MaxView

Timing Test Reference Updates
‹ 03_dropoutCase: 04_int805_fp8 ›

Metrics: main (b341212ce) vs fix/gate-fp8-nan-sanitization (cd5a4f898)

Metricmain  b341212cefix/gate-fp8-nan-sanitization  cd5a4f898Diff (fix/gate-fp8-nan-sanitization − main)
Parameters1.104 billion1.104 billion—
Final loss8.77408.77400
TFLOP/s109.279111.591+2.312
Tok/s16471.816820.2+348.488
Avg s/step2.4073.460+1.053
Memory %5.125.120
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 14:56:26.145835: 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 00:00:1776783386.155923 1482694 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776783386.159012 1482694 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776783386.168929 1482694 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776783386.168941 1482694 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776783386.168943 1482694 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776783386.168944 1482694 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-21 14:56:29.586228: 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 14:56:29.702314 139847166221440 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag.
I0421 14:56:55.670389 139847166221440 max_utils.py:800] System Information: Jax Version: 0.8.1
I0421 14:56:55.670576 139847166221440 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0421 14:56:55.670611 139847166221440 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 14:56:55.670643 139847166221440 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system
I0421 14:56:55.670758 139847166221440 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper.
I0421 14:56:56.024010 139847166221440 maxtext_utils.py:1551] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 14:56:56.024461 139847166221440 checkpointing.py:677] Setting up checkpoint logger...
I0421 14:56:56.024510 139847166221440 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0421 14:56:56.024558 139847166221440 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None
I0421 14:56:56.025286 139847166221440 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 0x7f30689a6fc0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0421 14:56:58.419216 139847166221440 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0421 14:56:58.419428 139847166221440 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 0x7f2ec0daa450>}, handler_registry=None
I0421 14:56:58.419732 139847166221440 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f2ec0daa450>` 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 14:56:58.419777 139847166221440 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f2ec0983290>` 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 14:56:58.419805 139847166221440 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 0x7f2ec0daa450>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f2ec0daa450>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f2ec0983290>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f2ec0983290>}).
I0421 14:56:58.420102 139847166221440 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28
I0421 14:56:58.420154 139847166221440 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7f2ec09a4ea0> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0421 14:56:58.552320 139847166221440 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints
I0421 14:56:58.552562 139847166221440 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_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f2ec0982fc0>
I0421 14:56:58.552644 139847166221440 checkpointing.py:301] Checkpoint manager created!
I0421 14:56:58.915533 139847166221440 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0421 14:56:58.918560 139847166221440 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 14:56:58.972261 139847166221440 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 14:56:58.994776 139847166221440 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 14:56:58.994837 139847166221440 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0421 14:56:59.841967 139847166221440 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0421 14:56:59.842081 139847166221440 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 14:57:00.142668 139847166221440 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 14:57:00.142816 139847166221440 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 14:57:00.155499 139847166221440 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 14:57:00.155551 139847166221440 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0421 14:57:00.220808 139847166221440 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 14:57:00.220890 139847166221440 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 14:57:00.233599 139847166221440 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 14:57:00.233653 139847166221440 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 14:57:00.246192 139847166221440 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 14:57:00.246245 139847166221440 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 14:57:00.265899 139847166221440 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0421 14:57:00.265960 139847166221440 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0421 14:57:00.332506 139847166221440 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0421 14:57:00.332582 139847166221440 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None).
I0421 14:57:01.098307 139847166221440 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0421 14:57:01.098414 139847166221440 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 14:57:03.450275 139847166221440 maxtext_utils.py:1654]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0421 14:57:03.450389 139847166221440 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 14:57:03.450432 139847166221440 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 14:57:03.450475 139847166221440 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 14:57:03.450515 139847166221440 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 14:57:03.450544 139847166221440 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 14:57:03.450583 139847166221440 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 14:57:03.450621 139847166221440 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 14:57:03.450649 139847166221440 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 14:57:03.450674 139847166221440 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 14:57:03.450711 139847166221440 maxtext_utils.py:1654]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0421 14:57:03.450746 139847166221440 maxtext_utils.py:1654]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0421 14:57:05.195523 139847166221440 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0421 14:57:05.195623 139847166221440 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None).
I0421 14:57:05.207661 139847166221440 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0421 14:57:05.207714 139847166221440 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None).
I0421 14:57:21.053590 139847166221440 max_utils.py:791] Total memory size: 3.7 GB, Output size: 1.5 GB, Temp size: 2.2 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB.
I0421 14:57:21.054398 139847166221440 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter.
I0421 14:57:21.054611 139847166221440 metric_logger.py:301] number parameters: 1.104 billion
I0421 14:57:41.376555 139847166221440 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0421 14:57:41.518123 139847166221440 checkpointing.py:776] Waited 0.14155936241149902 seconds for step 0 to finish before starting checkpointing.
I0421 14:57:41.518847 139847166221440 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 14:57:41.519030 139847166221440 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0
I0421 14:57:41.519561 139847166221440 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0.
I0421 14:57:41.598514 139847166221440 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient
I0421 14:57:41.691202 139755389978176 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0
I0421 14:57:41.692060 139847166221440 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0421 14:57:41.692231 139847166221440 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 14:57:43.256523 139755379492416 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0/items
I0421 14:57:43.401550 139755211720256 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783463119944461, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776783463119944461, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0/_CHECKPOINT_METADATA
I0421 14:57:43.557327 139847166221440 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.866452s
I0421 14:57:43.557795 139847166221440 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 6.302 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host)
I0421 14:57:43.557852 139847166221440 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.958206s (batch_requests_ready=0.087639s, total_serialization_initiated=1.870205s, others=0.000361s)
I0421 14:57:43.557912 139847166221440 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.958837s (all_items=0.000029s, per_item={'items': '0.00002933'}, temp_paths=1.958808)
I0421 14:57:43.558828 139755190748736 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0421 14:57:43.558936 139847166221440 async_checkpointer.py:561] Finished blocking save. Time taken: 2.039837s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0.
I0421 14:57:43.559136 139847166221440 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 14:57:43.559366 139755180262976 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0421 14:57:43.559482 139847166221440 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776783461.5188327, 'wait_for_prev_duration_secs': 4.601478576660156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776783461.5190709, 'checkpointer_blocking_duration_secs': 2.0399935245513916, 'get_old_steps_start_time': 1776783463.559079, 'get_old_steps_duration_secs': 2.2172927856445312e-05, 'checkpoint_manager_blocking_start_time': 1776783461.5187244, 'checkpoint_manager_blocking_duration_secs': 2.0407307147979736}
I0421 14:57:43.559666 139847166221440 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0421 14:57:43.559706 139847166221440 max_utils.py:750] 
Memstats: After params initialized:
I0421 14:57:43.559796 139847166221440 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0))
I0421 14:57:43.559829 139847166221440 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_1(process=0,(1,0,0,0))
I0421 14:57:43.559851 139847166221440 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_2(process=0,(0,1,0,0))
I0421 14:57:43.559869 139847166221440 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_3(process=0,(1,1,0,0))
I0421 14:57:43.559886 139847166221440 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_4(process=0,(0,2,0,0))
I0421 14:57:43.559904 139847166221440 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_5(process=0,(1,2,0,0))
I0421 14:57:43.559921 139847166221440 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_6(process=0,(0,3,0,0))
I0421 14:57:43.559938 139847166221440 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_7(process=0,(1,3,0,0))
I0421 14:57:43.716119 1486228 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json
I0421 14:57:43.716189 1486228 google_auth_provider.cc:156] Using OAuth2 AuthProvider
I0421 14:57:43.974198 139847166221440 metric_logger.py:196] completed step: 0, seconds: 20.321, TFLOP/s/device: 0.669, Tokens/s/device: 100.783, total_weights: 13328, loss: 10.840, lm_loss: 10.840, perplexity: 51025.957
I0421 14:57:43.975129 139847166221440 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/tensorboard/'
I0421 14:57:44.093073 139847166221440 metric_logger.py:196] completed step: 1, seconds: 2.595, TFLOP/s/device: 5.235, Tokens/s/device: 789.108, total_weights: 12332, loss: 10.869, lm_loss: 10.869, perplexity: 52515.969
I0421 14:57:44.214453 139847166221440 metric_logger.py:196] completed step: 2, seconds: 0.011, TFLOP/s/device: 1260.635, Tokens/s/device: 190016.701, total_weights: 15161, loss: 9.874, lm_loss: 9.874, perplexity: 19421.770
I0421 14:57:44.335500 139847166221440 metric_logger.py:196] completed step: 3, seconds: 0.121, TFLOP/s/device: 112.466, Tokens/s/device: 16952.099, total_weights: 13327, loss: 9.406, lm_loss: 9.406, perplexity: 12166.079
I0421 14:58:02.759711 139755389978176 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_04_int8/checkpoints/0/items/array_metadatas/process_0
I0421 14:58:02.760690 139847166221440 metric_logger.py:196] completed step: 4, seconds: 0.121, TFLOP/s/device: 112.064, Tokens/s/device: 16891.558, total_weights: 11939, loss: 9.149, lm_loss: 9.149, perplexity: 9403.998
I0421 14:58:02.788460 139847166221440 metric_logger.py:196] completed step: 5, seconds: 0.121, TFLOP/s/device: 112.039, Tokens/s/device: 16887.797, total_weights: 15502, loss: 9.058, lm_loss: 9.058, perplexity: 8587.564
I0421 14:58:02.908335 139847166221440 metric_logger.py:196] completed step: 6, seconds: 18.441, TFLOP/s/device: 0.737, Tokens/s/device: 111.057, total_weights: 13864, loss: 8.892, lm_loss: 8.892, perplexity: 7274.692
I0421 14:58:03.030149 139847166221440 metric_logger.py:196] completed step: 7, seconds: 0.009, TFLOP/s/device: 1525.273, Tokens/s/device: 229905.703, total_weights: 12988, loss: 8.751, lm_loss: 8.751, perplexity: 6314.503
I0421 14:58:03.152361 139847166221440 metric_logger.py:196] completed step: 8, seconds: 0.121, TFLOP/s/device: 112.092, Tokens/s/device: 16895.738, total_weights: 13820, loss: 8.791, lm_loss: 8.791, perplexity: 6573.798
I0421 14:58:03.273216 139847166221440 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0421 14:58:03.274688 139847166221440 checkpointing.py:776] Waited 0.0014448165893554688 seconds for step 9 to finish before starting checkpointing.
I0421 14:58:03.275284 139847166221440 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 14:58:30.397371 139755201234496 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 46.686020s (commit=46.209076s, array_metadata_write=0.476943s)
I0421 14:58:30.398674 139755190748736 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 258.956 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 48 seconds) (per-host)
I0421 14:58:30.398785 139755190748736 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 46.839803s.
I0421 14:58:30.688615 139755190748736 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783463119944461, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0/_CHECKPOINT_METADATA
I0421 14:58:30.882510 139755190748736 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 14:58:31.073186 139755211720256 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783463119944461, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0/_CHECKPOINT_METADATA
I0421 14:58:31.273030 139755190748736 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0421 14:58:31.273719 139755190748736 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.541075s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0/items
I0421 14:58:31.274453 139755190748736 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0/items
I0421 14:58:31.509359 139755190748736 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0
I0421 14:58:32.173053 139755190748736 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_04_int8/checkpoints/0`.
I0421 14:58:32.173798 139755190748736 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 50.654706s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/0
I0421 14:58:32.173883 139755190748736 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 48.614902s.
I0421 14:58:32.174074 139755180262976 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 14:58:32.174192 139755180262976 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0421 14:58:32.174251 139755180262976 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0421 14:58:32.174300 139755180262976 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0421 14:58:32.174428 139847166221440 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 14:58:32.174566 139847166221440 checkpoint_manager.py:1441] Waiting for previous save to complete took 28.899304 seconds. If this number is high, consider checkpointing less frequently.
I0421 14:58:32.175602 139847166221440 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9
I0421 14:58:32.175922 139847166221440 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/9.
I0421 14:58:32.348457 139755180262976 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/9
I0421 14:58:32.354729 139847166221440 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0421 14:58:32.354836 139847166221440 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 14:58:33.081053 139755159291456 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/9/items
I0421 14:58:43.290079 139847166221440 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.935769s
I0421 14:58:43.295006 139847166221440 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.119 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 11 seconds) (per-host)
I0421 14:58:43.295086 139847166221440 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 11.030595s (batch_requests_ready=0.086656s, total_serialization_initiated=10.939131s, others=0.004808s)
I0421 14:58:43.295186 139847166221440 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 11.031214s (all_items=0.000021s, per_item={'items': '0.00002098'}, temp_paths=11.031193)
I0421 14:58:43.296709 139755180262976 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0421 14:58:43.296796 139847166221440 async_checkpointer.py:561] Finished blocking save. Time taken: 11.121149s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/9.
I0421 14:58:43.297013 139847166221440 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 14:58:43.297311 139755190748736 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0421 14:58:43.297447 139847166221440 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776783483.275231, 'wait_for_prev_duration_secs': 28.899303674697876, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776783512.1756258, 'checkpointer_blocking_duration_secs': 11.121283531188965, 'get_old_steps_start_time': 1776783523.2969313, 'get_old_steps_duration_secs': 3.886222839355469e-05, 'checkpoint_manager_blocking_start_time': 1776783483.2751677, 'checkpoint_manager_blocking_duration_secs': 40.02225303649902}
I0421 14:58:43.297613 139847166221440 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0421 14:58:43.297649 139847166221440 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 14:58:44.049566 139755389978176 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_04_int8/checkpoints/9/items/array_metadatas/process_0
I0421 14:59:18.715035 139755295606336 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 35.419255s (commit=34.997578s, array_metadata_write=0.421677s)
I0421 14:59:18.716207 139755180262976 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 272.041 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 46 seconds) (per-host)
I0421 14:59:18.716261 139755180262976 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.419423s.
I0421 14:59:19.112247 139755180262976 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 14:59:19.509519 139755180262976 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0421 14:59:19.510263 139755180262976 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.547407s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/9/items
I0421 14:59:19.510987 139755180262976 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/9/items
I0421 14:59:19.771556 139755180262976 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/9
I0421 14:59:20.424469 139755180262976 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_04_int8/checkpoints/9`.
I0421 14:59:20.425232 139755180262976 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 48.249592s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_04_int8/checkpoints/9
I0421 14:59:20.425329 139755180262976 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 37.128490s.
I0421 14:59:20.425521 139755190748736 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 14:59:20.425647 139755190748736 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0421 14:59:20.425718 139755190748736 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0421 14:59:20.425761 139755190748736 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0421 14:59:20.425934 139847166221440 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 14:59:20.426107 139847166221440 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 14:59:20.426985 139847166221440 metric_logger.py:196] completed step: 9, seconds: 0.124, TFLOP/s/device: 109.279, Tokens/s/device: 16471.762, total_weights: 12300, loss: 8.774, lm_loss: 8.774, perplexity: 6465.313
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
fix/gate-fp8-nan-sanitization  ·  cd5a4f898  ·  fix_gate_fp8_nan_sanitization_20260422_174509  ·  full log 
2026-04-22 17:53:22.702177: 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 00:00:1776880402.715140 2025172 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776880402.718925 2025172 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776880402.730383 2025172 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776880402.730396 2025172 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776880402.730398 2025172 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776880402.730400 2025172 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'}
2026-04-22 17:53:26.417202: 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)
I0422 17:53:26.552702 127115095414592 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag.
I0422 17:53:56.468705 127115095414592 max_utils.py:800] System Information: Jax Version: 0.8.1
I0422 17:53:56.468968 127115095414592 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0422 17:53:56.469018 127115095414592 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
I0422 17:53:56.469064 127115095414592 train_utils.py:348] WARNING: 'dataset_path' might be pointing your local file system
I0422 17:53:56.469248 127115095414592 train.py:706] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper.
I0422 17:53:57.131097 127115095414592 maxtext_utils.py:1604] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0422 17:53:57.131461 127115095414592 checkpointing.py:677] Setting up checkpoint logger...
I0422 17:53:57.131508 127115095414592 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0422 17:53:57.131543 127115095414592 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None
I0422 17:53:57.131978 127115095414592 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 0x739c031b2c60>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0422 17:53:59.549970 127115095414592 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0422 17:53:59.550386 127115095414592 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 0x739a2df6da00>}, handler_registry=None
I0422 17:53:59.550620 127115095414592 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x739a2df6da00>` 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`.
I0422 17:53:59.550663 127115095414592 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x739aa29659a0>` 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`.
I0422 17:53:59.550692 127115095414592 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 0x739a2df6da00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x739a2df6da00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x739aa29659a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x739aa29659a0>}).
I0422 17:53:59.550977 127115095414592 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28
I0422 17:53:59.551031 127115095414592 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x739a2df7c540> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0422 17:53:59.696059 127115095414592 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints
I0422 17:53:59.696412 127115095414592 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_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x739a2df6d130>
I0422 17:53:59.696710 127115095414592 checkpointing.py:301] Checkpoint manager created!
I0422 17:53:59.776742 127115095414592 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0
I0422 17:53:59.780065 127115095414592 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.
I0422 17:53:59.848001 127115095414592 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
I0422 17:53:59.873940 127115095414592 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2
I0422 17:53:59.874018 127115095414592 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2
I0422 17:54:00.722592 127115095414592 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0422 17:54:00.722694 127115095414592 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0422 17:54:01.027283 127115095414592 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 17:54:01.027373 127115095414592 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0422 17:54:01.040079 127115095414592 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 17:54:01.040130 127115095414592 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None).
I0422 17:54:01.105660 127115095414592 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 17:54:01.105747 127115095414592 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0422 17:54:01.118586 127115095414592 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 17:54:01.118639 127115095414592 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0422 17:54:01.131297 127115095414592 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 17:54:01.131349 127115095414592 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0422 17:54:01.151290 127115095414592 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0422 17:54:01.151351 127115095414592 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None).
I0422 17:54:01.218608 127115095414592 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0422 17:54:01.218691 127115095414592 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None).
I0422 17:54:02.011803 127115095414592 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0422 17:54:02.011909 127115095414592 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
I0422 17:54:03.325756 127115095414592 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0422 17:54:03.325874 127115095414592 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)
I0422 17:54:03.325917 127115095414592 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)
I0422 17:54:03.325963 127115095414592 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')
I0422 17:54:03.326004 127115095414592 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)
I0422 17:54:03.326034 127115095414592 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)
I0422 17:54:03.326092 127115095414592 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)
I0422 17:54:03.326139 127115095414592 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')
I0422 17:54:03.326170 127115095414592 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)
I0422 17:54:03.326198 127115095414592 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)
I0422 17:54:03.326235 127115095414592 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0422 17:54:03.326271 127115095414592 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0422 17:54:05.093653 127115095414592 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0422 17:54:05.093752 127115095414592 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None).
I0422 17:54:05.106178 127115095414592 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0422 17:54:05.106230 127115095414592 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None).
I0422 17:54:20.492918 127115095414592 max_utils.py:791] Total memory size: 3.7 GB, Output size: 1.5 GB, Temp size: 2.2 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB.
I0422 17:54:20.493830 127115095414592 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter.
I0422 17:54:20.494133 127115095414592 metric_logger.py:301] number parameters: 1.104 billion
I0422 17:54:40.104556 127115095414592 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0422 17:54:40.249609 127115095414592 checkpointing.py:776] Waited 0.14502859115600586 seconds for step 0 to finish before starting checkpointing.
I0422 17:54:40.250585 127115095414592 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 17:54:40.250764 127115095414592 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0
I0422 17:54:40.251185 127115095414592 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0.
I0422 17:54:40.343284 127115095414592 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient
I0422 17:54:40.432654 126985873393216 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0
I0422 17:54:40.440151 127115095414592 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0422 17:54:40.440243 127115095414592 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
I0422 17:54:41.206478 126985862907456 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0/items
I0422 17:54:41.367580 126985831450176 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776880481109702982, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776880481109702982, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0/_CHECKPOINT_METADATA
I0422 17:54:41.681610 2028551 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json
I0422 17:54:41.681666 2028551 google_auth_provider.cc:156] Using OAuth2 AuthProvider
I0422 17:54:48.741010 127115095414592 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 8.301360s
I0422 17:54:48.745167 127115095414592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.469 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 8 seconds) (per-host)
I0422 17:54:48.745234 127115095414592 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 8.400542s (batch_requests_ready=0.091240s, total_serialization_initiated=8.305267s, others=0.004035s)
I0422 17:54:48.745307 127115095414592 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 8.401502s (all_items=0.000021s, per_item={'items': '0.00002050'}, temp_paths=8.401481)
I0422 17:54:48.747433 126985300870720 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0422 17:54:48.747607 127115095414592 async_checkpointer.py:561] Finished blocking save. Time taken: 8.496802s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0.
I0422 17:54:48.747814 127115095414592 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 17:54:48.748004 126985290384960 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0422 17:54:48.748082 127115095414592 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776880480.2505715, 'wait_for_prev_duration_secs': 4.482269287109375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776880480.250786, 'checkpointer_blocking_duration_secs': 8.496948957443237, 'get_old_steps_start_time': 1776880488.7477531, 'get_old_steps_duration_secs': 2.4557113647460938e-05, 'checkpoint_manager_blocking_start_time': 1776880480.2504919, 'checkpoint_manager_blocking_duration_secs': 8.497562408447266}
I0422 17:54:48.748195 127115095414592 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0422 17:54:48.748236 127115095414592 max_utils.py:750] 
Memstats: After params initialized:
I0422 17:54:48.748282 127115095414592 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0))
I0422 17:54:48.748314 127115095414592 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_1(process=0,(1,0,0,0))
I0422 17:54:48.748335 127115095414592 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_2(process=0,(0,1,0,0))
I0422 17:54:48.748354 127115095414592 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_3(process=0,(1,1,0,0))
I0422 17:54:48.748372 127115095414592 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_4(process=0,(0,2,0,0))
I0422 17:54:48.748389 127115095414592 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_5(process=0,(1,2,0,0))
I0422 17:54:48.748407 127115095414592 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_6(process=0,(0,3,0,0))
I0422 17:54:48.748424 127115095414592 max_utils.py:756] 	Using (GB) 1.6 / 31.25 (5.120000%) on TPU_7(process=0,(1,3,0,0))
I0422 17:54:49.146167 127115095414592 metric_logger.py:196] completed step: 0, seconds: 19.610, TFLOP/s/device: 0.693, Tokens/s/device: 104.438, total_weights: 13328, loss: 10.840, lm_loss: 10.840, perplexity: 51025.957
I0422 17:54:49.147149 127115095414592 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/tensorboard/'
I0422 17:54:49.265223 127115095414592 metric_logger.py:196] completed step: 1, seconds: 9.039, TFLOP/s/device: 1.503, Tokens/s/device: 226.570, total_weights: 12332, loss: 10.869, lm_loss: 10.869, perplexity: 52515.969
I0422 17:54:49.386942 127115095414592 metric_logger.py:196] completed step: 2, seconds: 0.011, TFLOP/s/device: 1220.218, Tokens/s/device: 183924.562, total_weights: 15161, loss: 9.874, lm_loss: 9.874, perplexity: 19421.770
I0422 17:54:49.508568 127115095414592 metric_logger.py:196] completed step: 3, seconds: 0.122, TFLOP/s/device: 111.807, Tokens/s/device: 16852.777, total_weights: 13327, loss: 9.406, lm_loss: 9.406, perplexity: 12166.079
I0422 17:54:49.517397 126985841935936 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0/items/array_metadatas/process_0
I0422 17:55:10.984322 127115095414592 metric_logger.py:196] completed step: 4, seconds: 0.121, TFLOP/s/device: 112.671, Tokens/s/device: 16983.025, total_weights: 11939, loss: 9.149, lm_loss: 9.149, perplexity: 9403.998
I0422 17:55:10.997410 127115095414592 metric_logger.py:196] completed step: 5, seconds: 0.122, TFLOP/s/device: 111.286, Tokens/s/device: 16774.236, total_weights: 15502, loss: 9.058, lm_loss: 9.058, perplexity: 8587.564
I0422 17:55:11.117255 127115095414592 metric_logger.py:196] completed step: 6, seconds: 21.477, TFLOP/s/device: 0.633, Tokens/s/device: 95.359, total_weights: 13864, loss: 8.892, lm_loss: 8.892, perplexity: 7274.692
I0422 17:55:11.239088 127115095414592 metric_logger.py:196] completed step: 7, seconds: 0.008, TFLOP/s/device: 1724.912, Tokens/s/device: 259997.461, total_weights: 12988, loss: 8.751, lm_loss: 8.751, perplexity: 6314.503
I0422 17:55:11.360756 127115095414592 metric_logger.py:196] completed step: 8, seconds: 0.121, TFLOP/s/device: 111.974, Tokens/s/device: 16877.915, total_weights: 13820, loss: 8.791, lm_loss: 8.791, perplexity: 6573.798
I0422 17:55:11.481523 127115095414592 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0422 17:55:11.482607 127115095414592 checkpointing.py:776] Waited 0.0011129379272460938 seconds for step 9 to finish before starting checkpointing.
I0422 17:55:11.482936 127115095414592 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 17:55:19.718795 126985502197312 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 30.972005s (commit=30.502137s, array_metadata_write=0.469868s)
I0422 17:55:19.720003 126985300870720 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 320.932 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 39 seconds) (per-host)
I0422 17:55:19.720124 126985300870720 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 30.972472s.
I0422 17:55:19.985646 126985300870720 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776880481109702982, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0/_CHECKPOINT_METADATA
I0422 17:55:20.183809 126985300870720 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0422 17:55:20.407426 126985831450176 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776880481109702982, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0/_CHECKPOINT_METADATA
I0422 17:55:20.634741 126985300870720 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0422 17:55:20.634978 126985300870720 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.611836s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0/items
I0422 17:55:20.635738 126985300870720 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0/items
I0422 17:55:20.906257 126985300870720 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0
I0422 17:55:21.631678 126985300870720 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0`.
I0422 17:55:21.632420 126985300870720 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 41.381622s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/0
I0422 17:55:21.632500 126985300870720 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 32.884852s.
I0422 17:55:21.632655 126985290384960 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 17:55:21.632713 126985290384960 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0422 17:55:21.632773 126985290384960 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0422 17:55:21.632814 126985290384960 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0422 17:55:21.632936 127115095414592 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.
W0422 17:55:21.633089 127115095414592 checkpoint_manager.py:1441] Waiting for previous save to complete took 10.150181 seconds. If this number is high, consider checkpointing less frequently.
I0422 17:55:21.634394 127115095414592 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9
I0422 17:55:21.634713 127115095414592 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9.
I0422 17:55:21.804852 127115095414592 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array.
I0422 17:55:21.804959 127115095414592 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
I0422 17:55:21.806040 126985300870720 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9
I0422 17:55:22.568146 126985841935936 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9/items
I0422 17:55:24.477926 127115095414592 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 2.673692s
I0422 17:55:24.482075 127115095414592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.465 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 2 seconds) (per-host)
I0422 17:55:24.482142 127115095414592 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 2.764265s (batch_requests_ready=0.083095s, total_serialization_initiated=2.677150s, others=0.004020s)
I0422 17:55:24.482227 127115095414592 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.765321s (all_items=0.000021s, per_item={'items': '0.00002050'}, temp_paths=2.765300)
I0422 17:55:24.483881 126985300870720 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started.
I0422 17:55:24.483986 127115095414592 async_checkpointer.py:561] Finished blocking save. Time taken: 2.849543s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9.
I0422 17:55:24.484203 127115095414592 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 17:55:24.484415 126985290384960 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0422 17:55:24.484538 127115095414592 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776880511.4828787, 'wait_for_prev_duration_secs': 10.150180578231812, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776880521.6344225, 'checkpointer_blocking_duration_secs': 2.8496882915496826, 'get_old_steps_start_time': 1776880524.4841287, 'get_old_steps_duration_secs': 3.3855438232421875e-05, 'checkpoint_manager_blocking_start_time': 1776880511.48284, 'checkpoint_manager_blocking_duration_secs': 13.00167179107666}
I0422 17:55:24.484661 127115095414592 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0422 17:55:24.484694 127115095414592 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 17:55:25.248933 126986022291008 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9/items/array_metadatas/process_0
I0422 17:56:16.351314 126985841935936 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 51.868388s (commit=51.419465s, array_metadata_write=0.448923s)
I0422 17:56:16.352473 126985300870720 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 231.297 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 54 seconds) (per-host)
I0422 17:56:16.352596 126985300870720 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 51.868562s.
I0422 17:56:16.842790 126985300870720 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0.
I0422 17:56:17.242178 126985300870720 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389).
I0422 17:56:17.242354 126985300870720 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.574592s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9/items
I0422 17:56:17.243236 126985300870720 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9/items
I0422 17:56:17.519065 126985300870720 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9
I0422 17:56:18.232869 126985300870720 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9`.
I0422 17:56:18.233615 126985300870720 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 56.599180s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_04_int8/checkpoints/9
I0422 17:56:18.233696 126985300870720 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 53.749665s.
I0422 17:56:18.233892 126985290384960 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 17:56:18.234010 126985290384960 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0422 17:56:18.234076 126985290384960 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0422 17:56:18.234116 126985290384960 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0422 17:56:18.234264 127115095414592 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.
I0422 17:56:18.234395 127115095414592 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 17:56:18.235268 127115095414592 metric_logger.py:196] completed step: 9, seconds: 0.122, TFLOP/s/device: 111.591, Tokens/s/device: 16820.250, total_weights: 12300, loss: 8.774, lm_loss: 8.774, perplexity: 6465.313
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops