MaxView

Timing Test Reference Updates
‹ 06_grad_accumCase: 07_eval08_checkpoint_async_true_resume ›

Metrics: main (172d0f10a) vs fix/gate-fp8-nan-sanitization (3d212094d)

Metricmain  172d0f10afix/gate-fp8-nan-sanitization  3d212094dDiff (fix/gate-fp8-nan-sanitization − main)
Parameters1.104 billion1.104 billion—
Final loss8.71008.71000
TFLOP/s115.558116.005+0.447
Tok/s17418.217485.6+67.368
Avg s/step2.8472.829-0.018
Memory %5.095.090
JAX0.8.10.8.1—

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

main  ·  172d0f10a  ·  main_20260423_012248  ·  full log 
2026-04-23 01:38:23.078725: 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 00:00:1776908303.091663 2275839 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776908303.095468 2275839 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776908303.106941 2275839 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776908303.106955 2275839 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776908303.106957 2275839 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776908303.106959 2275839 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'}
2026-04-23 01:38:26.777635: 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)
I0423 01:38:26.914352 124229166630720 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag.
I0423 01:38:58.850556 124229166630720 max_utils.py:800] System Information: Jax Version: 0.8.1
I0423 01:38:58.850844 124229166630720 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0423 01:38:58.850891 124229166630720 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
I0423 01:38:58.850920 124229166630720 train_utils.py:348] WARNING: 'dataset_path' might be pointing your local file system
I0423 01:38:58.851124 124229166630720 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper.
I0423 01:38:59.511929 124229166630720 maxtext_utils.py:1604] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 01:38:59.512218 124229166630720 checkpointing.py:677] Setting up checkpoint logger...
I0423 01:38:59.512264 124229166630720 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 01:38:59.512300 124229166630720 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None
I0423 01:38:59.512723 124229166630720 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 0x70fc165b1eb0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 01:39:01.945936 124229166630720 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0423 01:39:01.946435 124229166630720 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 0x70fa44bf9be0>}, handler_registry=None
I0423 01:39:01.946771 124229166630720 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x70fa44bf9be0>` 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`.
I0423 01:39:01.946811 124229166630720 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x70fa44bfa1e0>` 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`.
I0423 01:39:01.946843 124229166630720 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 0x70fa44bf9be0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x70fa44bf9be0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x70fa44bfa1e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x70fa44bfa1e0>}).
I0423 01:39:01.947326 124229166630720 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28
I0423 01:39:01.947391 124229166630720 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x70fa3f70c540> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0423 01:39:02.075163 124229166630720 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints
I0423 01:39:02.075405 124229166630720 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_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x70fa44fd9340>
I0423 01:39:02.075693 124229166630720 checkpointing.py:301] Checkpoint manager created!
I0423 01:39:02.156064 124229166630720 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0423 01:39:02.159137 124229166630720 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.
I0423 01:39:02.226764 124229166630720 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
I0423 01:39:02.252614 124229166630720 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0423 01:39:02.252681 124229166630720 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0423 01:39:02.796333 124229166630720 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0423 01:39:02.797987 124229166630720 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.
I0423 01:39:02.812908 124229166630720 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
I0423 01:39:02.815434 124229166630720 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0423 01:39:02.815486 124229166630720 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0423 01:39:03.124824 124229166630720 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 01:39:03.124924 124229166630720 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0423 01:39:03.432300 124229166630720 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 01:39:03.432394 124229166630720 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0423 01:39:03.445219 124229166630720 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 01:39:03.445269 124229166630720 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0423 01:39:03.464416 124229166630720 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 01:39:03.464478 124229166630720 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0423 01:39:03.477397 124229166630720 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 01:39:03.477453 124229166630720 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0423 01:39:03.490242 124229166630720 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 01:39:03.490292 124229166630720 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0423 01:39:03.510465 124229166630720 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 01:39:03.510527 124229166630720 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0423 01:39:03.526497 124229166630720 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 01:39:03.526552 124229166630720 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None).
I0423 01:39:03.681525 124229166630720 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 01:39:03.681630 124229166630720 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
I0423 01:39:04.260119 124229166630720 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0423 01:39:04.260240 124229166630720 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 01:39:04.260283 124229166630720 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 01:39:04.260328 124229166630720 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   PartitionSpec('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 01:39:04.260369 124229166630720 maxtext_utils.py:1707]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0423 01:39:04.260396 124229166630720 maxtext_utils.py:1707]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0423 01:39:04.260436 124229166630720 maxtext_utils.py:1707]  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)
I0423 01:39:04.260475 124229166630720 maxtext_utils.py:1707]  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')
I0423 01:39:04.260505 124229166630720 maxtext_utils.py:1707]  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)
I0423 01:39:04.260533 124229166630720 maxtext_utils.py:1707]  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)
I0423 01:39:04.260567 124229166630720 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0423 01:39:04.260602 124229166630720 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0423 01:39:04.965895 124229166630720 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0423 01:39:04.965991 124229166630720 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None).
I0423 01:39:04.978470 124229166630720 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0423 01:39:04.978524 124229166630720 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None).
I0423 01:39:05.767811 124229166630720 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.
I0423 01:39:05.768421 124229166630720 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter.
I0423 01:39:05.768570 124229166630720 metric_logger.py:301] number parameters: 1.104 billion
I0423 01:39:10.261955 124229166630720 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0423 01:39:10.383416 124229166630720 checkpointing.py:776] Waited 0.12144136428833008 seconds for step 0 to finish before starting checkpointing.
I0423 01:39:10.384093 124229166630720 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 01:39:10.384244 124229166630720 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0
I0423 01:39:10.384627 124229166630720 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0.
I0423 01:39:10.469206 124229166630720 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient
I0423 01:39:10.563465 124098436924992 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0
I0423 01:39:10.568840 124229166630720 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0423 01:39:10.568936 124229166630720 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
I0423 01:39:11.313932 124098426439232 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0/items
I0423 01:39:11.484929 124098394981952 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776908351224506792, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776908351224506792, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0423 01:39:11.845874 2278510 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json
I0423 01:39:11.845938 2278510 google_auth_provider.cc:156] Using OAuth2 AuthProvider
I0423 01:39:12.563257 124229166630720 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.994866s
I0423 01:39:12.568409 124229166630720 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.882 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 2 seconds) (per-host)
I0423 01:39:12.568478 124229166630720 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 2.098179s (batch_requests_ready=0.094234s, total_serialization_initiated=1.998907s, others=0.005038s)
I0423 01:39:12.568553 124229166630720 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.098796s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=2.098777)
I0423 01:39:12.569982 124098447410752 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0423 01:39:12.570077 124229166630720 async_checkpointer.py:561] Finished blocking save. Time taken: 2.185793s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0.
I0423 01:39:12.570263 124229166630720 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 01:39:12.570439 124097673561664 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0423 01:39:12.570540 124229166630720 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776908350.3840804, 'wait_for_prev_duration_secs': 4.3392181396484375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776908350.3842654, 'checkpointer_blocking_duration_secs': 2.1859171390533447, 'get_old_steps_start_time': 1776908352.5702, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1776908350.3839824, 'checkpoint_manager_blocking_duration_secs': 2.1865296363830566}
I0423 01:39:12.570653 124229166630720 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0423 01:39:12.570694 124229166630720 max_utils.py:750] 
Memstats: After params initialized:
I0423 01:39:12.570758 124229166630720 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0))
I0423 01:39:12.570784 124229166630720 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0))
I0423 01:39:12.570807 124229166630720 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0))
I0423 01:39:12.570825 124229166630720 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0))
I0423 01:39:12.570842 124229166630720 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0))
I0423 01:39:12.570858 124229166630720 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0))
I0423 01:39:12.570875 124229166630720 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0))
I0423 01:39:12.570894 124229166630720 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0))
I0423 01:39:12.948951 124229166630720 metric_logger.py:196] completed step: 0, seconds: 4.493, TFLOP/s/device: 3.024, Tokens/s/device: 455.837, total_weights: 13328, loss: 10.828, lm_loss: 10.828, perplexity: 50404.086
I0423 01:39:12.949765 124229166630720 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/tensorboard/'
I0423 01:39:13.065030 124229166630720 metric_logger.py:196] completed step: 1, seconds: 2.685, TFLOP/s/device: 5.061, Tokens/s/device: 762.792, total_weights: 12332, loss: 10.828, lm_loss: 10.828, perplexity: 50391.758
I0423 01:39:13.182838 124229166630720 metric_logger.py:196] completed step: 2, seconds: 0.010, TFLOP/s/device: 1375.215, Tokens/s/device: 207287.449, total_weights: 15161, loss: 9.823, lm_loss: 9.823, perplexity: 18444.867
I0423 01:39:13.309140 124098405467712 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0/items/array_metadatas/process_0
I0423 01:39:16.076078 124229166630720 train.py:620] Completed eval step 0
I0423 01:39:16.118221 124229166630720 train.py:620] Completed eval step 1
I0423 01:39:16.123603 124229166630720 metric_logger.py:216] eval metrics after step: 4, loss=8.979, perplexity=7936.996, total_weights=26703.0
I0423 01:39:16.124174 124229166630720 metric_logger.py:196] completed step: 3, seconds: 0.119, TFLOP/s/device: 114.562, Tokens/s/device: 17267.983, total_weights: 13327, loss: 9.405, lm_loss: 9.405, perplexity: 12149.331
I0423 01:39:35.628219 124229166630720 metric_logger.py:196] completed step: 4, seconds: 0.116, TFLOP/s/device: 116.900, Tokens/s/device: 17620.387, total_weights: 11939, loss: 9.045, lm_loss: 9.045, perplexity: 8471.842
I0423 01:39:35.639559 124229166630720 metric_logger.py:196] completed step: 5, seconds: 2.942, TFLOP/s/device: 4.619, Tokens/s/device: 696.185, total_weights: 15502, loss: 8.953, lm_loss: 8.953, perplexity: 7728.546
I0423 01:39:35.756791 124229166630720 metric_logger.py:196] completed step: 6, seconds: 19.505, TFLOP/s/device: 0.697, Tokens/s/device: 104.999, total_weights: 13864, loss: 8.775, lm_loss: 8.775, perplexity: 6468.536
I0423 01:39:35.874651 124229166630720 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 1817.676, Tokens/s/device: 273979.933, total_weights: 12988, loss: 8.688, lm_loss: 8.688, perplexity: 5928.887
I0423 01:39:36.587506 124229166630720 train.py:620] Completed eval step 0
I0423 01:39:36.627360 124229166630720 train.py:620] Completed eval step 1
I0423 01:39:36.630609 124229166630720 metric_logger.py:216] eval metrics after step: 9, loss=8.653, perplexity=5725.923, total_weights=26703.0
I0423 01:39:36.631062 124229166630720 metric_logger.py:196] completed step: 8, seconds: 0.118, TFLOP/s/device: 114.835, Tokens/s/device: 17309.139, total_weights: 13820, loss: 8.740, lm_loss: 8.740, perplexity: 6250.123
I0423 01:39:36.633745 124229166630720 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0423 01:39:36.634369 124229166630720 checkpointing.py:776] Waited 0.0006263256072998047 seconds for step 9 to finish before starting checkpointing.
I0423 01:39:36.634663 124229166630720 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 01:39:59.265682 124097874888256 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 46.696591s (commit=46.210472s, array_metadata_write=0.486119s)
I0423 01:39:59.266945 124098447410752 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 258.969 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 48 seconds) (per-host)
I0423 01:39:59.267087 124098447410752 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 46.696963s.
I0423 01:39:59.525120 124098447410752 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776908351224506792, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0423 01:39:59.730056 124098447410752 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0423 01:39:59.932265 124098394981952 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776908351224506792, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0423 01:40:00.119952 124098447410752 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0423 01:40:00.120162 124098447410752 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.548407s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0/items
I0423 01:40:00.120977 124098447410752 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0/items
I0423 01:40:00.352367 124098447410752 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0
I0423 01:40:01.034518 124098447410752 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0`.
I0423 01:40:01.035313 124098447410752 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 50.651031s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/0
I0423 01:40:01.035400 124098447410752 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 48.465281s.
I0423 01:40:01.035574 124097673561664 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 01:40:01.035698 124097673561664 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0423 01:40:01.035760 124097673561664 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 01:40:01.035812 124097673561664 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 01:40:01.035961 124229166630720 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.
W0423 01:40:01.036134 124229166630720 checkpoint_manager.py:1441] Waiting for previous save to complete took 24.401482 seconds. If this number is high, consider checkpointing less frequently.
I0423 01:40:01.037429 124229166630720 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9
I0423 01:40:01.037774 124229166630720 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9.
I0423 01:40:01.201992 124229166630720 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0423 01:40:01.202108 124229166630720 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
I0423 01:40:01.204583 124098447410752 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9
I0423 01:40:01.933970 124098405467712 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9/items
I0423 01:40:04.149242 124229166630720 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 2.947744s
I0423 01:40:04.153859 124229166630720 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.058 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 3 seconds) (per-host)
I0423 01:40:04.153929 124229166630720 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 3.041417s (batch_requests_ready=0.084903s, total_serialization_initiated=2.952021s, others=0.004494s)
I0423 01:40:04.154010 124229166630720 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 3.042056s (all_items=0.000023s, per_item={'items': '0.00002313'}, temp_paths=3.042033)
I0423 01:40:04.155935 124097673561664 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0423 01:40:04.156022 124229166630720 async_checkpointer.py:561] Finished blocking save. Time taken: 3.118541s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9.
I0423 01:40:04.156249 124229166630720 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 01:40:04.156452 124097874888256 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0423 01:40:04.156589 124229166630720 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776908376.6346116, 'wait_for_prev_duration_secs': 24.401482105255127, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776908401.0374577, 'checkpointer_blocking_duration_secs': 3.1186940670013428, 'get_old_steps_start_time': 1776908404.1561666, 'get_old_steps_duration_secs': 4.172325134277344e-05, 'checkpoint_manager_blocking_start_time': 1776908376.634552, 'checkpoint_manager_blocking_duration_secs': 27.522013187408447}
I0423 01:40:04.156672 124229166630720 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0423 01:40:04.156703 124229166630720 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 01:40:04.883353 124098246084160 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9/items/array_metadatas/process_0
I0423 01:40:39.287545 124098235598400 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 35.132838s (commit=34.656495s, array_metadata_write=0.476343s)
I0423 01:40:39.288647 124097673561664 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 331.013 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 38 seconds) (per-host)
I0423 01:40:39.288704 124097673561664 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.132616s.
I0423 01:40:39.814014 124097673561664 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0423 01:40:40.194952 124097673561664 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0423 01:40:40.195231 124097673561664 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.568984s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9/items
I0423 01:40:40.196089 124097673561664 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9/items
I0423 01:40:40.437267 124097673561664 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9
I0423 01:40:41.138599 124097673561664 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9`.
I0423 01:40:41.139356 124097673561664 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 40.101881s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_07_eval/checkpoints/9
I0423 01:40:41.139429 124097673561664 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 36.983342s.
I0423 01:40:41.139599 124097874888256 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 01:40:41.139719 124097874888256 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0423 01:40:41.139785 124097874888256 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 01:40:41.139830 124097874888256 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 01:40:41.139959 124229166630720 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.
I0423 01:40:41.140113 124229166630720 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 01:40:41.140980 124229166630720 metric_logger.py:196] completed step: 9, seconds: 0.118, TFLOP/s/device: 115.558, Tokens/s/device: 17418.224, 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
fix/gate-fp8-nan-sanitization  ·  3d212094d  ·  fix_gate_fp8_nan_sanitization_20260423_024158  ·  full log 
2026-04-23 03:00:58.060663: 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 00:00:1776913258.074238 2353184 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776913258.078232 2353184 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776913258.090633 2353184 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776913258.090647 2353184 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776913258.090649 2353184 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776913258.090651 2353184 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'}
2026-04-23 03:01:01.750205: 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)
I0423 03:01:01.885316 130111404267328 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag.
~/maxtext/maxtext_venv/lib/python3.12/site-packages/jax/_src/xla_bridge.py:204: UserWarning: TPU backend initialization is taking more than 60.0 seconds. Did you run your code on all TPU hosts? See https://docs.jax.dev/en/latest/multi_process.html for more information.
  warnings.warn(
I0423 03:02:07.398036 130111404267328 max_utils.py:800] System Information: Jax Version: 0.8.1
I0423 03:02:07.398289 130111404267328 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0423 03:02:07.398339 130111404267328 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
I0423 03:02:07.398369 130111404267328 train_utils.py:348] WARNING: 'dataset_path' might be pointing your local file system
I0423 03:02:07.398537 130111404267328 train.py:672] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper.
I0423 03:02:08.058944 130111404267328 maxtext_utils.py:1604] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 03:02:08.059216 130111404267328 checkpointing.py:677] Setting up checkpoint logger...
I0423 03:02:08.059262 130111404267328 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 03:02:08.059298 130111404267328 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None
I0423 03:02:08.059709 130111404267328 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 0x7655a27c0dd0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 03:02:10.480212 130111404267328 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0423 03:02:10.480566 130111404267328 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 0x76544514af00>}, handler_registry=None
I0423 03:02:10.480794 130111404267328 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x76544514af00>` 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`.
I0423 03:02:10.480835 130111404267328 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7653d0f3d9a0>` 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`.
I0423 03:02:10.480865 130111404267328 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 0x76544514af00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x76544514af00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7653d0f3d9a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7653d0f3d9a0>}).
I0423 03:02:10.481152 130111404267328 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28
I0423 03:02:10.481200 130111404267328 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7653d0f5c540> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0423 03:02:10.606956 130111404267328 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints
I0423 03:02:10.607196 130111404267328 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_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7653d0f3e210>
I0423 03:02:10.607469 130111404267328 checkpointing.py:301] Checkpoint manager created!
I0423 03:02:10.684584 130111404267328 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0423 03:02:10.687353 130111404267328 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.
I0423 03:02:10.746462 130111404267328 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
I0423 03:02:10.769121 130111404267328 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0423 03:02:10.769183 130111404267328 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0423 03:02:11.311237 130111404267328 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0423 03:02:11.312768 130111404267328 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.
I0423 03:02:11.327715 130111404267328 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
I0423 03:02:11.330220 130111404267328 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0423 03:02:11.330269 130111404267328 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0423 03:02:11.639095 130111404267328 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 03:02:11.639186 130111404267328 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0423 03:02:11.938846 130111404267328 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 03:02:11.938924 130111404267328 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0423 03:02:11.951611 130111404267328 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 03:02:11.951658 130111404267328 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0423 03:02:11.970729 130111404267328 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 03:02:11.970785 130111404267328 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0423 03:02:11.983589 130111404267328 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 03:02:11.983643 130111404267328 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0423 03:02:11.996317 130111404267328 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 03:02:11.996364 130111404267328 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0423 03:02:12.016350 130111404267328 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 03:02:12.016408 130111404267328 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0423 03:02:12.032288 130111404267328 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 03:02:12.032343 130111404267328 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None).
I0423 03:02:12.186434 130111404267328 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 03:02:12.186529 130111404267328 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
I0423 03:02:12.761853 130111404267328 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0423 03:02:12.761970 130111404267328 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 03:02:12.762014 130111404267328 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 03:02:12.762077 130111404267328 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   PartitionSpec('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 03:02:12.762124 130111404267328 maxtext_utils.py:1707]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0423 03:02:12.762154 130111404267328 maxtext_utils.py:1707]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0423 03:02:12.762193 130111404267328 maxtext_utils.py:1707]  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)
I0423 03:02:12.762233 130111404267328 maxtext_utils.py:1707]  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')
I0423 03:02:12.762262 130111404267328 maxtext_utils.py:1707]  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)
I0423 03:02:12.762288 130111404267328 maxtext_utils.py:1707]  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)
I0423 03:02:12.762323 130111404267328 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0423 03:02:12.762356 130111404267328 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0423 03:02:13.441434 130111404267328 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0423 03:02:13.441516 130111404267328 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None).
I0423 03:02:13.453581 130111404267328 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0423 03:02:13.453636 130111404267328 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None).
I0423 03:02:14.214479 130111404267328 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.
I0423 03:02:14.215000 130111404267328 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter.
I0423 03:02:14.215166 130111404267328 metric_logger.py:301] number parameters: 1.104 billion
I0423 03:02:18.200451 130111404267328 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0423 03:02:18.321992 130111404267328 checkpointing.py:776] Waited 0.12152314186096191 seconds for step 0 to finish before starting checkpointing.
I0423 03:02:18.322613 130111404267328 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 03:02:18.322763 130111404267328 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0
I0423 03:02:18.323142 130111404267328 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0.
I0423 03:02:18.420466 130111404267328 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient
I0423 03:02:18.504361 129980941993536 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0
I0423 03:02:18.517459 130111404267328 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0423 03:02:18.517547 130111404267328 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
I0423 03:02:19.258768 129980931507776 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0/items
I0423 03:02:19.419449 129980900050496 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776913339155659179, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776913339155659179, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0423 03:02:19.783698 2355778 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json
I0423 03:02:19.783760 2355778 google_auth_provider.cc:156] Using OAuth2 AuthProvider
I0423 03:02:19.893012 130111404267328 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.375934s
I0423 03:02:19.897825 130111404267328 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 8.360 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host)
I0423 03:02:19.897890 130111404267328 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.476225s (batch_requests_ready=0.092031s, total_serialization_initiated=1.379495s, others=0.004699s)
I0423 03:02:19.897969 130111404267328 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.477062s (all_items=0.000020s, per_item={'items': '0.00002027'}, temp_paths=1.477042)
I0423 03:02:19.899172 129981090891328 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0423 03:02:19.899256 130111404267328 async_checkpointer.py:561] Finished blocking save. Time taken: 1.576453s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0.
I0423 03:02:19.899469 130111404267328 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 03:02:19.899682 129980291876416 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0423 03:02:19.899785 130111404267328 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776913338.3226001, 'wait_for_prev_duration_secs': 4.315376281738281e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776913338.3227851, 'checkpointer_blocking_duration_secs': 1.5765843391418457, 'get_old_steps_start_time': 1776913339.8993974, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776913338.3225195, 'checkpoint_manager_blocking_duration_secs': 1.5772395133972168}
I0423 03:02:19.899922 130111404267328 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0423 03:02:19.899967 130111404267328 max_utils.py:750] 
Memstats: After params initialized:
I0423 03:02:19.900018 130111404267328 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0))
I0423 03:02:19.900059 130111404267328 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0))
I0423 03:02:19.900080 130111404267328 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0))
I0423 03:02:19.900101 130111404267328 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0))
I0423 03:02:19.900119 130111404267328 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0))
I0423 03:02:19.900137 130111404267328 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0))
I0423 03:02:19.900155 130111404267328 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0))
I0423 03:02:19.900171 130111404267328 max_utils.py:756] 	Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0))
I0423 03:02:20.268949 130111404267328 metric_logger.py:196] completed step: 0, seconds: 3.985, TFLOP/s/device: 3.410, Tokens/s/device: 513.957, total_weights: 13328, loss: 10.828, lm_loss: 10.828, perplexity: 50404.086
I0423 03:02:20.269859 130111404267328 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/tensorboard/'
I0423 03:02:20.385098 130111404267328 metric_logger.py:196] completed step: 1, seconds: 2.066, TFLOP/s/device: 6.576, Tokens/s/device: 991.178, total_weights: 12332, loss: 10.828, lm_loss: 10.828, perplexity: 50391.758
I0423 03:02:20.502637 130111404267328 metric_logger.py:196] completed step: 2, seconds: 0.010, TFLOP/s/device: 1361.572, Tokens/s/device: 205230.985, total_weights: 15161, loss: 9.823, lm_loss: 9.823, perplexity: 18444.867
I0423 03:02:20.773231 129980436579904 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0/items/array_metadatas/process_0
I0423 03:02:23.901276 130111404267328 train.py:589] Completed eval step 0
I0423 03:02:23.943542 130111404267328 train.py:589] Completed eval step 1
I0423 03:02:23.948479 130111404267328 metric_logger.py:216] eval metrics after step: 4, loss=8.979, perplexity=7936.996, total_weights=26703.0
I0423 03:02:23.949029 130111404267328 metric_logger.py:196] completed step: 3, seconds: 0.117, TFLOP/s/device: 115.920, Tokens/s/device: 17472.763, total_weights: 13327, loss: 9.405, lm_loss: 9.405, perplexity: 12149.331
I0423 03:02:43.400371 130111404267328 metric_logger.py:196] completed step: 4, seconds: 0.118, TFLOP/s/device: 114.672, Tokens/s/device: 17284.597, total_weights: 11939, loss: 9.045, lm_loss: 9.045, perplexity: 8471.842
I0423 03:02:43.418684 130111404267328 metric_logger.py:196] completed step: 5, seconds: 3.446, TFLOP/s/device: 3.943, Tokens/s/device: 594.304, total_weights: 15502, loss: 8.953, lm_loss: 8.953, perplexity: 7728.546
I0423 03:02:43.534975 130111404267328 metric_logger.py:196] completed step: 6, seconds: 19.458, TFLOP/s/device: 0.698, Tokens/s/device: 105.254, total_weights: 13864, loss: 8.775, lm_loss: 8.775, perplexity: 6468.536
I0423 03:02:43.652716 130111404267328 metric_logger.py:196] completed step: 7, seconds: 0.010, TFLOP/s/device: 1424.825, Tokens/s/device: 214765.101, total_weights: 12988, loss: 8.688, lm_loss: 8.688, perplexity: 5928.887
I0423 03:02:44.361766 130111404267328 train.py:589] Completed eval step 0
I0423 03:02:44.401183 130111404267328 train.py:589] Completed eval step 1
I0423 03:02:44.404327 130111404267328 metric_logger.py:216] eval metrics after step: 9, loss=8.653, perplexity=5725.923, total_weights=26703.0
I0423 03:02:44.404771 130111404267328 metric_logger.py:196] completed step: 8, seconds: 0.117, TFLOP/s/device: 116.345, Tokens/s/device: 17536.799, total_weights: 13820, loss: 8.740, lm_loss: 8.740, perplexity: 6250.123
I0423 03:02:44.407283 130111404267328 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0423 03:02:44.407888 130111404267328 checkpointing.py:776] Waited 0.000606536865234375 seconds for step 9 to finish before starting checkpointing.
I0423 03:02:44.408200 130111404267328 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 03:03:00.571747 129980426094144 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 40.673132s (commit=40.169070s, array_metadata_write=0.504062s)
I0423 03:03:00.572725 129981090891328 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 299.798 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 42 seconds) (per-host)
I0423 03:03:00.572788 129981090891328 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.673485s.
I0423 03:03:00.853941 129981090891328 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776913339155659179, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0423 03:03:01.067203 129981090891328 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0423 03:03:01.305716 129980900050496 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776913339155659179, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0423 03:03:01.490377 129981090891328 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0423 03:03:01.490547 129981090891328 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.586846s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0/items
I0423 03:03:01.491387 129981090891328 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0/items
I0423 03:03:01.731660 129981090891328 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0
I0423 03:03:02.515763 129981090891328 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0`.
I0423 03:03:02.516397 129981090891328 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 44.193597s. directory=gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/0
I0423 03:03:02.516483 129981090891328 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 42.617184s.
I0423 03:03:02.516643 129980291876416 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 03:03:02.516752 129980291876416 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0423 03:03:02.516823 129980291876416 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 03:03:02.516883 129980291876416 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 03:03:02.517005 130111404267328 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.
W0423 03:03:02.517134 130111404267328 checkpoint_manager.py:1441] Waiting for previous save to complete took 18.108960 seconds. If this number is high, consider checkpointing less frequently.
I0423 03:03:02.518811 130111404267328 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9
I0423 03:03:02.519143 130111404267328 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9.
I0423 03:03:02.702732 129981090891328 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9
I0423 03:03:02.717307 130111404267328 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0423 03:03:02.717405 130111404267328 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
I0423 03:03:03.473541 129980436579904 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9/items
I0423 03:03:05.635302 130111404267328 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 2.918426s
I0423 03:03:05.635825 130111404267328 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.080 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 3 seconds) (per-host)
I0423 03:03:05.635885 130111404267328 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 3.024715s (batch_requests_ready=0.102739s, total_serialization_initiated=2.921556s, others=0.000421s)
I0423 03:03:05.635957 130111404267328 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 3.025247s (all_items=0.000026s, per_item={'items': '0.00002599'}, temp_paths=3.025221)
I0423 03:03:05.641724 129980291876416 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0423 03:03:05.641828 130111404267328 async_checkpointer.py:561] Finished blocking save. Time taken: 3.122959s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9.
I0423 03:03:05.642080 130111404267328 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 03:03:05.642352 129963984422464 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0423 03:03:05.642469 130111404267328 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776913364.408148, 'wait_for_prev_duration_secs': 18.108959913253784, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776913382.51884, 'checkpointer_blocking_duration_secs': 3.1231253147125244, 'get_old_steps_start_time': 1776913385.6419804, 'get_old_steps_duration_secs': 4.315376281738281e-05, 'checkpoint_manager_blocking_start_time': 1776913364.408082, 'checkpoint_manager_blocking_duration_secs': 21.234359979629517}
I0423 03:03:05.642599 130111404267328 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0423 03:03:05.642635 130111404267328 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 03:03:06.409457 129980807775808 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9/items/array_metadatas/process_0
I0423 03:03:47.107976 129981090891328 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.467252s (commit=40.905947s, array_metadata_write=0.561305s)
I0423 03:03:47.109009 129980291876416 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 283.987 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 44 seconds) (per-host)
I0423 03:03:47.109125 129980291876416 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.467238s.
I0423 03:03:47.637553 129980291876416 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0423 03:03:48.056031 129980291876416 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0423 03:03:48.056274 129980291876416 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.616476s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9/items
I0423 03:03:48.057037 129980291876416 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9/items
I0423 03:03:48.317542 129980291876416 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9
I0423 03:03:49.051246 129980291876416 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9`.
I0423 03:03:49.051870 129980291876416 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 46.533015s. directory=gs://wanglance-maxtext/linen_ckpt_fix_gate_fp8_nan_sanitization_20260423_024158/linen_fix_gate_fp8_nan_sanitization_20260423_024158_07_eval/checkpoints/9
I0423 03:03:49.051948 129980291876416 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 43.410065s.
I0423 03:03:49.052143 129963984422464 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 03:03:49.052242 129963984422464 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0423 03:03:49.052294 129963984422464 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 03:03:49.052333 129963984422464 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 03:03:49.052474 130111404267328 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.
I0423 03:03:49.052609 130111404267328 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 03:03:49.053428 130111404267328 metric_logger.py:196] completed step: 9, seconds: 0.117, TFLOP/s/device: 116.005, Tokens/s/device: 17485.592, 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