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 | 116.020 | 115.591 | -0.429 |
| Tok/s | 17487.8 | 17423.1 | -64.717 |
| Avg s/step | 2.400 | 2.353 | -0.047 |
| 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 21:11:33.703749: 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 21:11:33.819891 126656459193472 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 21:12:15.844906 126656459193472 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 21:12:15.845007 126656459193472 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 21:12:15.845041 126656459193472 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 21:12:15.845081 126656459193472 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 21:12:15.845155 126656459193472 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 21:12:15.937455 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:12:16.342119 126656459193472 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0415 21:12:16.342369 126656459193472 checkpointing.py:677] Setting up checkpoint logger... I0415 21:12:16.342413 126656459193472 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 21:12:16.342453 126656459193472 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 21:12:16.342889 126656459193472 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 0x7330ed712f30>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 21:12:19.148542 126656459193472 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 21:12:19.148702 126656459193472 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 0x732a6bf5ddf0>}, handler_registry=None I0415 21:12:19.148934 126656459193472 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x732a6bf5ddf0>` 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 21:12:19.148973 126656459193472 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x732a6bf5e030>` 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 21:12:19.149002 126656459193472 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 0x732a6bf5ddf0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x732a6bf5ddf0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x732a6bf5e030>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x732a6bf5e030>}). I0415 21:12:19.149300 126656459193472 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 21:12:19.149356 126656459193472 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x732a6bf56020> timeout: 600 secs and primary_host=0 for async checkpoint writes I0415 21:12:19.274700 126656459193472 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints I0415 21:12:19.274919 126656459193472 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_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x732a6bf5d4f0> I0415 21:12:19.274993 126656459193472 checkpointing.py:301] Checkpoint manager created! I0415 21:12:19.353533 126656459193472 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0415 21:12:19.356778 126656459193472 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 21:12:19.407432 126656459193472 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 21:12:19.432666 126656459193472 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:12:19.432731 126656459193472 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:12:20.402855 126656459193472 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 21:12:20.402995 126656459193472 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:12:20.505620 126656459193472 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:12:20.505695 126656459193472 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:12:20.520683 126656459193472 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:12:20.520732 126656459193472 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:12:20.542182 126656459193472 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 21:12:20.542239 126656459193472 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:12:20.557163 126656459193472 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 21:12:20.557209 126656459193472 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:12:20.572261 126656459193472 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 21:12:20.572309 126656459193472 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:12:20.594783 126656459193472 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 21:12:20.594839 126656459193472 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:12:20.613653 126656459193472 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 21:12:20.613708 126656459193472 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0415 21:12:20.800975 126656459193472 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 21:12:20.801084 126656459193472 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0415 21:12:20.851137 3223486 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 21:12:21.259603 126656459193472 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0415 21:12:21.259719 126656459193472 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 21:12:21.259760 126656459193472 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 21:12:21.259804 126656459193472 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 21:12:21.259848 126656459193472 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 21:12:21.259877 126656459193472 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 21:12:21.259915 126656459193472 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 21:12:21.259953 126656459193472 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 21:12:21.259982 126656459193472 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 21:12:21.260007 126656459193472 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 21:12:21.260057 126656459193472 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0415 21:12:21.260096 126656459193472 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0415 21:12:21.695218 126656459193472 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:12:21.695367 126656459193472 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0415 21:12:21.709610 126656459193472 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:12:21.709659 126656459193472 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0415 21:12:21.980259 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:12:22.094864 126656459193472 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 21:12:22.095367 126656459193472 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 21:12:22.095527 126656459193472 metric_logger.py:289] number parameters: 1.104 billion W0415 21:12:25.134171 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:12:25.264807 126656459193472 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0415 21:12:25.384924 126656459193472 checkpointing.py:776] Waited 0.12009882926940918 seconds for step 0 to finish before starting checkpointing. I0415 21:12:25.385491 126656459193472 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:12:25.385674 126656459193472 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0415 21:12:25.386199 126656459193472 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0. I0415 21:12:25.477458 126656459193472 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0415 21:12:25.572492 126543305115200 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0 I0415 21:12:25.577911 126656459193472 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:12:25.577968 126656459193472 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 21:12:25.794204 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:25.808933 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:25.821613 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:25.836955 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:25.844947 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:25.852704 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:25.858785 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:25.863160 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:12:26.223332 126656459193472 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.645824s I0415 21:12:26.223773 126656459193472 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 16.567 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 744 milliseconds) (per-host) I0415 21:12:26.223825 126656459193472 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.745023s (batch_requests_ready=0.095298s, total_serialization_initiated=0.649383s, others=0.000342s) I0415 21:12:26.223885 126656459193472 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.745933s (all_items=0.000040s, per_item={'items': '0.00003982'}, temp_paths=0.745893) I0415 21:12:26.224735 126543233812032 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:12:26.224828 126656459193472 async_checkpointer.py:561] Finished blocking save. Time taken: 0.839111s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0. I0415 21:12:26.225013 126656459193472 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:12:26.225272 126543315600960 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:12:26.225385 126656459193472 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776287545.3854744, 'wait_for_prev_duration_secs': 4.4345855712890625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776287545.3856971, 'checkpointer_blocking_duration_secs': 0.8392300605773926, 'get_old_steps_start_time': 1776287546.2249463, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776287545.385346, 'checkpoint_manager_blocking_duration_secs': 0.8400068283081055} I0415 21:12:26.225498 126656459193472 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0415 21:12:26.225539 126656459193472 max_utils.py:750] Memstats: After params initialized: I0415 21:12:26.225615 126656459193472 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0415 21:12:26.225642 126656459193472 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0415 21:12:26.225668 126656459193472 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0415 21:12:26.225686 126656459193472 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0415 21:12:26.225703 126656459193472 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0415 21:12:26.225718 126656459193472 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0415 21:12:26.225735 126656459193472 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0415 21:12:26.225751 126656459193472 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0415 21:12:26.231562 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:26.237226 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:12:26.241925 3223486 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:12:26.338111 126543294629440 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0/items I0415 21:12:26.566017 126543223326272 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776287546242374923, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776287546242374923, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0/_CHECKPOINT_METADATA I0415 21:12:26.575262 126656459193472 metric_logger.py:185] completed step: 0, seconds: 3.169, TFLOP/s/device: 4.288, Tokens/s/device: 646.317, total_weights: 13328, loss: 10.828 I0415 21:12:26.576273 126656459193472 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/tensorboard/' I0415 21:12:26.691093 126656459193472 metric_logger.py:185] completed step: 1, seconds: 1.308, TFLOP/s/device: 10.385, Tokens/s/device: 1565.335, total_weights: 12332, loss: 10.828 I0415 21:12:26.808559 126656459193472 metric_logger.py:185] completed step: 2, seconds: 0.010, TFLOP/s/device: 1410.038, Tokens/s/device: 212536.322, total_weights: 15161, loss: 9.823 I0415 21:12:26.926216 126656459193472 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.462, Tokens/s/device: 17554.386, total_weights: 13327, loss: 9.405 I0415 21:12:27.022715 3226238 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 21:12:27.022782 3226238 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 21:12:27.850978 126543254783552 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_01_base/checkpoints/0/items/array_metadatas/process_0 I0415 21:12:46.619843 126656459193472 metric_logger.py:185] completed step: 4, seconds: 0.117, TFLOP/s/device: 115.697, Tokens/s/device: 17439.138, total_weights: 11939, loss: 9.045 I0415 21:12:46.628441 126656459193472 metric_logger.py:185] completed step: 5, seconds: 0.118, TFLOP/s/device: 115.622, Tokens/s/device: 17427.859, total_weights: 15502, loss: 8.953 I0415 21:12:46.745103 126656459193472 metric_logger.py:185] completed step: 6, seconds: 19.693, TFLOP/s/device: 0.690, Tokens/s/device: 103.996, total_weights: 13864, loss: 8.775 I0415 21:12:46.862621 126656459193472 metric_logger.py:185] completed step: 7, seconds: 0.006, TFLOP/s/device: 2271.336, Tokens/s/device: 342360.415, total_weights: 12988, loss: 8.688 I0415 21:12:46.979918 126656459193472 metric_logger.py:185] completed step: 8, seconds: 0.118, TFLOP/s/device: 115.370, Tokens/s/device: 17389.828, total_weights: 13820, loss: 8.740 I0415 21:12:47.097076 126656459193472 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0415 21:12:47.097909 126656459193472 checkpointing.py:776] Waited 0.0008687973022460938 seconds for step 9 to finish before starting checkpointing. I0415 21:12:47.098199 126656459193472 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:12:55.295166 126543244297792 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 28.270729s (commit=27.809231s, array_metadata_write=0.461498s) I0415 21:12:55.296311 126543233812032 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 423.805 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 29 seconds) (per-host) I0415 21:12:55.296425 126543233812032 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 29.071548s. I0415 21:12:55.551182 126543233812032 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776287546242374923, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0/_CHECKPOINT_METADATA I0415 21:12:55.724934 126543233812032 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 21:12:55.930019 126543223326272 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776287546242374923, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0/_CHECKPOINT_METADATA I0415 21:12:56.142522 126543233812032 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:12:56.143217 126543233812032 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.561481s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0/items I0415 21:12:56.143884 126543233812032 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0/items I0415 21:12:56.402038 126543233812032 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0 I0415 21:12:57.090754 126543233812032 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_01_base/checkpoints/0`. I0415 21:12:57.091398 126543233812032 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 31.705687s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/0 I0415 21:12:57.091477 126543233812032 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 30.866601s. I0415 21:12:57.091656 126543315600960 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:12:57.091766 126543315600960 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:12:57.091831 126543315600960 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:12:57.091886 126543315600960 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0415 21:12:57.092010 126656459193472 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 21:12:57.092133 126656459193472 checkpoint_manager.py:1441] Waiting for previous save to complete took 9.993946 seconds. If this number is high, consider checkpointing less frequently. I0415 21:12:57.093126 126656459193472 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0415 21:12:57.093399 126656459193472 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/9. I0415 21:12:57.267694 126543315600960 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/9 I0415 21:12:57.276693 126656459193472 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:12:57.276782 126656459193472 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 21:12:58.005787 126543087011392 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/9/items I0415 21:13:07.928303 126656459193472 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.651998s I0415 21:13:07.932434 126656459193472 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.148 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 10 seconds) (per-host) I0415 21:13:07.932499 126656459193472 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 10.745671s (batch_requests_ready=0.086284s, total_serialization_initiated=10.655352s, others=0.004035s) I0415 21:13:07.932569 126656459193472 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 10.746162s (all_items=0.000020s, per_item={'items': '0.00002003'}, temp_paths=10.746142) I0415 21:13:07.934025 126543315600960 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:13:07.934138 126656459193472 async_checkpointer.py:561] Finished blocking save. Time taken: 10.840965s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/9. I0415 21:13:07.934392 126656459193472 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:13:07.934537 126543233812032 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:13:07.934621 126656459193472 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776287567.0981588, 'wait_for_prev_duration_secs': 9.993945598602295, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776287577.0931523, 'checkpointer_blocking_duration_secs': 10.841151475906372, 'get_old_steps_start_time': 1776287587.934319, 'get_old_steps_duration_secs': 3.4809112548828125e-05, 'checkpoint_manager_blocking_start_time': 1776287567.0981073, 'checkpoint_manager_blocking_duration_secs': 20.836487293243408} I0415 21:13:07.934708 126656459193472 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0415 21:13:07.934741 126656459193472 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:13:08.690636 126543305115200 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_01_base/checkpoints/9/items/array_metadatas/process_0 I0415 21:13:36.737218 126543087011392 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 28.804125s (commit=28.355352s, array_metadata_write=0.448773s) I0415 21:13:36.738251 126543315600960 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 319.503 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 39 seconds) (per-host) I0415 21:13:36.738359 126543315600960 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 28.804128s. I0415 21:13:37.147952 126543315600960 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 21:13:37.549684 126543315600960 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:13:37.550525 126543315600960 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.547024s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/9/items I0415 21:13:37.551178 126543315600960 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/9/items I0415 21:13:37.804370 126543315600960 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/9 I0415 21:13:38.505670 126543315600960 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_01_base/checkpoints/9`. I0415 21:13:38.506307 126543315600960 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 41.413140s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_01_base/checkpoints/9 I0415 21:13:38.506384 126543315600960 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 30.572154s. I0415 21:13:38.506483 126543233812032 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:13:38.506532 126543233812032 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:13:38.506585 126543233812032 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:13:38.506625 126543233812032 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0415 21:13:38.506824 126656459193472 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 21:13:38.507255 126656459193472 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:13:38.507886 126656459193472 metric_logger.py:185] completed step: 9, seconds: 0.117, TFLOP/s/device: 116.020, Tokens/s/device: 17487.832, 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 03:35:39.316802: 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 03:35:39.433112 132919054130304 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 03:36:25.880275 132919054130304 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 03:36:25.880378 132919054130304 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 03:36:25.880416 132919054130304 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 03:36:25.880442 132919054130304 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 03:36:25.880526 132919054130304 train.py:818] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 03:36:25.974440 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:36:26.369920 132919054130304 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 03:36:26.370102 132919054130304 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 03:36:26.370327 132919054130304 checkpointing.py:688] Setting up checkpoint logger... I0416 03:36:26.370365 132919054130304 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 03:36:26.370404 132919054130304 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 03:36:26.370945 132919054130304 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 0x78e30abd7bf0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 03:36:29.183788 132919054130304 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 03:36:29.183964 132919054130304 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 0x78dc8d745010>}, handler_registry=None I0416 03:36:29.184214 132919054130304 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78dc8d745010>` 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 03:36:29.184257 132919054130304 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78dc8d746ab0>` 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 03:36:29.184285 132919054130304 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 0x78dc8d745010>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78dc8d745010>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78dc8d746ab0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78dc8d746ab0>}). I0416 03:36:29.184570 132919054130304 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 03:36:29.184621 132919054130304 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x78dc8d74a160> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 03:36:29.315937 132919054130304 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_01_base/checkpoints I0416 03:36:29.316189 132919054130304 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_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x78de94580530> I0416 03:36:29.316269 132919054130304 checkpointing.py:302] Checkpoint manager created! I0416 03:36:29.394538 132919054130304 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 03:36:29.397434 132919054130304 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 03:36:29.449227 132919054130304 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 03:36:29.475140 132919054130304 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 03:36:29.475208 132919054130304 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 03:36:30.475619 132919054130304 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 03:36:30.475715 132919054130304 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 03:36:30.578325 132919054130304 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 03:36:30.578401 132919054130304 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 03:36:30.593317 132919054130304 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 03:36:30.593365 132919054130304 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 03:36:30.614469 132919054130304 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 03:36:30.614524 132919054130304 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 03:36:30.629497 132919054130304 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 03:36:30.629546 132919054130304 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 03:36:30.644467 132919054130304 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 03:36:30.644518 132919054130304 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 03:36:30.686302 132919054130304 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 03:36:30.686379 132919054130304 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 03:36:30.705153 132919054130304 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 03:36:30.705210 132919054130304 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 03:36:30.900231 132919054130304 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 03:36:30.900328 132919054130304 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0416 03:36:30.950376 3623483 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 03:36:31.337977 132919054130304 maxtext_utils.py:1790] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 03:36:31.338104 132919054130304 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 03:36:31.338148 132919054130304 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 03:36:31.338193 132919054130304 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 03:36:31.338233 132919054130304 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 03:36:31.338260 132919054130304 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 03:36:31.338300 132919054130304 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 03:36:31.338338 132919054130304 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 03:36:31.338366 132919054130304 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 03:36:31.338392 132919054130304 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 03:36:31.338426 132919054130304 maxtext_utils.py:1790] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 03:36:31.338459 132919054130304 maxtext_utils.py:1790] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 03:36:31.773137 132919054130304 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 03:36:31.773218 132919054130304 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 03:36:31.787851 132919054130304 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 03:36:31.787903 132919054130304 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 03:36:32.081462 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:36:32.198503 132919054130304 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 03:36:32.198979 132919054130304 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 03:36:32.199149 132919054130304 metric_logger.py:289] number parameters: 1.104 billion W0416 03:36:35.254917 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:36:35.389869 132919054130304 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 03:36:35.510268 132919054130304 checkpointing.py:798] Waited 0.12038660049438477 seconds for step 0 to finish before starting checkpointing. I0416 03:36:35.510707 132919054130304 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 03:36:35.510862 132919054130304 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 03:36:35.511245 132919054130304 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_01_base/checkpoints/0. I0416 03:36:35.597910 132919054130304 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 03:36:35.689556 132805822514752 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_01_base/checkpoints/0 I0416 03:36:35.694649 132919054130304 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 03:36:35.694740 132919054130304 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 03:36:35.909268 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:35.923226 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:35.936552 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:35.951368 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:35.958478 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:35.963732 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:35.968683 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:35.974561 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:36:36.251748 132919054130304 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.557512s I0416 03:36:36.252455 132919054130304 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 18.894 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 653 milliseconds) (per-host) I0416 03:36:36.252614 132919054130304 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.653468s (batch_requests_ready=0.091725s, total_serialization_initiated=0.561133s, others=0.000610s) I0416 03:36:36.252741 132919054130304 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.654377s (all_items=0.000035s, per_item={'items': '0.00003529'}, temp_paths=0.654342) I0416 03:36:36.253656 132805751211584 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 03:36:36.253753 132919054130304 async_checkpointer.py:561] Finished blocking save. Time taken: 0.742851s. 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_01_base/checkpoints/0. I0416 03:36:36.253935 132919054130304 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 03:36:36.254184 132805833000512 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 03:36:36.254293 132919054130304 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_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776310595.5106952, 'wait_for_prev_duration_secs': 4.100799560546875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776310595.510883, 'checkpointer_blocking_duration_secs': 0.7429747581481934, 'get_old_steps_start_time': 1776310596.253878, 'get_old_steps_duration_secs': 2.1696090698242188e-05, 'checkpoint_manager_blocking_start_time': 1776310595.5106122, 'checkpoint_manager_blocking_duration_secs': 0.7436504364013672} I0416 03:36:36.254411 132919054130304 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 03:36:36.254459 132919054130304 max_utils.py:750] Memstats: After params initialized: I0416 03:36:36.254568 132919054130304 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0416 03:36:36.254603 132919054130304 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 03:36:36.254625 132919054130304 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 03:36:36.254644 132919054130304 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 03:36:36.254664 132919054130304 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 03:36:36.254682 132919054130304 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 03:36:36.254701 132919054130304 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 03:36:36.254718 132919054130304 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 03:36:36.260820 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:36.266226 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:36:36.270831 3623483 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:36:36.565803 132805812028992 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_01_base/checkpoints/0/items I0416 03:36:36.574652 132919054130304 metric_logger.py:185] completed step: 0, seconds: 3.190, TFLOP/s/device: 4.259, Tokens/s/device: 641.965, total_weights: 13328, loss: 10.828 I0416 03:36:36.575635 132919054130304 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_01_base/tensorboard/' I0416 03:36:36.620292 132805732337216 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776310596368309281, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776310596368309281, "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_01_base/checkpoints/0/_CHECKPOINT_METADATA I0416 03:36:36.690776 132919054130304 metric_logger.py:185] completed step: 1, seconds: 1.183, TFLOP/s/device: 11.487, Tokens/s/device: 1731.461, total_weights: 12332, loss: 10.828 I0416 03:36:36.808161 132919054130304 metric_logger.py:185] completed step: 2, seconds: 0.009, TFLOP/s/device: 1445.132, Tokens/s/device: 217825.994, total_weights: 15161, loss: 9.823 I0416 03:36:36.925319 132919054130304 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 115.800, Tokens/s/device: 17454.595, total_weights: 13327, loss: 9.405 I0416 03:36:37.069293 3626261 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 03:36:37.069363 3626261 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 03:36:37.858186 132805772183104 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_01_base/checkpoints/0/items/array_metadatas/process_0 I0416 03:36:56.319576 132919054130304 metric_logger.py:185] completed step: 4, seconds: 0.117, TFLOP/s/device: 115.851, Tokens/s/device: 17462.334, total_weights: 11939, loss: 9.045 I0416 03:36:56.329315 132919054130304 metric_logger.py:185] completed step: 5, seconds: 0.118, TFLOP/s/device: 115.423, Tokens/s/device: 17397.805, total_weights: 15502, loss: 8.953 I0416 03:36:56.446124 132919054130304 metric_logger.py:185] completed step: 6, seconds: 19.394, TFLOP/s/device: 0.701, Tokens/s/device: 105.600, total_weights: 13864, loss: 8.775 I0416 03:36:56.563624 132919054130304 metric_logger.py:185] completed step: 7, seconds: 0.006, TFLOP/s/device: 2224.845, Tokens/s/device: 335352.874, total_weights: 12988, loss: 8.688 I0416 03:36:56.680843 132919054130304 metric_logger.py:185] completed step: 8, seconds: 0.118, TFLOP/s/device: 115.626, Tokens/s/device: 17428.452, total_weights: 13820, loss: 8.740 I0416 03:36:56.797989 132919054130304 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0416 03:36:56.798802 132919054130304 checkpointing.py:798] Waited 0.0008423328399658203 seconds for step 9 to finish before starting checkpointing. I0416 03:36:56.799089 132919054130304 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 03:37:02.338809 132805761697344 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 25.266909s (commit=24.799941s, array_metadata_write=0.466968s) I0416 03:37:02.339947 132805751211584 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 472.566 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 26 seconds) (per-host) I0416 03:37:02.340067 132805751211584 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 26.086267s. I0416 03:37:02.654901 132805751211584 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776310596368309281, '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_01_base/checkpoints/0/_CHECKPOINT_METADATA I0416 03:37:02.850572 132805751211584 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 03:37:03.042834 132805732337216 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776310596368309281, '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_01_base/checkpoints/0/_CHECKPOINT_METADATA I0416 03:37:03.310146 132805751211584 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 03:37:03.310358 132805751211584 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.617896s. 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_01_base/checkpoints/0/items I0416 03:37:03.311031 132805751211584 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_01_base/checkpoints/0/items I0416 03:37:03.567805 132805751211584 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_01_base/checkpoints/0 I0416 03:37:04.249632 132805751211584 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_01_base/checkpoints/0`. I0416 03:37:04.250277 132805751211584 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 28.739382s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_01_base/checkpoints/0 I0416 03:37:04.250349 132805751211584 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 27.996551s. I0416 03:37:04.250495 132805833000512 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 03:37:04.250595 132805833000512 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 03:37:04.250650 132805833000512 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 03:37:04.250700 132805833000512 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 03:37:04.250897 132919054130304 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 03:37:04.250973 132919054130304 checkpoint_manager.py:1441] Waiting for previous save to complete took 7.451911 seconds. If this number is high, consider checkpointing less frequently. I0416 03:37:04.252143 132919054130304 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 03:37:04.252375 132919054130304 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_01_base/checkpoints/9. I0416 03:37:04.435815 132919054130304 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 03:37:04.435905 132919054130304 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 03:37:04.443014 132805751211584 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_01_base/checkpoints/9 I0416 03:37:05.203789 132805596022336 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_01_base/checkpoints/9/items I0416 03:37:13.567322 132919054130304 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.131912s I0416 03:37:13.567843 132919054130304 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.338 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0416 03:37:13.567893 132919054130304 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.223506s (batch_requests_ready=0.087967s, total_serialization_initiated=9.135133s, others=0.000406s) I0416 03:37:13.567962 132919054130304 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.224284s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=9.224265) I0416 03:37:13.573035 132805772183104 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 03:37:13.573152 132919054130304 async_checkpointer.py:561] Finished blocking save. Time taken: 9.320964s. 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_01_base/checkpoints/9. I0416 03:37:13.573347 132919054130304 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 03:37:13.573534 132805833000512 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 03:37:13.573646 132919054130304 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_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776310616.7990408, 'wait_for_prev_duration_secs': 7.451910734176636, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776310624.2521675, 'checkpointer_blocking_duration_secs': 9.321096181869507, 'get_old_steps_start_time': 1776310633.5732808, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1776310616.7989962, 'checkpoint_manager_blocking_duration_secs': 16.77462410926819} I0416 03:37:13.573768 132919054130304 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0416 03:37:13.573802 132919054130304 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 03:37:14.320261 132805801543232 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_01_base/checkpoints/9/items/array_metadatas/process_0 I0416 03:37:48.457056 132805596022336 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 34.884643s (commit=34.458855s, array_metadata_write=0.425788s) I0416 03:37:48.458166 132805772183104 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 286.459 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 44 seconds) (per-host) I0416 03:37:48.458282 132805772183104 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 34.885076s. I0416 03:37:48.837134 132805772183104 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 03:37:49.266421 132805772183104 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 03:37:49.267141 132805772183104 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.552763s. 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_01_base/checkpoints/9/items I0416 03:37:49.267759 132805772183104 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_01_base/checkpoints/9/items I0416 03:37:49.514639 132805772183104 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_01_base/checkpoints/9 I0416 03:37:50.206533 132805772183104 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_01_base/checkpoints/9`. I0416 03:37:50.207194 132805772183104 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 45.955013s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_01_base/checkpoints/9 I0416 03:37:50.207277 132805772183104 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 36.634079s. I0416 03:37:50.207458 132805833000512 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 03:37:50.207568 132805833000512 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 03:37:50.207648 132805833000512 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 03:37:50.207696 132805833000512 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 03:37:50.208072 132919054130304 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 03:37:50.208206 132919054130304 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 03:37:50.208778 132919054130304 metric_logger.py:185] completed step: 9, seconds: 0.118, TFLOP/s/device: 115.591, Tokens/s/device: 17423.115, 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