444442295424 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.152 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 10 seconds) (per-host) I0416 01:20:55.450700 128444442295424 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 10.709807s (batch_requests_ready=0.092308s, total_serialization_initiated=10.612659s, others=0.004840s) I0416 01:20:55.450780 128444442295424 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 10.710386s (all_items=0.000022s, per_item={'items': '0.00002217'}, temp_paths=10.710364) I0416 01:20:55.452610 128328533018176 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 01:20:55.452702 128444442295424 async_checkpointer.py:561] Finished blocking save. Time taken: 10.804831s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9. I0416 01:20:55.452905 128444442295424 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 01:20:55.453105 128329654994496 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 01:20:55.453220 128444442295424 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776302441.7730856, 'wait_for_prev_duration_secs': 2.8735289573669434, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776302444.6478453, 'checkpointer_blocking_duration_secs': 10.804969549179077, 'get_old_steps_start_time': 1776302455.4528356, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1776302441.7730227, 'checkpoint_manager_blocking_duration_secs': 13.680165529251099} I0416 01:20:55.453352 128444442295424 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0416 01:20:55.453389 128444442295424 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 01:20:56.215225 128328040187456 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_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9/items/array_metadatas/process_0 I0416 01:21:26.952153 128328449132096 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 31.500533s (commit=31.083344s, array_metadata_write=0.417189s) I0416 01:21:26.953285 128328533018176 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 299.362 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 42 seconds) (per-host) I0416 01:21:26.953360 128328533018176 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 31.500609s. I0416 01:21:27.376347 128328533018176 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 01:21:27.759970 128328533018176 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 01:21:27.760181 128328533018176 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.526819s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9/items I0416 01:21:27.760904 128328533018176 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9/items I0416 01:21:27.983854 128328533018176 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9 I0416 01:21:28.638581 128328533018176 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_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9`. I0416 01:21:28.639334 128328533018176 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 43.991473s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_10_shardy_false/checkpoints/9 I0416 01:21:28.639405 128328533018176 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 33.186657s. I0416 01:21:28.639605 128329654994496 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 01:21:28.639731 128329654994496 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 01:21:28.639810 128329654994496 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 01:21:28.639858 128329654994496 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 01:21:28.640063 128444442295424 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 01:21:28.640209 128444442295424 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 01:21:28.640925 128444442295424 metric_logger.py:185] completed step: 9, seconds: 0.118, TFLOP/s/device: 114.907, Tokens/s/device: 17319.971, 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