MaxView

Timing Test Reference Updates
‹ 15_ocdbt_false_resumeCase: 15_ocdbt_false_save15_ocdbt_true_resume ›

Metrics: main (e27fc1e97) vs feat/nnx-post-train-fixes (e27fc1e97)

Metricmain  e27fc1e97feat/nnx-post-train-fixes  e27fc1e97Diff (feat/nnx-post-train-fixes − main)
Parameters1.104 billion1.104 billion—
Final loss7.18107.18100
TFLOP/s92.16492.084-0.08
Tok/s13892.013879.9-12.051
Avg s/step4.4654.627+0.162
Memory %1.381.380
JAX0.9.20.9.2—

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

main  ·  e27fc1e97  ·  main_20260421_061359  ·  full log 
XPK Start: Tue Apr 21 07:25:41 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-21 07:26:05.920419: 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 07:26:06.130208 135870948595520 max_utils.py:273] Attempting to initialize the jax distributed system...
I0421 07:26:15.171269 135870948595520 distributed.py:149] Starting JAX distributed service on [::]:8482
I0421 07:26:15.173539 135870948595520 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-etyex-slice-job-0-0.mt-15-ocdbt-false-sv-etyex:8482
I0421 07:26:16.364723 135870948595520 max_utils.py:284] Jax distributed system initialized!
I0421 07:26:21.432838 135870948595520 max_utils.py:800] System Information: Jax Version: 0.9.2
I0421 07:26:21.432945 135870948595520 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0421 07:26:21.432985 135870948595520 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0421 07:26:21.433021 135870948595520 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0421 07:26:22.141453 135870948595520 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 07:26:22.141751 135870948595520 checkpointing.py:677] Setting up checkpoint logger...
I0421 07:26:22.141810 135870948595520 checkpointing.py:233] Creating checkpoint manager with ocdbt=False and zarr3=True
I0421 07:26:22.141855 135870948595520 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0421 07:26:22.142198 135870948595520 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7b9233f4c6e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0421 07:26:24.987134 135870948595520 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0421 07:26:24.987374 135870948595520 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b7ed0369f10>}, handler_registry=None
I0421 07:26:24.987614 135870948595520 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b7ed0369f10>` 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 07:26:24.987664 135870948595520 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b7ed036c1a0>` 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 07:26:24.987713 135870948595520 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 0x7b7ed0369f10>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b7ed0369f10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b7ed036c1a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b7ed036c1a0>}).
I0421 07:26:24.988049 135870948595520 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0421 07:26:24.988120 135870948595520 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b7ea849d800> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0421 07:26:26.543112 135870948595520 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints
I0421 07:26:26.975852 135870948595520 checkpoint_manager.py:929] [process=4][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_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=5), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b7f2416fbf0>
I0421 07:26:26.976022 135870948595520 checkpointing.py:301] Checkpoint manager created!
I0421 07:26:27.900840 135870948595520 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0421 07:26:27.900958 135870948595520 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 07:26:28.282256 135870948595520 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 07:26:28.282354 135870948595520 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 07:26:28.298802 135870948595520 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 07:26:28.298864 135870948595520 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 07:26:28.322644 135870948595520 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 07:26:28.322722 135870948595520 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 07:26:28.339312 135870948595520 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 07:26:28.339375 135870948595520 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 07:26:28.356032 135870948595520 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 07:26:28.356098 135870948595520 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 07:26:28.380692 135870948595520 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0421 07:26:28.380776 135870948595520 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 07:26:28.401887 135870948595520 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0421 07:26:28.401955 135870948595520 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0421 07:26:28.618778 135870948595520 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0421 07:26:28.618890 135870948595520 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0421 07:26:30.051906 135870948595520 maxtext_utils.py:1668]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0421 07:26:30.052035 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 07:26:30.052089 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 07:26:30.052148 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0421 07:26:30.052201 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0421 07:26:30.052241 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0421 07:26:30.052294 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 07:26:30.052347 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0421 07:26:30.052388 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0421 07:26:30.052426 135870948595520 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 07:26:30.052482 135870948595520 maxtext_utils.py:1668]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0421 07:26:30.052533 135870948595520 maxtext_utils.py:1668]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]

    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0421 07:26:30.545062 135870948595520 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0421 07:26:30.545157 135870948595520 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0421 07:26:30.560477 135870948595520 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0421 07:26:30.560537 135870948595520 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0421 07:26:41.335864 135870948595520 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0421 07:26:41.336685 135870948595520 metric_logger.py:301] number parameters: 1.104 billion
I0421 07:26:53.114420 135870948595520 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0421 07:26:53.267866 135870948595520 checkpointing.py:776] Waited 0.1534271240234375 seconds for step 0 to finish before starting checkpointing.
