MaxView

Timing Test Reference Updates
‹ 06_grad_accumCase: 07_eval08_checkpoint_async_true_resume ›

Metrics: main (b341212ce) vs feat/nnx-trainstate-and-training-loop (29bbe47ed)

Metricmain  b341212cefeat/nnx-trainstate-and-training-loop  29bbe47edDiff (feat/nnx-trainstate-and-training-loop − main)
Parameters1.104 billion1.104 billion—
Final loss8.71008.71000
TFLOP/s115.180115.322+0.142
Tok/s17361.217382.6+21.366
Avg s/step3.0592.449-0.61
Memory %5.095.090
JAX0.8.10.8.1—

Diff = branch value − main value. Green = branch improved. Red = branch regressed.

main  ·  b341212ce  ·  main_20260421_144657  ·  full log 
2026-04-21 15:05:52.016311: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1776783952.026598 1493670 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776783952.029580 1493670 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776783952.039321 1493670 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776783952.039331 1493670 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776783952.039333 1493670 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776783952.039334 1493670 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-21 15:05:55.453527: 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)
I0421 15:05:55.571375 131315320347776 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag.
I0421 15:06:22.480978 131315320347776 max_utils.py:800] System Information: Jax Version: 0.8.1
I0421 15:06:22.481242 131315320347776 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0421 15:06:22.481289 131315320347776 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709
I0421 15:06:22.481319 131315320347776 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system
I0421 15:06:22.481446 131315320347776 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper.
I0421 15:06:22.837010 131315320347776 maxtext_utils.py:1551] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 15:06:22.837411 131315320347776 checkpointing.py:677] Setting up checkpoint logger...
I0421 15:06:22.837460 131315320347776 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0421 15:06:22.837547 131315320347776 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None
I0421 15:06:22.838434 131315320347776 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 0x776deeba4650>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0421 15:06:25.266310 131315320347776 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0421 15:06:25.266610 131315320347776 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 0x776c493d6870>}, handler_registry=None
I0421 15:06:25.267040 131315320347776 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x776c493d6870>` 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`.
I0421 15:06:25.267103 131315320347776 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x776c48381f10>` 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`.
I0421 15:06:25.267133 131315320347776 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 0x776c493d6870>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x776c493d6870>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x776c48381f10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x776c48381f10>}).
I0421 15:06:25.267461 131315320347776 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28
I0421 15:06:25.267523 131315320347776 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x776c483a4ea0> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0421 15:06:25.397320 131315320347776 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints
I0421 15:06:25.397606 131315320347776 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_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x776c4873b710>
I0421 15:06:25.397701 131315320347776 checkpointing.py:301] Checkpoint manager created!
I0421 15:06:25.746380 131315320347776 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0421 15:06:25.749346 131315320347776 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.
I0421 15:06:25.802343 131315320347776 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
I0421 15:06:25.824531 131315320347776 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 15:06:25.824594 131315320347776 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 15:06:26.353594 131315320347776 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0421 15:06:26.355216 131315320347776 reader.py:262] Creating a tf.data.Dataset reading 2 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0.
I0421 15:06:26.371138 131315320347776 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split validation, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0421 15:06:26.373711 131315320347776 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 15:06:26.373760 131315320347776 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 15:06:26.707358 131315320347776 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0421 15:06:26.707451 131315320347776 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 15:06:27.008454 131315320347776 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 15:06:27.008544 131315320347776 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 15:06:27.021054 131315320347776 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 15:06:27.021106 131315320347776 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 15:06:27.053697 131315320347776 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 15:06:27.053778 131315320347776 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 15:06:27.066441 131315320347776 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 15:06:27.066496 131315320347776 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 15:06:27.078966 131315320347776 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 15:06:27.079014 131315320347776 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 15:06:27.098854 131315320347776 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0421 15:06:27.098916 131315320347776 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 15:06:27.114620 131315320347776 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0421 15:06:27.114674 131315320347776 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None).