I0421 07:26:53.270535 135870948595520 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 07:26:53.272153 135870948595520 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0
I0421 07:26:53.273508 135870948595520 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/0.
I0421 07:26:53.601502 135870948595520 signaling_client.py:364] Using JaxDistributedSignalingClient
I0421 07:26:53.602434 135870948595520 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 07:26:53.602491 135870948595520 replica_slices.py:424] 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 07:26:53.880653 135870948595520 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279223s
I0421 07:26:53.880837 135870948595520 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.416 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28481459617614746 s) (per-host)
I0421 07:26:53.880891 135870948595520 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.284879s (batch_requests_ready=0.002304s, total_serialization_initiated=0.282502s, others=0.000072s)
I0421 07:26:53.881004 135870948595520 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288933s (all_items=0.000020s, per_item={'items': '0.00001979'}, temp_paths=0.288913)
I0421 07:26:53.881829 135870948595520 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/0.
I0421 07:26:53.882081 135745308735232 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 07:46:53.882056
I0421 07:26:53.887176 135870948595520 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 07:26:53.887393 135743658862336 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0421 07:26:53.887507 135870948595520 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776756413.2705166, 'wait_for_prev_duration_secs': 6.365776062011719e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776756413.2721915, 'checkpointer_blocking_duration_secs': 0.6099975109100342, 'get_old_steps_start_time': 1776756413.8822138, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776756413.268408, 'checkpoint_manager_blocking_duration_secs': 0.6190617084503174}
I0421 07:26:53.887611 135870948595520 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0421 07:26:53.887659 135870948595520 max_utils.py:750] 
Memstats: After params initialized:
I0421 07:26:53.887723 135870948595520 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0))
I0421 07:26:53.887758 135870948595520 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0))
I0421 07:26:53.887784 135870948595520 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0))
I0421 07:26:53.887810 135870948595520 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0))
I0421 07:26:54.195849 135870948595520 metric_logger.py:196] completed step: 0, seconds: 11.778, TFLOP/s/device: 1.154, Tokens/s/device: 173.889, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0421 07:26:54.384649 135870948595520 metric_logger.py:196] completed step: 1, seconds: 1.080, TFLOP/s/device: 12.582, Tokens/s/device: 1896.544, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0421 07:26:54.806681 135870948595520 metric_logger.py:196] completed step: 2, seconds: 0.042, TFLOP/s/device: 327.219, Tokens/s/device: 49322.062, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719
I0421 07:26:54.954085 135870948595520 metric_logger.py:196] completed step: 3, seconds: 0.422, TFLOP/s/device: 32.188, Tokens/s/device: 4851.758, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506
I0421 07:26:54.958668 135870948595520 checkpointing.py:772] Waiting for step 5 to finish before checkpoint...
I0421 07:26:55.248247 135870948595520 checkpointing.py:776] Waited 0.28953981399536133 seconds for step 5 to finish before starting checkpointing.
I0421 07:26:55.251225 135870948595520 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 07:26:56.772023    2755 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0421 07:26:58.803049 135743667255040 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_4
I0421 07:27:04.317764 135745308735232 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 147.327 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.72169017791748 s) (per-host)
I0421 07:27:04.317886 135745308735232 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.435726s.
I0421 07:27:10.119973 135745308735232 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 16.237797s.
I0421 07:27:10.120299 135743658862336 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 07:27:10.120414 135743658862336 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0421 07:27:10.120465 135743658862336 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0421 07:27:10.122087 135743658862336 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0421 07:27:10.122268 135870948595520 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0421 07:27:10.122399 135870948595520 checkpoint_manager.py:1452] Waiting for previous save to complete took 14.871174 seconds. If this number is high, consider checkpointing less frequently.
I0421 07:27:10.124447 135870948595520 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 5
I0421 07:27:10.126737 135870948595520 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/5.
I0421 07:27:10.838508 135870948595520 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 07:27:10.838603 135870948595520 replica_slices.py:424] 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 07:27:10.876905 135870948595520 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039477s
I0421 07:27:10.877073 135870948595520 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 35.871 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04300355911254883 s) (per-host)
I0421 07:27:10.877125 135870948595520 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043066s (batch_requests_ready=0.001892s, total_serialization_initiated=0.041104s, others=0.000070s)
I0421 07:27:10.877223 135870948595520 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048027s (all_items=0.000016s, per_item={'items': '0.00001645'}, temp_paths=0.048010)
I0421 07:27:10.879240 135870948595520 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/5.
I0421 07:27:10.879542 135744790198016 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 07:47:10.879506
I0421 07:27:10.882876 135870948595520 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 07:27:10.883115 135743658862336 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0421 07:27:10.883250 135870948595520 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776756415.2511964, 'wait_for_prev_duration_secs': 14.871173620223999, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776756430.1244855, 'checkpointer_blocking_duration_secs': 0.7551939487457275, 'get_old_steps_start_time': 1776756430.8797133, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776756415.2489247, 'checkpoint_manager_blocking_duration_secs': 15.634290933609009}
I0421 07:27:10.883359 135870948595520 checkpointing.py:408] Started an asynchronous checkpoint save for step 5
I0421 07:27:10.884173 135870948595520 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.643, Tokens/s/device: 13361.169, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355
I0421 07:27:10.889859 135870948595520 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.201, Tokens/s/device: 13897.465, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467
I0421 07:27:16.336888 135743667255040 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_4
I0421 07:27:22.073878 135744790198016 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 140.536 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.239787101745605 s) (per-host)
I0421 07:27:22.073987 135744790198016 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.194332s.
I0421 07:27:27.393532 135744790198016 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 16.513860s.
I0421 07:27:27.393849 135743658862336 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 07:27:27.393967 135743658862336 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0421 07:27:27.394016 135743658862336 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts...
I0421 07:27:27.395692 135743658862336 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts.
I0421 07:27:33.149177 135870948595520 metric_logger.py:196] completed step: 6, seconds: 15.930, TFLOP/s/device: 0.853, Tokens/s/device: 128.565, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740
I0421 07:27:33.296606 135870948595520 metric_logger.py:196] completed step: 7, seconds: 22.113, TFLOP/s/device: 0.614, Tokens/s/device: 92.615, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151
I0421 07:27:33.443819 135870948595520 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.261, Tokens/s/device: 13454.388, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098
I0421 07:27:33.590780 135870948595520 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0421 07:27:33.591456 135870948595520 checkpointing.py:776] Waited 0.0006947517395019531 seconds for step 9 to finish before starting checkpointing.
I0421 07:27:33.593542 135870948595520 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 07:27:33.595449 135870948595520 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9
I0421 07:27:33.596755 135870948595520 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/9.
I0421 07:27:33.927691 135870948595520 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 07:27:33.927797 135870948595520 replica_slices.py:424] 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 07:27:33.969850 135870948595520 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042957s
I0421 07:27:33.970007 135870948595520 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 33.164 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04651355743408203 s) (per-host)
I0421 07:27:33.970056 135870948595520 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046573s (batch_requests_ready=0.001766s, total_serialization_initiated=0.044743s, others=0.000064s)
I0421 07:27:33.970141 135870948595520 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050579s (all_items=0.000012s, per_item={'items': '0.00001168'}, temp_paths=0.050567)
I0421 07:27:33.970848 135870948595520 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.38 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/9.
I0421 07:27:33.971115 135743658862336 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 07:47:33.971091
I0421 07:27:33.973026 135870948595520 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 07:27:33.973250 135744211363584 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0421 07:27:33.973364 135870948595520 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776756453.5935094, 'wait_for_prev_duration_secs': 7.462501525878906e-05, 'time_between_consecutive_saves_sec': 6.197761535644531, 'checkpointer_blocking_start_time': 1776756453.5954876, 'checkpointer_blocking_duration_secs': 0.3757305145263672, 'get_old_steps_start_time': 1776756453.9712393, 'get_old_steps_duration_secs': 2.4318695068359375e-05, 'checkpoint_manager_blocking_start_time': 1776756453.591732, 'checkpoint_manager_blocking_duration_secs': 0.38160228729248047}
I0421 07:27:33.973476 135870948595520 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0421 07:27:33.973521 135870948595520 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 07:27:39.966474 135743667255040 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_4
I0421 07:27:45.189615 135743658862336 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 140.208 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.266082286834717 s) (per-host)
I0421 07:27:45.189749 135743658862336 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.218555s.
I0421 07:27:50.241838 135743658862336 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 16.270627s.
I0421 07:27:50.242046 135744211363584 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 07:27:50.242108 135744211363584 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0421 07:27:50.242162 135744211363584 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0421 07:27:50.243682 135744211363584 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0421 07:27:50.243873 135870948595520 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0421 07:27:50.244025 135870948595520 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 07:27:50.245054 135870948595520 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.164, Tokens/s/device: 13891.998, total_weights: 65536, loss: 7.181, lm_loss: 7.181, perplexity: 1314.495
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Tue Apr 21 07:28:01 UTC 2026
EXIT_CODE=0
feat/nnx-post-train-fixes  ·  e27fc1e97  ·  feat_nnx_post_train_fixes_20260421_114057  ·  full log 
XPK Start: Tue Apr 21 12:51:14 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-21 12:51:38.407136: 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 12:51:38.588892 136269990262592 max_utils.py:273] Attempting to initialize the jax distributed system...
I0421 12:51:47.631157 136269990262592 distributed.py:149] Starting JAX distributed service on [::]:8482
I0421 12:51:47.633521 136269990262592 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-cti1k-slice-job-0-0.mt-15-ocdbt-false-sv-cti1k:8482
I0421 12:51:48.645175 136269990262592 max_utils.py:284] Jax distributed system initialized!
I0421 12:51:54.904298 136269990262592 max_utils.py:800] System Information: Jax Version: 0.9.2
I0421 12:51:54.904400 136269990262592 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0421 12:51:54.904438 136269990262592 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0421 12:51:54.904472 136269990262592 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0421 12:51:55.603109 136269990262592 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 12:51:55.603677 136269990262592 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 12:51:55.603863 136269990262592 checkpointing.py:688] Setting up checkpoint logger...
I0421 12:51:55.603914 136269990262592 checkpointing.py:234] Creating checkpoint manager with ocdbt=False and zarr3=True
I0421 12:51:55.603955 136269990262592 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0421 12:51:55.604284 136269990262592 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7bef1cd389b0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0421 12:51:58.556478 136269990262592 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0421 12:51:58.556704 136269990262592 checkpoint_manager.py:708] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bdac0192870>}, handler_registry=None
I0421 12:51:58.556951 136269990262592 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bdac0192870>` 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 12:51:58.556999 136269990262592 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bdac01953d0>` 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 12:51:58.557038 136269990262592 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 0x7bdac0192870>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bdac0192870>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bdac01953d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bdac01953d0>}).
I0421 12:51:58.557362 136269990262592 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0421 12:51:58.557433 136269990262592 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7bd9e0650540> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0421 12:51:59.491287 136269990262592 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints
I0421 12:51:59.899472 136269990262592 checkpoint_manager.py:929] [process=7][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_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=5), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7bdac0192e40>
I0421 12:51:59.899646 136269990262592 checkpointing.py:302] Checkpoint manager created!
I0421 12:52:00.842280 136269990262592 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0421 12:52:00.842385 136269990262592 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 12:52:01.218624 136269990262592 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 12:52:01.218712 136269990262592 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 12:52:01.235129 136269990262592 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 12:52:01.235189 136269990262592 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 12:52:01.258672 136269990262592 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 12:52:01.258750 136269990262592 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 12:52:01.275079 136269990262592 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 12:52:01.275139 136269990262592 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 12:52:01.291379 136269990262592 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 12:52:01.291439 136269990262592 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 12:52:01.315918 136269990262592 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0421 12:52:01.315988 136269990262592 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 12:52:01.337033 136269990262592 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0421 12:52:01.337097 136269990262592 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0421 12:52:01.556711 136269990262592 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0421 12:52:01.556872 136269990262592 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0421 12:52:02.995968 136269990262592 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0421 12:52:02.996097 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 12:52:02.996155 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 12:52:02.996212 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0421 12:52:02.996264 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0421 12:52:02.996307 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0421 12:52:02.996360 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 12:52:02.996410 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0421 12:52:02.996448 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0421 12:52:02.996484 136269990262592 maxtext_utils.py:1827]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 12:52:02.996532 136269990262592 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0421 12:52:02.996581 136269990262592 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]

    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0421 12:52:03.486347 136269990262592 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0421 12:52:03.486440 136269990262592 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0421 12:52:03.501857 136269990262592 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0421 12:52:03.501919 136269990262592 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0421 12:52:14.504951 136269990262592 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0421 12:52:14.505721 136269990262592 metric_logger.py:301] number parameters: 1.104 billion
I0421 12:52:26.040673 136269990262592 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0421 12:52:26.193514 136269990262592 checkpointing.py:798] Waited 0.15282320976257324 seconds for step 0 to finish before starting checkpointing.
I0421 12:52:26.195785 136269990262592 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 12:52:26.197769 136269990262592 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0
I0421 12:52:26.199482 136269990262592 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/0.
I0421 12:52:26.534559 136269990262592 signaling_client.py:364] Using JaxDistributedSignalingClient
I0421 12:52:26.535423 136269990262592 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 12:52:26.535479 136269990262592 replica_slices.py:424] 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 12:52:26.808121 136269990262592 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273635s
I0421 12:52:26.808304 136269990262592 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 5.527 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2790951728820801 s) (per-host)
I0421 12:52:26.808376 136269990262592 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279174s (batch_requests_ready=0.002292s, total_serialization_initiated=0.276794s, others=0.000088s)
I0421 12:52:26.808494 136269990262592 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283436s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.283419)
I0421 12:52:26.809413 136269990262592 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/0.
I0421 12:52:26.809772 136140991764224 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 13:12:26.809718
I0421 12:52:26.816238 136269990262592 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 12:52:26.816521 136136697571072 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save.
I0421 12:52:26.816648 136269990262592 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776775946.1957655, 'wait_for_prev_duration_secs': 6.771087646484375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776775946.1978588, 'checkpointer_blocking_duration_secs': 0.612065315246582, 'get_old_steps_start_time': 1776775946.8099496, 'get_old_steps_duration_secs': 3.647804260253906e-05, 'checkpoint_manager_blocking_start_time': 1776775946.1939406, 'checkpoint_manager_blocking_duration_secs': 0.6226673126220703}
I0421 12:52:26.816762 136269990262592 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0421 12:52:26.816813 136269990262592 max_utils.py:750] 
Memstats: After params initialized:
I0421 12:52:26.816862 136269990262592 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_26(process=7,(2,6,0,0))
I0421 12:52:26.816894 136269990262592 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0))
I0421 12:52:26.816919 136269990262592 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0))
I0421 12:52:26.816943 136269990262592 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0))
I0421 12:52:27.132510 136269990262592 metric_logger.py:196] completed step: 0, seconds: 11.535, TFLOP/s/device: 1.178, Tokens/s/device: 177.550, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0421 12:52:27.309985 136269990262592 metric_logger.py:196] completed step: 1, seconds: 1.090, TFLOP/s/device: 12.461, Tokens/s/device: 1878.272, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0421 12:52:27.726788 136269990262592 metric_logger.py:196] completed step: 2, seconds: 0.030, TFLOP/s/device: 450.188, Tokens/s/device: 67857.261, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719
I0421 12:52:27.874309 136269990262592 metric_logger.py:196] completed step: 3, seconds: 0.417, TFLOP/s/device: 32.572, Tokens/s/device: 4909.552, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506
I0421 12:52:27.878898 136269990262592 checkpointing.py:794] Waiting for step 5 to finish before checkpoint...
I0421 12:52:28.168301 136269990262592 checkpointing.py:798] Waited 0.2893664836883545 seconds for step 5 to finish before starting checkpointing.
I0421 12:52:28.171124 136269990262592 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 12:52:30.408005    2846 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0421 12:52:32.759861 136136705963776 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_7
I0421 12:52:37.535475 136140991764224 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 143.518 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.006227970123291 s) (per-host)
I0421 12:52:37.535593 136140991764224 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.725708s.
I0421 12:52:44.348442 136140991764224 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 17.538541s.
I0421 12:52:44.348753 136136697571072 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 12:52:44.348886 136136697571072 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save.
I0421 12:52:44.348942 136136697571072 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0421 12:52:44.351263 136136697571072 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0421 12:52:44.351462 136269990262592 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0421 12:52:44.351597 136269990262592 checkpoint_manager.py:1452] Waiting for previous save to complete took 16.180470 seconds. If this number is high, consider checkpointing less frequently.
I0421 12:52:44.353630 136269990262592 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 5
I0421 12:52:44.356122 136269990262592 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/5.
I0421 12:52:45.065706 136269990262592 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 12:52:45.065817 136269990262592 replica_slices.py:424] 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 12:52:45.102578 136269990262592 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037909s
I0421 12:52:45.102756 136269990262592 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 37.270 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04138898849487305 s) (per-host)
I0421 12:52:45.102806 136269990262592 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.041459s (batch_requests_ready=0.001860s, total_serialization_initiated=0.039522s, others=0.000077s)
I0421 12:52:45.102897 136269990262592 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.046579s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.046562)
I0421 12:52:45.103537 136269990262592 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/5.
I0421 12:52:45.103891 136140464060160 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 13:12:45.103851
I0421 12:52:45.111544 136269990262592 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 12:52:45.111866 136136697571072 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save.
I0421 12:52:45.112035 136269990262592 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776775948.1710956, 'wait_for_prev_duration_secs': 16.18046998977661, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776775964.353669, 'checkpointer_blocking_duration_secs': 0.7503652572631836, 'get_old_steps_start_time': 1776775965.1040547, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1776775948.1689193, 'checkpoint_manager_blocking_duration_secs': 16.94308090209961}
I0421 12:52:45.112145 136269990262592 checkpointing.py:409] Started an asynchronous checkpoint save for step 5
I0421 12:52:45.112910 136269990262592 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.838, Tokens/s/device: 13390.610, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355
I0421 12:52:45.118766 136269990262592 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.124, Tokens/s/device: 13885.875, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467
I0421 12:52:50.178042 136136705963776 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_7
I0421 12:52:55.875663 136140464060160 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 146.066 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.814286470413208 s) (per-host)
I0421 12:52:55.875792 136140464060160 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.771786s.
I0421 12:53:04.079220 136140464060160 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 18.975199s.
I0421 12:53:04.079527 136136697571072 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 12:53:04.079653 136136697571072 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save.
I0421 12:53:04.079700 136136697571072 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts...
I0421 12:53:04.081281 136136697571072 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts.
I0421 12:53:07.529022 136269990262592 metric_logger.py:196] completed step: 6, seconds: 17.238, TFLOP/s/device: 0.788, Tokens/s/device: 118.805, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740
I0421 12:53:07.676401 136269990262592 metric_logger.py:196] completed step: 7, seconds: 22.264, TFLOP/s/device: 0.610, Tokens/s/device: 91.986, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151
I0421 12:53:07.826074 136269990262592 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.397, Tokens/s/device: 13474.925, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098
I0421 12:53:07.972754 136269990262592 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0421 12:53:07.973402 136269990262592 checkpointing.py:798] Waited 0.0006880760192871094 seconds for step 9 to finish before starting checkpointing.
I0421 12:53:07.975373 136269990262592 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 12:53:07.976907 136269990262592 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 9
I0421 12:53:07.978354 136269990262592 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/9.
I0421 12:53:08.616569 136269990262592 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 12:53:08.616658 136269990262592 replica_slices.py:424] 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 12:53:08.654746 136269990262592 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039005s
I0421 12:53:08.654889 136269990262592 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 36.311 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.042482852935791016 s) (per-host)
I0421 12:53:08.654935 136269990262592 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.042538s (batch_requests_ready=0.001721s, total_serialization_initiated=0.040756s, others=0.000061s)
I0421 12:53:08.655012 136269990262592 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.046687s (all_items=0.000012s, per_item={'items': '0.00001168'}, temp_paths=0.046675)
I0421 12:53:08.655625 136269990262592 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.68 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/9.
I0421 12:53:08.655896 136116363585280 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 13:13:08.655873
I0421 12:53:08.657744 136269990262592 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 12:53:08.658108 136136697571072 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save.
I0421 12:53:08.658280 136269990262592 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776775987.9753442, 'wait_for_prev_duration_secs': 7.033348083496094e-05, 'time_between_consecutive_saves_sec': 3.8940227031707764, 'checkpointer_blocking_start_time': 1776775987.9769452, 'checkpointer_blocking_duration_secs': 0.6790478229522705, 'get_old_steps_start_time': 1776775988.6560087, 'get_old_steps_duration_secs': 2.5510787963867188e-05, 'checkpoint_manager_blocking_start_time': 1776775987.9736195, 'checkpoint_manager_blocking_duration_secs': 0.6846210956573486}
I0421 12:53:08.658407 136269990262592 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0421 12:53:08.658456 136269990262592 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 12:53:14.230376 136116455839488 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_7
I0421 12:53:18.920628 136116363585280 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 153.237 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.308181285858154 s) (per-host)
I0421 12:53:18.920762 136116363585280 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.264791s.
I0421 12:53:25.291363 136116363585280 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 16.635377s.
I0421 12:53:25.291640 136136697571072 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 12:53:25.291808 136136697571072 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save.
I0421 12:53:25.291861 136136697571072 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0421 12:53:25.293382 136136697571072 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0421 12:53:25.293569 136269990262592 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0421 12:53:25.293724 136269990262592 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 12:53:25.294724 136269990262592 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.084, Tokens/s/device: 13879.947, total_weights: 65536, loss: 7.181, lm_loss: 7.181, perplexity: 1314.495
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Tue Apr 21 12:53:34 UTC 2026
EXIT_CODE=0