I0421 15:06:27.268534 131315320347776 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0421 15:06:27.268633 131315320347776 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
I0421 15:06:27.883403 131315320347776 maxtext_utils.py:1654]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0421 15:06:27.883517 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 15:06:27.883559 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 15:06:27.883605 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   PartitionSpec('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0421 15:06:27.883646 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0421 15:06:27.883677 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0421 15:06:27.883719 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 15:06:27.883767 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   PartitionSpec('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0421 15:06:27.883800 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0421 15:06:27.883830 131315320347776 maxtext_utils.py:1654]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 15:06:27.883868 131315320347776 maxtext_utils.py:1654]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0421 15:06:27.883905 131315320347776 maxtext_utils.py:1654]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0421 15:06:28.294680 131315320347776 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0421 15:06:28.294775 131315320347776 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None).
I0421 15:06:28.306895 131315320347776 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0421 15:06:28.306947 131315320347776 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None).
I0421 15:06:29.308348 131315320347776 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.
I0421 15:06:29.308892 131315320347776 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter.
I0421 15:06:29.309037 131315320347776 metric_logger.py:301] number parameters: 1.104 billion
I0421 15:06:33.221704 131315320347776 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0421 15:06:33.343138 131315320347776 checkpointing.py:776] Waited 0.12141656875610352 seconds for step 0 to finish before starting checkpointing.
I0421 15:06:33.343608 131315320347776 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 15:06:33.343758 131315320347776 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0
I0421 15:06:33.344180 131315320347776 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0.
I0421 15:06:33.424947 131315320347776 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient
I0421 15:06:33.504574 131222781036096 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0
I0421 15:06:33.521018 131315320347776 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0421 15:06:33.521131 131315320347776 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
I0421 15:06:34.187656 131315320347776 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.667113s
I0421 15:06:34.188330 131315320347776 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 16.190 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 762 milliseconds) (per-host)
I0421 15:06:34.188388 131315320347776 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.762356s (batch_requests_ready=0.091009s, total_serialization_initiated=0.670806s, others=0.000541s)
I0421 15:06:34.188458 131315320347776 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.762948s (all_items=0.000019s, per_item={'items': '0.00001860'}, temp_paths=0.762929)
I0421 15:06:34.189380 131222709732928 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0421 15:06:34.189517 131315320347776 async_checkpointer.py:561] Finished blocking save. Time taken: 0.845715s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0.
I0421 15:06:34.189760 131315320347776 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 15:06:34.189969 131315320347776 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776783993.3435946, 'wait_for_prev_duration_secs': 4.100799560546875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776783993.3437781, 'checkpointer_blocking_duration_secs': 0.8458857536315918, 'get_old_steps_start_time': 1776783994.1896873, 'get_old_steps_duration_secs': 2.6941299438476562e-05, 'checkpoint_manager_blocking_start_time': 1776783993.343508, 'checkpoint_manager_blocking_duration_secs': 0.8464305400848389}
I0421 15:06:34.190065 131315320347776 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0421 15:06:34.190113 131315320347776 max_utils.py:750] 
Memstats: After params initialized:
I0421 15:06:34.190186 131315320347776 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0))
I0421 15:06:34.190213 131315320347776 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0))
I0421 15:06:34.190234 131315320347776 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0))
I0421 15:06:34.190253 131315320347776 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0))
I0421 15:06:34.190270 131315320347776 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0))
I0421 15:06:34.190287 131315320347776 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0))
I0421 15:06:34.190303 131315320347776 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0))
I0421 15:06:34.190320 131315320347776 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0))
I0421 15:06:34.252394 131222770550336 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0/items
I0421 15:06:34.404864 131222730704448 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783994156629522, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776783994156629522, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0421 15:06:34.405198 131222791521856 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0421 15:06:34.618212 131315320347776 metric_logger.py:196] completed step: 0, seconds: 3.912, TFLOP/s/device: 3.473, Tokens/s/device: 523.501, total_weights: 13328, loss: 10.828, lm_loss: 10.828, perplexity: 50404.086
I0421 15:06:34.619251 131315320347776 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/tensorboard/'
I0421 15:06:34.732764 131315320347776 metric_logger.py:196] completed step: 1, seconds: 1.393, TFLOP/s/device: 9.754, Tokens/s/device: 1470.193, total_weights: 12332, loss: 10.828, lm_loss: 10.828, perplexity: 50391.758
I0421 15:06:34.787795 1496298 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json
I0421 15:06:34.787887 1496298 google_auth_provider.cc:156] Using OAuth2 AuthProvider
I0421 15:06:34.850549 131315320347776 metric_logger.py:196] completed step: 2, seconds: 0.012, TFLOP/s/device: 1160.103, Tokens/s/device: 174863.388, total_weights: 15161, loss: 9.823, lm_loss: 9.823, perplexity: 18444.867
I0421 15:06:36.020914 131222781036096 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0/items/array_metadatas/process_0
I0421 15:06:42.649302 131315320347776 train.py:620] Completed eval step 0
I0421 15:06:42.691614 131315320347776 train.py:620] Completed eval step 1
I0421 15:06:42.696554 131315320347776 metric_logger.py:216] eval metrics after step: 4, loss=8.979, perplexity=7936.996, total_weights=26703.0
I0421 15:06:42.697111 131315320347776 metric_logger.py:196] completed step: 3, seconds: 0.116, TFLOP/s/device: 116.649, Tokens/s/device: 17582.569, total_weights: 13327, loss: 9.405, lm_loss: 9.405, perplexity: 12149.331
I0421 15:07:00.493808 131315320347776 metric_logger.py:196] completed step: 4, seconds: 0.118, TFLOP/s/device: 115.548, Tokens/s/device: 17416.595, total_weights: 11939, loss: 9.045, lm_loss: 9.045, perplexity: 8471.842
I0421 15:07:00.511221 131315320347776 metric_logger.py:196] completed step: 5, seconds: 7.845, TFLOP/s/device: 1.732, Tokens/s/device: 261.058, total_weights: 15502, loss: 8.953, lm_loss: 8.953, perplexity: 7728.546
I0421 15:07:00.625275 131315320347776 metric_logger.py:196] completed step: 6, seconds: 17.801, TFLOP/s/device: 0.763, Tokens/s/device: 115.047, total_weights: 13864, loss: 8.775, lm_loss: 8.775, perplexity: 6468.536
I0421 15:07:00.743087 131315320347776 metric_logger.py:196] completed step: 7, seconds: 0.013, TFLOP/s/device: 1012.378, Tokens/s/device: 152596.677, total_weights: 12988, loss: 8.688, lm_loss: 8.688, perplexity: 5928.887
I0421 15:07:01.441814 131315320347776 train.py:620] Completed eval step 0
I0421 15:07:01.481650 131315320347776 train.py:620] Completed eval step 1
I0421 15:07:01.484435 131315320347776 metric_logger.py:216] eval metrics after step: 9, loss=8.653, perplexity=5725.923, total_weights=26703.0
I0421 15:07:01.484850 131315320347776 metric_logger.py:196] completed step: 8, seconds: 0.111, TFLOP/s/device: 122.414, Tokens/s/device: 18451.614, total_weights: 13820, loss: 8.740, lm_loss: 8.740, perplexity: 6250.123
I0421 15:07:01.486784 131315320347776 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0421 15:07:01.487377 131315320347776 checkpointing.py:776] Waited 0.0005931854248046875 seconds for step 9 to finish before starting checkpointing.
I0421 15:07:01.487618 131315320347776 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 15:07:18.935664 131222760064576 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 44.154273s (commit=43.569222s, array_metadata_write=0.585051s)
I0421 15:07:18.937060 131222709732928 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 277.665 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 45 seconds) (per-host)
I0421 15:07:18.937209 131222709732928 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.747633s.
I0421 15:07:19.185860 131222709732928 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783994156629522, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0421 15:07:19.370462 131222709732928 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0421 15:07:19.570655 131222730704448 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783994156629522, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0421 15:07:19.809830 131222709732928 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0421 15:07:19.810088 131222709732928 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.586430s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0/items
I0421 15:07:19.810913 131222709732928 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0/items
I0421 15:07:20.052954 131222709732928 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0
I0421 15:07:20.763590 131222709732928 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0`.
I0421 15:07:20.764378 131222709732928 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 47.420586s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/0
I0421 15:07:20.764521 131222709732928 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 46.574948s.
I0421 15:07:20.764667 131222791521856 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 15:07:20.764720 131222791521856 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0421 15:07:20.764797 131222791521856 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0421 15:07:20.764851 131222791521856 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0421 15:07:20.765025 131315320347776 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.
W0421 15:07:20.765211 131315320347776 checkpoint_manager.py:1441] Waiting for previous save to complete took 19.277590 seconds. If this number is high, consider checkpointing less frequently.
I0421 15:07:20.766143 131315320347776 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9
I0421 15:07:20.766526 131315320347776 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9.
I0421 15:07:20.941793 131222791521856 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9
I0421 15:07:20.945756 131315320347776 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0421 15:07:20.945867 131315320347776 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
I0421 15:07:21.708382 131222770550336 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9/items
I0421 15:07:28.607589 131315320347776 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 7.662425s
I0421 15:07:28.612406 131315320347776 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.589 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 7 seconds) (per-host)
I0421 15:07:28.612478 131315320347776 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 7.764165s (batch_requests_ready=0.093682s, total_serialization_initiated=7.665818s, others=0.004664s)
I0421 15:07:28.612564 131315320347776 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 7.764817s (all_items=0.000026s, per_item={'items': '0.00002575'}, temp_paths=7.764792)
I0421 15:07:28.615355 131222709732928 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0421 15:07:28.615490 131315320347776 async_checkpointer.py:561] Finished blocking save. Time taken: 7.849272s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9.
I0421 15:07:28.615789 131315320347776 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 15:07:28.616106 131222760064576 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0421 15:07:28.616249 131315320347776 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776784021.487585, 'wait_for_prev_duration_secs': 19.277589559555054, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776784040.7661703, 'checkpointer_blocking_duration_secs': 7.849505662918091, 'get_old_steps_start_time': 1776784048.6156948, 'get_old_steps_duration_secs': 3.62396240234375e-05, 'checkpoint_manager_blocking_start_time': 1776784021.4875474, 'checkpoint_manager_blocking_duration_secs': 27.128668069839478}
I0421 15:07:28.616411 131315320347776 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0421 15:07:28.616451 131315320347776 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 15:07:29.347483 131222781036096 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9/items/array_metadatas/process_0
I0421 15:08:03.650440 131220899890752 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 35.037137s (commit=34.499063s, array_metadata_write=0.538074s)
I0421 15:08:03.651647 131222709732928 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 295.229 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 42 seconds) (per-host)
I0421 15:08:03.651710 131222709732928 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.036144s.
I0421 15:08:04.052982 131222709732928 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0421 15:08:04.485822 131222709732928 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0421 15:08:04.486030 131222709732928 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.588916s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9/items
I0421 15:08:04.486841 131222709732928 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9/items
I0421 15:08:04.714742 131222709732928 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9
I0421 15:08:05.414950 131222709732928 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9`.
I0421 15:08:05.415766 131222709732928 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 44.649578s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_07_eval/checkpoints/9
I0421 15:08:05.415856 131222709732928 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 36.800291s.
I0421 15:08:05.416106 131222760064576 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 15:08:05.416229 131222760064576 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0421 15:08:05.416300 131222760064576 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0421 15:08:05.416344 131222760064576 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0421 15:08:05.416825 131315320347776 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.
I0421 15:08:05.416974 131315320347776 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 15:08:05.417882 131315320347776 metric_logger.py:196] completed step: 9, seconds: 0.118, TFLOP/s/device: 115.180, Tokens/s/device: 17361.229, total_weights: 12300, loss: 8.710, lm_loss: 8.710, perplexity: 6065.008
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
feat/nnx-trainstate-and-training-loop  ·  29bbe47ed  ·  feat_nnx_trainstate_and_training_loop_20260421_161934  ·  full log 
2026-04-21 16:36:18.288179: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1776789378.298534 1608939 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776789378.301570 1608939 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776789378.311371 1608939 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776789378.311381 1608939 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776789378.311383 1608939 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776789378.311384 1608939 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'}
`rope_scaling`'s factor field must be a float >= 1, got 40
`rope_scaling`'s beta_fast field must be a float, got 32
`rope_scaling`'s beta_slow field must be a float, got 1
Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'}
Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'}
Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'}
2026-04-21 16:36:22.244235: 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)
I0421 16:36:22.379707 135347584011392 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag.
I0421 16:37:18.745735 135347584011392 max_utils.py:800] System Information: Jax Version: 0.8.1
I0421 16:37:18.745997 135347584011392 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0421 16:37:18.746057 135347584011392 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709
I0421 16:37:18.746086 135347584011392 train_utils.py:378] WARNING: 'dataset_path' might be pointing your local file system
I0421 16:37:18.746287 135347584011392 train.py:806] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper.
I0421 16:37:19.388098 135347584011392 maxtext_utils.py:1732] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 16:37:19.388268 135347584011392 maxtext_utils.py:1732] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 16:37:19.388424 135347584011392 checkpointing.py:688] Setting up checkpoint logger...
I0421 16:37:19.388471 135347584011392 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0421 16:37:19.388510 135347584011392 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None
I0421 16:37:19.388948 135347584011392 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 0x7b18c49a7200>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0421 16:37:21.770352 135347584011392 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0421 16:37:21.770548 135347584011392 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 0x7b176efdd070>}, handler_registry=None
I0421 16:37:21.770793 135347584011392 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b176efdd070>` 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`.
I0421 16:37:21.770837 135347584011392 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b171b98d880>` 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`.
I0421 16:37:21.770866 135347584011392 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 0x7b176efdd070>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b176efdd070>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b171b98d880>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b171b98d880>}).
I0421 16:37:21.771170 135347584011392 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28
I0421 16:37:21.771224 135347584011392 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7b171b9b4ae0> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0421 16:37:21.895629 135347584011392 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints
I0421 16:37:21.895862 135347584011392 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_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b171b98dc70>
I0421 16:37:21.895941 135347584011392 checkpointing.py:302] Checkpoint manager created!
I0421 16:37:21.972535 135347584011392 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0421 16:37:21.975385 135347584011392 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.
I0421 16:37:22.033484 135347584011392 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
I0421 16:37:22.056293 135347584011392 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 16:37:22.056356 135347584011392 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 16:37:22.599852 135347584011392 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0421 16:37:22.601411 135347584011392 reader.py:262] Creating a tf.data.Dataset reading 2 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0.
I0421 16:37:22.617791 135347584011392 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split validation, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0421 16:37:22.620410 135347584011392 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 16:37:22.620462 135347584011392 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 16:37:22.941170 135347584011392 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0421 16:37:22.941267 135347584011392 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 16:37:23.239837 135347584011392 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 16:37:23.239927 135347584011392 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 16:37:23.252493 135347584011392 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 16:37:23.252541 135347584011392 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 16:37:23.271451 135347584011392 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 16:37:23.271506 135347584011392 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 16:37:23.284118 135347584011392 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 16:37:23.284172 135347584011392 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 16:37:23.296765 135347584011392 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 16:37:23.296818 135347584011392 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 16:37:23.316714 135347584011392 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0421 16:37:23.316774 135347584011392 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 16:37:23.332590 135347584011392 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0421 16:37:23.332644 135347584011392 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None).
I0421 16:37:23.485569 135347584011392 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0421 16:37:23.485669 135347584011392 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
I0421 16:37:24.037858 135347584011392 maxtext_utils.py:1835]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0421 16:37:24.037978 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 16:37:24.038022 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 16:37:24.038085 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   PartitionSpec('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0421 16:37:24.038130 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0421 16:37:24.038160 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0421 16:37:24.038201 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 16:37:24.038243 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   PartitionSpec('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0421 16:37:24.038275 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0421 16:37:24.038304 135347584011392 maxtext_utils.py:1835]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 16:37:24.038344 135347584011392 maxtext_utils.py:1835]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0421 16:37:24.038382 135347584011392 maxtext_utils.py:1835]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0421 16:37:24.710378 135347584011392 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0421 16:37:24.710459 135347584011392 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None).
I0421 16:37:24.722981 135347584011392 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0421 16:37:24.723033 135347584011392 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None).
I0421 16:37:25.458782 135347584011392 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.
I0421 16:37:25.459320 135347584011392 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter.
I0421 16:37:25.459471 135347584011392 metric_logger.py:301] number parameters: 1.104 billion
I0421 16:37:29.403737 135347584011392 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0421 16:37:29.525249 135347584011392 checkpointing.py:798] Waited 0.12150192260742188 seconds for step 0 to finish before starting checkpointing.
I0421 16:37:29.525722 135347584011392 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 16:37:29.525880 135347584011392 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0
I0421 16:37:29.526294 135347584011392 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0.
I0421 16:37:29.607434 135347584011392 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient
I0421 16:37:29.695709 135347584011392 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0421 16:37:29.695795 135347584011392 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
I0421 16:37:29.702327 135255082141248 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0
I0421 16:37:30.460232 135347584011392 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.764913s
I0421 16:37:30.460667 135347584011392 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 14.486 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 851 milliseconds) (per-host)
I0421 16:37:30.460726 135347584011392 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.852018s (batch_requests_ready=0.082978s, total_serialization_initiated=0.768754s, others=0.000286s)
I0421 16:37:30.460798 135347584011392 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.852884s (all_items=0.000023s, per_item={'items': '0.00002289'}, temp_paths=0.852861)
I0421 16:37:30.461638 135255019226688 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0421 16:37:30.461761 135347584011392 async_checkpointer.py:561] Finished blocking save. Time taken: 0.935836s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0.
I0421 16:37:30.462009 135347584011392 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 16:37:30.462286 135347584011392 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776789449.525708, 'wait_for_prev_duration_secs': 4.6253204345703125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776789449.5259006, 'checkpointer_blocking_duration_secs': 0.9360036849975586, 'get_old_steps_start_time': 1776789450.4619296, 'get_old_steps_duration_secs': 3.457069396972656e-05, 'checkpoint_manager_blocking_start_time': 1776789449.5256166, 'checkpoint_manager_blocking_duration_secs': 0.9366393089294434}
I0421 16:37:30.462374 135347584011392 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0421 16:37:30.462422 135347584011392 max_utils.py:750] 
Memstats: After params initialized:
I0421 16:37:30.462512 135347584011392 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0))
I0421 16:37:30.462538 135347584011392 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0))
I0421 16:37:30.462558 135347584011392 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0))
I0421 16:37:30.462577 135347584011392 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0))
I0421 16:37:30.462594 135347584011392 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0))
I0421 16:37:30.462612 135347584011392 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0))
I0421 16:37:30.462629 135347584011392 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0))
I0421 16:37:30.462647 135347584011392 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0))
I0421 16:37:30.484913 135255071655488 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0/items
I0421 16:37:30.621014 135255040198208 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776789450389915197, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776789450389915197, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0421 16:37:30.621294 135255092627008 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0421 16:37:30.892756 135347584011392 metric_logger.py:196] completed step: 0, seconds: 3.944, TFLOP/s/device: 3.445, Tokens/s/device: 519.305, total_weights: 13328, loss: 10.828, lm_loss: 10.828, perplexity: 50404.086
I0421 16:37:30.893772 135347584011392 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/tensorboard/'
I0421 16:37:31.008757 135347584011392 metric_logger.py:196] completed step: 1, seconds: 1.487, TFLOP/s/device: 9.138, Tokens/s/device: 1377.403, total_weights: 12332, loss: 10.828, lm_loss: 10.828, perplexity: 50391.758
I0421 16:37:31.021780 1611508 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json
I0421 16:37:31.021862 1611508 google_auth_provider.cc:156] Using OAuth2 AuthProvider
I0421 16:37:31.126513 135347584011392 metric_logger.py:196] completed step: 2, seconds: 0.010, TFLOP/s/device: 1390.557, Tokens/s/device: 209599.836, total_weights: 15161, loss: 9.823, lm_loss: 9.823, perplexity: 18444.867
I0421 16:37:32.038318 135255061169728 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_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0/items/array_metadatas/process_0
I0421 16:37:34.488804 135347584011392 train.py:723] Completed eval step 0
I0421 16:37:34.530539 135347584011392 train.py:723] Completed eval step 1
I0421 16:37:34.535788 135347584011392 metric_logger.py:216] eval metrics after step: 4, loss=8.979, perplexity=7936.996, total_weights=26703.0
I0421 16:37:34.536318 135347584011392 metric_logger.py:196] completed step: 3, seconds: 0.127, TFLOP/s/device: 106.799, Tokens/s/device: 16097.845, total_weights: 13327, loss: 9.405, lm_loss: 9.405, perplexity: 12149.331
I0421 16:37:51.191070 135347584011392 metric_logger.py:196] completed step: 4, seconds: 0.108, TFLOP/s/device: 126.270, Tokens/s/device: 19032.750, total_weights: 11939, loss: 9.045, lm_loss: 9.045, perplexity: 8471.842
I0421 16:37:51.202180 135347584011392 metric_logger.py:196] completed step: 5, seconds: 3.409, TFLOP/s/device: 3.986, Tokens/s/device: 600.749, total_weights: 15502, loss: 8.953, lm_loss: 8.953, perplexity: 7728.546
I0421 16:37:51.319123 135347584011392 metric_logger.py:196] completed step: 6, seconds: 16.656, TFLOP/s/device: 0.816, Tokens/s/device: 122.956, total_weights: 13864, loss: 8.775, lm_loss: 8.775, perplexity: 6468.536
I0421 16:37:51.436712 135347584011392 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 2058.344, Tokens/s/device: 310256.022, total_weights: 12988, loss: 8.688, lm_loss: 8.688, perplexity: 5928.887
I0421 16:37:52.108635 135347584011392 train.py:723] Completed eval step 0
I0421 16:37:52.149190 135347584011392 train.py:723] Completed eval step 1
I0421 16:37:52.151926 135347584011392 metric_logger.py:216] eval metrics after step: 9, loss=8.653, perplexity=5725.923, total_weights=26703.0
I0421 16:37:52.152322 135347584011392 metric_logger.py:196] completed step: 8, seconds: 0.118, TFLOP/s/device: 115.138, Tokens/s/device: 17354.903, total_weights: 13820, loss: 8.740, lm_loss: 8.740, perplexity: 6250.123
I0421 16:37:52.154173 135347584011392 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0421 16:37:52.154684 135347584011392 checkpointing.py:798] Waited 0.0005147457122802734 seconds for step 9 to finish before starting checkpointing.
I0421 16:37:52.154992 135347584011392 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 16:38:07.697259 135255082141248 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 36.673669s (commit=36.183261s, array_metadata_write=0.490408s)
I0421 16:38:07.698337 135255019226688 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 331.765 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 38 seconds) (per-host)
I0421 16:38:07.698408 135255019226688 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.236584s.
I0421 16:38:07.941687 135255019226688 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776789450389915197, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0421 16:38:08.133136 135255019226688 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0421 16:38:08.332057 135255040198208 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776789450389915197, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0421 16:38:08.540414 135255019226688 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0421 16:38:08.541100 135255019226688 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.558321s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0/items
I0421 16:38:08.541798 135255019226688 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0/items
I0421 16:38:08.785357 135255019226688 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0
I0421 16:38:09.466142 135255019226688 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_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0`.
I0421 16:38:09.466817 135255019226688 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 39.940899s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/0
I0421 16:38:09.466910 135255019226688 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 39.005091s.
I0421 16:38:09.467122 135255092627008 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 16:38:09.467240 135255092627008 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0421 16:38:09.467307 135255092627008 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0421 16:38:09.467370 135255092627008 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0421 16:38:09.467535 135347584011392 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.
W0421 16:38:09.467669 135347584011392 checkpoint_manager.py:1441] Waiting for previous save to complete took 17.312702 seconds. If this number is high, consider checkpointing less frequently.
I0421 16:38:09.468930 135347584011392 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9
I0421 16:38:09.469285 135347584011392 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9.
I0421 16:38:09.637300 135347584011392 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0421 16:38:09.637401 135347584011392 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
I0421 16:38:09.647562 135255092627008 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9
I0421 16:38:10.363227 135255082141248 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9/items
I0421 16:38:24.645347 135347584011392 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 15.008488s
I0421 16:38:24.650009 135347584011392 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 836.837 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 15 seconds) (per-host)
I0421 16:38:24.650089 135347584011392 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 15.100760s (batch_requests_ready=0.084701s, total_serialization_initiated=15.011495s, others=0.004564s)
I0421 16:38:24.650169 135347584011392 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 15.101309s (all_items=0.000026s, per_item={'items': '0.00002551'}, temp_paths=15.101284)
I0421 16:38:24.651621 135255092627008 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0421 16:38:24.651713 135347584011392 async_checkpointer.py:561] Finished blocking save. Time taken: 15.182732s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9.
I0421 16:38:24.651933 135347584011392 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 16:38:24.652126 135255019226688 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0421 16:38:24.652229 135347584011392 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776789472.1549335, 'wait_for_prev_duration_secs': 17.312702178955078, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776789489.468957, 'checkpointer_blocking_duration_secs': 15.182868003845215, 'get_old_steps_start_time': 1776789504.6518419, 'get_old_steps_duration_secs': 4.839897155761719e-05, 'checkpoint_manager_blocking_start_time': 1776789472.1548533, 'checkpoint_manager_blocking_duration_secs': 32.497349977493286}
I0421 16:38:24.652352 135347584011392 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0421 16:38:24.652386 135347584011392 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 16:38:25.416421 135255061169728 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_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9/items/array_metadatas/process_0
I0421 16:39:06.925241 135255082141248 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 42.274524s (commit=41.786199s, array_metadata_write=0.488325s)
I0421 16:39:06.926412 135255092627008 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 220.241 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 57 seconds) (per-host)
I0421 16:39:06.926483 135255092627008 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.274722s.
I0421 16:39:07.387413 135255092627008 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0421 16:39:07.825585 135255092627008 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0421 16:39:07.826309 135255092627008 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.588284s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9/items
I0421 16:39:07.827000 135255092627008 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9/items
I0421 16:39:08.090519 135255092627008 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9
I0421 16:39:08.776619 135255092627008 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_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9`.
I0421 16:39:08.777299 135255092627008 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 59.308324s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/linen_feat_nnx_trainstate_and_training_loop_20260421_161934_07_eval/checkpoints/9
I0421 16:39:08.777389 135255092627008 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 44.125630s.
I0421 16:39:08.777616 135255019226688 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 16:39:08.777728 135255019226688 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0421 16:39:08.777807 135255019226688 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0421 16:39:08.777856 135255019226688 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0421 16:39:08.778773 135347584011392 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.
I0421 16:39:08.779650 135347584011392 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 16:39:08.780727 135347584011392 metric_logger.py:196] completed step: 9, seconds: 0.118, TFLOP/s/device: 115.322, Tokens/s/device: 17382.595, total_weights: 12300, loss: 8.710, lm_loss: 8.710, perplexity: 6065.008
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops