MaxView

Timing Test Reference Updates
‹ 15_ocdbt_false_resumeCase: 15_ocdbt_false_save15_ocdbt_true_resume ›

Metrics: main (7f7822822) vs feat/nnx-post-train-fixes (2fbc8c2f1)

Metricmain  7f7822822feat/nnx-post-train-fixes  2fbc8c2f1Diff (feat/nnx-post-train-fixes − main)
Parameters1.104 billion1.104 billion—
Final loss7.18107.18100
TFLOP/s92.20192.169-0.032
Tok/s13897.513892.8-4.713
Avg s/step4.4774.701+0.224
Memory %1.381.380
JAX0.9.20.9.2—

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

main  ·  7f7822822  ·  main_20260417_125726  ·  full log 
XPK Start: Fri Apr 17 14:40:40 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-17 14:41:04.589682: 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)
I0417 14:41:04.768588 138766995720000 max_utils.py:273] Attempting to initialize the jax distributed system...
I0417 14:41:13.809290 138766995720000 distributed.py:149] Starting JAX distributed service on [::]:8482
I0417 14:41:13.814925 138766995720000 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-t1cme-slice-job-0-0.mt-15-ocdbt-false-sv-t1cme:8482
I0417 14:41:15.073329 138766995720000 max_utils.py:284] Jax distributed system initialized!
I0417 14:41:21.109717 138766995720000 max_utils.py:800] System Information: Jax Version: 0.9.2
I0417 14:41:21.109817 138766995720000 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0417 14:41:21.109876 138766995720000 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
I0417 14:41:21.109912 138766995720000 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0417 14:41:22.152187 138766995720000 maxtext_utils.py:1548] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 14:41:22.152466 138766995720000 checkpointing.py:677] Setting up checkpoint logger...
I0417 14:41:22.152517 138766995720000 checkpointing.py:233] Creating checkpoint manager with ocdbt=False and zarr3=True
I0417 14:41:22.152561 138766995720000 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0417 14:41:22.152911 138766995720000 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 0x7e347db50b00>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0417 14:41:24.659680 138766995720000 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0417 14:41:24.659924 138766995720000 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e20381421b0>}, handler_registry=None
I0417 14:41:24.660162 138766995720000 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e20381421b0>` 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`.
I0417 14:41:24.660211 138766995720000 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e337171c410>` 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`.
I0417 14:41:24.660247 138766995720000 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 0x7e20381421b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e20381421b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e337171c410>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e337171c410>}).
I0417 14:41:24.660557 138766995720000 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0417 14:41:24.660624 138766995720000 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e3371968d60> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0417 14:41:25.964010 138766995720000 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints
I0417 14:41:26.428918 138766995720000 checkpoint_manager.py:929] [process=6][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_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e33716c8d10>
I0417 14:41:26.429085 138766995720000 checkpointing.py:301] Checkpoint manager created!
I0417 14:41:27.363312 138766995720000 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0417 14:41:27.363421 138766995720000 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 14:41:27.745445 138766995720000 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 14:41:27.745537 138766995720000 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 14:41:27.761785 138766995720000 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 14:41:27.761856 138766995720000 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 14:41:27.785401 138766995720000 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 14:41:27.785466 138766995720000 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 14:41:27.801888 138766995720000 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 14:41:27.801947 138766995720000 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 14:41:27.818103 138766995720000 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 14:41:27.818162 138766995720000 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 14:41:27.842741 138766995720000 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0417 14:41:27.842817 138766995720000 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 14:41:27.863687 138766995720000 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0417 14:41:27.863751 138766995720000 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0417 14:41:28.080990 138766995720000 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0417 14:41:28.081099 138766995720000 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0417 14:41:29.509171 138766995720000 maxtext_utils.py:1651]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0417 14:41:29.509298 138766995720000 maxtext_utils.py:1651]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 14:41:29.509353 138766995720000 maxtext_utils.py:1651]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 14:41:29.509407 138766995720000 maxtext_utils.py:1651]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0417 14:41:29.509488 138766995720000 maxtext_utils.py:1651]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 14:41:29.509542 138766995720000 maxtext_utils.py:1651]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 14:41:29.509597 138766995720000 maxtext_utils.py:1651]  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)
I0417 14:41:29.509654 138766995720000 maxtext_utils.py:1651]  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')
I0417 14:41:29.509693 138766995720000 maxtext_utils.py:1651]  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)
I0417 14:41:29.509729 138766995720000 maxtext_utils.py:1651]  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)
I0417 14:41:29.509774 138766995720000 maxtext_utils.py:1651]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0417 14:41:29.509845 138766995720000 maxtext_utils.py:1651]  params/params/token_embedder/embedding

    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0417 14:41:30.000048 138766995720000 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 14:41:30.000140 138766995720000 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 14:41:30.015516 138766995720000 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 14:41:30.015581 138766995720000 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 14:41:40.893399 138766995720000 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.
I0417 14:41:40.894142 138766995720000 metric_logger.py:301] number parameters: 1.104 billion
I0417 14:41:52.112492 138766995720000 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0417 14:41:52.264868 138766995720000 checkpointing.py:776] Waited 0.15234160423278809 seconds for step 0 to finish before starting checkpointing.
I0417 14:41:52.267365 138766995720000 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 14:41:52.269361 138766995720000 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0417 14:41:52.270945 138766995720000 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/0.
I0417 14:41:52.997176 138766995720000 signaling_client.py:364] Using JaxDistributedSignalingClient
I0417 14:41:52.998108 138766995720000 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 14:41:52.998165 138766995720000 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
I0417 14:41:53.271600 138766995720000 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274501s
I0417 14:41:53.271778 138766995720000 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.500 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2804756164550781 s) (per-host)
I0417 14:41:53.271840 138766995720000 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.280548s (batch_requests_ready=0.002351s, total_serialization_initiated=0.278116s, others=0.000080s)
I0417 14:41:53.271932 138766995720000 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.284612s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=0.284592)
I0417 14:41:53.272663 138766995720000 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.00 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/0.
I0417 14:41:53.273006 138637769430784 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 15:01:53.272968
I0417 14:41:53.283648 138766995720000 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 14:41:53.283950 138634028119808 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 14:41:53.284110 138766995720000 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776436912.2673478, 'wait_for_prev_duration_secs': 6.198883056640625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776436912.2693994, 'checkpointer_blocking_duration_secs': 1.0037121772766113, 'get_old_steps_start_time': 1776436913.273135, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776436912.265276, 'checkpoint_manager_blocking_duration_secs': 1.0187945365905762}
I0417 14:41:53.284215 138766995720000 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0417 14:41:53.284266 138766995720000 max_utils.py:750] 
Memstats: After params initialized:
I0417 14:41:53.284315 138766995720000 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0417 14:41:53.284347 138766995720000 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0417 14:41:53.284373 138766995720000 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0417 14:41:53.284398 138766995720000 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0417 14:41:53.596275 138766995720000 metric_logger.py:196] completed step: 0, seconds: 11.218, TFLOP/s/device: 1.211, Tokens/s/device: 182.560, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0417 14:41:53.766290 138766995720000 metric_logger.py:196] completed step: 1, seconds: 1.482, TFLOP/s/device: 9.166, Tokens/s/device: 1381.653, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0417 14:41:54.156874 138766995720000 metric_logger.py:196] completed step: 2, seconds: 0.023, TFLOP/s/device: 595.639, Tokens/s/device: 89781.246, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719
I0417 14:41:54.304244 138766995720000 metric_logger.py:196] completed step: 3, seconds: 0.391, TFLOP/s/device: 34.774, Tokens/s/device: 5241.525, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506
I0417 14:41:54.308584 138766995720000 checkpointing.py:772] Waiting for step 5 to finish before checkpoint...
I0417 14:41:54.598308 138766995720000 checkpointing.py:776] Waited 0.2896864414215088 seconds for step 5 to finish before starting checkpointing.
I0417 14:41:54.601174 138766995720000 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 14:41:56.244004    2832 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0417 14:41:58.491591 138635044902656 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_6
I0417 14:42:03.267443 138637769430784 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 153.716 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.27610182762146 s) (per-host)
I0417 14:42:03.267574 138637769430784 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 9.993927s.
I0417 14:42:09.418234 138637769430784 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 16.144571s.
I0417 14:42:09.418511 138634028119808 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 14:42:09.418639 138634028119808 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 14:42:09.418689 138634028119808 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0417 14:42:09.420356 138634028119808 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0417 14:42:09.420543 138766995720000 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0417 14:42:09.420678 138766995720000 checkpoint_manager.py:1452] Waiting for previous save to complete took 14.819500 seconds. If this number is high, consider checkpointing less frequently.
I0417 14:42:09.422741 138766995720000 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5
I0417 14:42:09.425032 138766995720000 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/5.
I0417 14:42:09.707449 138766995720000 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 14:42:09.707541 138766995720000 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
I0417 14:42:09.750710 138766995720000 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.044203s
I0417 14:42:09.750895 138766995720000 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 32.354 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04767799377441406 s) (per-host)
I0417 14:42:09.750950 138766995720000 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.047742s (batch_requests_ready=0.001866s, total_serialization_initiated=0.045805s, others=0.000071s)
I0417 14:42:09.751042 138766995720000 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.052623s (all_items=0.000016s, per_item={'items': '0.00001645'}, temp_paths=0.052607)
I0417 14:42:09.751802 138766995720000 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/5.
I0417 14:42:09.752120 138634028119808 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 15:02:09.752084
I0417 14:42:09.754126 138766995720000 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 14:42:09.754359 138637752645376 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 14:42:09.754462 138766995720000 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776436914.6011465, 'wait_for_prev_duration_secs': 14.81950044631958, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776436929.4227855, 'checkpointer_blocking_duration_secs': 0.3294847011566162, 'get_old_steps_start_time': 1776436929.7522938, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776436914.598971, 'checkpoint_manager_blocking_duration_secs': 15.155457019805908}
I0417 14:42:09.754621 138766995720000 checkpointing.py:408] Started an asynchronous checkpoint save for step 5
I0417 14:42:09.755433 138766995720000 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.825, Tokens/s/device: 13388.684, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355
I0417 14:42:09.761381 138766995720000 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.298, Tokens/s/device: 13912.098, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467
I0417 14:42:14.692293 138635044902656 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_6
I0417 14:42:20.190516 138634028119808 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 150.621 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.487210035324097 s) (per-host)
I0417 14:42:20.190663 138634028119808 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.438421s.
I0417 14:42:27.106611 138634028119808 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 17.354355s.
I0417 14:42:27.106854 138637752645376 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 14:42:27.106949 138637752645376 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 14:42:27.107013 138637752645376 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts...
I0417 14:42:27.108971 138637752645376 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts.
I0417 14:42:32.254137 138766995720000 metric_logger.py:196] completed step: 6, seconds: 15.451, TFLOP/s/device: 0.879, Tokens/s/device: 132.545, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740
I0417 14:42:32.401558 138766995720000 metric_logger.py:196] completed step: 7, seconds: 22.347, TFLOP/s/device: 0.608, Tokens/s/device: 91.647, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151
I0417 14:42:32.548875 138766995720000 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.375, Tokens/s/device: 13471.646, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098
I0417 14:42:32.695574 138766995720000 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0417 14:42:32.696271 138766995720000 checkpointing.py:776] Waited 0.0007200241088867188 seconds for step 9 to finish before starting checkpointing.
I0417 14:42:32.698945 138766995720000 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 14:42:32.700553 138766995720000 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0417 14:42:32.701921 138766995720000 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/9.
I0417 14:42:33.004989 138766995720000 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 14:42:33.005082 138766995720000 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
I0417 14:42:33.045332 138766995720000 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041157s
I0417 14:42:33.045497 138766995720000 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 34.533 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04467010498046875 s) (per-host)
I0417 14:42:33.045547 138766995720000 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044729s (batch_requests_ready=0.001741s, total_serialization_initiated=0.042924s, others=0.000063s)
I0417 14:42:33.045643 138766995720000 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048783s (all_items=0.000010s, per_item={'items': '0.00001049'}, temp_paths=0.048773)
I0417 14:42:33.046352 138766995720000 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/9.
I0417 14:42:33.046672 138637752645376 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 15:02:33.046636
I0417 14:42:33.048642 138766995720000 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 14:42:33.048882 138616235845376 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 14:42:33.049005 138766995720000 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776436952.6989155, 'wait_for_prev_duration_secs': 7.2479248046875e-05, 'time_between_consecutive_saves_sec': 5.589905261993408, 'checkpointer_blocking_start_time': 1776436952.7005956, 'checkpointer_blocking_duration_secs': 0.34621381759643555, 'get_old_steps_start_time': 1776436953.0468378, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1776436952.6964998, 'checkpoint_manager_blocking_duration_secs': 0.35247325897216797}
I0417 14:42:33.049118 138766995720000 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0417 14:42:33.049166 138766995720000 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 14:42:38.749496 138616428779264 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_6
I0417 14:42:43.964496 138637752645376 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 144.076 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.963629722595215 s) (per-host)
I0417 14:42:43.964631 138637752645376 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.917851s.
I0417 14:42:49.374224 138637752645376 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 16.327429s.
I0417 14:42:49.374531 138616235845376 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 14:42:49.374658 138616235845376 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 14:42:49.374701 138616235845376 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0417 14:42:49.376147 138616235845376 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0417 14:42:49.376333 138766995720000 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0417 14:42:49.376482 138766995720000 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 14:42:49.377438 138766995720000 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.201, Tokens/s/device: 13897.465, 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: Fri Apr 17 14:43:01 UTC 2026
EXIT_CODE=0
feat/nnx-post-train-fixes  ·  2fbc8c2f1  ·  feat_nnx_post_train_fixes_20260417_023710  ·  full log 
XPK Start: Fri Apr 17 04:10:43 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-17 04:11:07.885519: 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)
I0417 04:11:08.063299 138967777109824 max_utils.py:273] Attempting to initialize the jax distributed system...
I0417 04:11:17.104900 138967777109824 distributed.py:149] Starting JAX distributed service on [::]:8482
I0417 04:11:17.107190 138967777109824 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-edpmg-slice-job-0-0.mt-15-ocdbt-false-sv-edpmg:8482
I0417 04:11:18.249209 138967777109824 max_utils.py:284] Jax distributed system initialized!
I0417 04:11:24.481379 138967777109824 max_utils.py:800] System Information: Jax Version: 0.9.2
I0417 04:11:24.481482 138967777109824 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0417 04:11:24.481521 138967777109824 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
I0417 04:11:24.481555 138967777109824 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0417 04:11:25.162989 138967777109824 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 04:11:25.163564 138967777109824 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 04:11:25.163748 138967777109824 checkpointing.py:688] Setting up checkpoint logger...
I0417 04:11:25.163797 138967777109824 checkpointing.py:234] Creating checkpoint manager with ocdbt=False and zarr3=True
I0417 04:11:25.163851 138967777109824 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0417 04:11:25.164179 138967777109824 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 0x7e633d534140>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0417 04:11:28.551488 138967777109824 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0417 04:11:28.551728 138967777109824 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4f7c2e3bf0>}, handler_registry=None
I0417 04:11:28.551979 138967777109824 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4f7c2e3bf0>` 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`.
I0417 04:11:28.552028 138967777109824 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4f7c2e7da0>` 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`.
I0417 04:11:28.552065 138967777109824 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 0x7e4f7c2e3bf0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4f7c2e3bf0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4f7c2e7da0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4f7c2e7da0>}).
I0417 04:11:28.552386 138967777109824 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0417 04:11:28.552456 138967777109824 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e4f647cc7c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0417 04:11:31.284612 138967777109824 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints
I0417 04:11:31.738005 138967777109824 checkpoint_manager.py:929] [process=6][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_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e4f7c44a3f0>
I0417 04:11:31.738177 138967777109824 checkpointing.py:302] Checkpoint manager created!
I0417 04:11:32.666146 138967777109824 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0417 04:11:32.666253 138967777109824 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 04:11:33.047747 138967777109824 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 04:11:33.047846 138967777109824 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 04:11:33.064203 138967777109824 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 04:11:33.064261 138967777109824 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 04:11:33.087953 138967777109824 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 04:11:33.088021 138967777109824 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 04:11:33.104372 138967777109824 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 04:11:33.104429 138967777109824 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 04:11:33.120736 138967777109824 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 04:11:33.120793 138967777109824 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 04:11:33.145416 138967777109824 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0417 04:11:33.145487 138967777109824 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 04:11:33.166934 138967777109824 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0417 04:11:33.166996 138967777109824 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0417 04:11:33.379423 138967777109824 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0417 04:11:33.379528 138967777109824 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0417 04:11:34.799221 138967777109824 maxtext_utils.py:1796]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0417 04:11:34.799347 138967777109824 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 04:11:34.799399 138967777109824 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 04:11:34.799457 138967777109824 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0417 04:11:34.799509 138967777109824 maxtext_utils.py:1796]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 04:11:34.799547 138967777109824 maxtext_utils.py:1796]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 04:11:34.799601 138967777109824 maxtext_utils.py:1796]  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)
I0417 04:11:34.799651 138967777109824 maxtext_utils.py:1796]  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')
I0417 04:11:34.799692 138967777109824 maxtext_utils.py:1796]  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)
I0417 04:11:34.799728 138967777109824 maxtext_utils.py:1796]  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)
I0417 04:11:34.799776 138967777109824 maxtext_utils.py:1796]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0417 04:11:34.799841 138967777109824 maxtext_utils.py:1796]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0417 04:11:35.290681 138967777109824 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 04:11:35.290772 138967777109824 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 04:11:35.306513 138967777109824 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 04:11:35.306579 138967777109824 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 04:11:46.468322 138967777109824 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.
I0417 04:11:46.469192 138967777109824 metric_logger.py:289] number parameters: 1.104 billion
I0417 04:11:57.791619 138967777109824 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0417 04:11:57.944831 138967777109824 checkpointing.py:798] Waited 0.15317010879516602 seconds for step 0 to finish before starting checkpointing.
I0417 04:11:57.947464 138967777109824 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 04:11:57.949076 138967777109824 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0417 04:11:57.950525 138967777109824 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/0.
I0417 04:11:58.696605 138967777109824 signaling_client.py:364] Using JaxDistributedSignalingClient
I0417 04:11:58.697613 138967777109824 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 04:11:58.697669 138967777109824 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
I0417 04:11:58.975924 138967777109824 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279391s
I0417 04:11:58.976098 138967777109824 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.418 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2847330570220947 s) (per-host)
I0417 04:11:58.976153 138967777109824 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.284797s (batch_requests_ready=0.002164s, total_serialization_initiated=0.282561s, others=0.000072s)
I0417 04:11:58.976248 138967777109824 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.289013s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.288994)
I0417 04:11:58.977039 138967777109824 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.03 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/0.
I0417 04:11:58.977365 138840497649408 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 04:31:58.977329
I0417 04:11:58.983627 138967777109824 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 04:11:58.983911 138837782853376 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 04:11:58.984058 138967777109824 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776399117.9474452, 'wait_for_prev_duration_secs': 6.103515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776399117.949116, 'checkpointer_blocking_duration_secs': 1.028404712677002, 'get_old_steps_start_time': 1776399118.9775457, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1776399117.9453363, 'checkpoint_manager_blocking_duration_secs': 1.038680076599121}
I0417 04:11:58.984227 138967777109824 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0417 04:11:58.984279 138967777109824 max_utils.py:750] 
Memstats: After params initialized:
I0417 04:11:58.984327 138967777109824 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0417 04:11:58.984359 138967777109824 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0417 04:11:58.984387 138967777109824 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0417 04:11:58.984412 138967777109824 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0417 04:11:59.296235 138967777109824 metric_logger.py:185] completed step: 0, seconds: 11.322, TFLOP/s/device: 1.200, Tokens/s/device: 180.882, total_weights: 65536, loss: 10.874
I0417 04:11:59.481142 138967777109824 metric_logger.py:185] completed step: 1, seconds: 1.503, TFLOP/s/device: 9.038, Tokens/s/device: 1362.269, total_weights: 65536, loss: 10.874
I0417 04:11:59.891276 138967777109824 metric_logger.py:185] completed step: 2, seconds: 0.038, TFLOP/s/device: 360.077, Tokens/s/device: 54274.659, total_weights: 65536, loss: 10.560
I0417 04:12:00.038671 138967777109824 metric_logger.py:185] completed step: 3, seconds: 0.410, TFLOP/s/device: 33.108, Tokens/s/device: 4990.399, total_weights: 65536, loss: 9.980
I0417 04:12:00.043261 138967777109824 checkpointing.py:794] Waiting for step 5 to finish before checkpoint...
I0417 04:12:00.332992 138967777109824 checkpointing.py:798] Waited 0.28969883918762207 seconds for step 5 to finish before starting checkpointing.
I0417 04:12:00.335933 138967777109824 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 04:12:02.792108    2807 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0417 04:12:05.732255 138837791246080 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_6
I0417 04:12:10.218732 138840497649408 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 137.031 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.527328491210938 s) (per-host)
I0417 04:12:10.218864 138840497649408 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.241386s.
I0417 04:12:17.227772 138840497649408 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 18.250278s.
I0417 04:12:17.228076 138837782853376 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 04:12:17.228210 138837782853376 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 04:12:17.228260 138837782853376 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0417 04:12:17.230314 138837782853376 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0417 04:12:17.230543 138967777109824 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0417 04:12:17.230678 138967777109824 checkpoint_manager.py:1452] Waiting for previous save to complete took 16.894744 seconds. If this number is high, consider checkpointing less frequently.
I0417 04:12:17.232525 138967777109824 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5
I0417 04:12:17.234951 138967777109824 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/5.
I0417 04:12:17.520714 138967777109824 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 04:12:17.520805 138967777109824 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
I0417 04:12:17.554634 138967777109824 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035161s
I0417 04:12:17.554789 138967777109824 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 40.056 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03851056098937988 s) (per-host)
I0417 04:12:17.554853 138967777109824 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038582s (batch_requests_ready=0.001745s, total_serialization_initiated=0.036758s, others=0.000079s)
I0417 04:12:17.554941 138967777109824 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043509s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=0.043495)
I0417 04:12:17.555629 138967777109824 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/5.
I0417 04:12:17.555915 138837782853376 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 04:32:17.555890
I0417 04:12:17.565148 138967777109824 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 04:12:17.565400 138839985223424 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 04:12:17.565559 138967777109824 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776399120.3359032, 'wait_for_prev_duration_secs': 16.894744157791138, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776399137.232564, 'checkpointer_blocking_duration_secs': 0.3234539031982422, 'get_old_steps_start_time': 1776399137.5560412, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776399120.333596, 'checkpoint_manager_blocking_duration_secs': 17.23192834854126}
I0417 04:12:17.565678 138967777109824 checkpointing.py:409] Started an asynchronous checkpoint save for step 5
I0417 04:12:17.566256 138967777109824 metric_logger.py:185] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.932, Tokens/s/device: 13404.808, total_weights: 65536, loss: 9.460
I0417 04:12:17.571866 138967777109824 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.270, Tokens/s/device: 13907.941, total_weights: 65536, loss: 8.959
I0417 04:12:23.411878 138837791246080 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_6
I0417 04:12:29.047161 138837782853376 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 136.989 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.530832290649414 s) (per-host)
I0417 04:12:29.047278 138837782853376 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.491283s.
I0417 04:12:36.174468 138837782853376 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 18.618457s.
I0417 04:12:36.174769 138839985223424 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 04:12:36.174942 138839985223424 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 04:12:36.174997 138839985223424 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts...
I0417 04:12:36.177684 138839985223424 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts.
I0417 04:12:39.947576 138967777109824 metric_logger.py:185] completed step: 6, seconds: 17.527, TFLOP/s/device: 0.775, Tokens/s/device: 116.846, total_weights: 65536, loss: 8.469
I0417 04:12:40.094947 138967777109824 metric_logger.py:185] completed step: 7, seconds: 22.230, TFLOP/s/device: 0.611, Tokens/s/device: 92.129, total_weights: 65536, loss: 8.003
I0417 04:12:40.242676 138967777109824 metric_logger.py:185] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.476, Tokens/s/device: 13486.727, total_weights: 65536, loss: 7.572
I0417 04:12:40.389899 138967777109824 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0417 04:12:40.390558 138967777109824 checkpointing.py:798] Waited 0.0006887912750244141 seconds for step 9 to finish before starting checkpointing.
I0417 04:12:40.392505 138967777109824 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 04:12:40.394176 138967777109824 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0417 04:12:40.395563 138967777109824 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/9.
I0417 04:12:41.102972 138967777109824 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 04:12:41.103063 138967777109824 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
I0417 04:12:41.137221 138967777109824 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035311s
I0417 04:12:41.137388 138967777109824 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 39.845 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03871417045593262 s) (per-host)
I0417 04:12:41.137437 138967777109824 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038774s (batch_requests_ready=0.001631s, total_serialization_initiated=0.037078s, others=0.000065s)
I0417 04:12:41.137525 138967777109824 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043044s (all_items=0.000011s, per_item={'items': '0.00001144'}, temp_paths=0.043032)
I0417 04:12:41.138180 138967777109824 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/9.
I0417 04:12:41.138509 138839985223424 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 04:32:41.138473
I0417 04:12:41.146571 138967777109824 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 04:12:41.146880 138838330504960 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 04:12:41.147040 138967777109824 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776399160.3924744, 'wait_for_prev_duration_secs': 7.176399230957031e-05, 'time_between_consecutive_saves_sec': 4.21474814414978, 'checkpointer_blocking_start_time': 1776399160.3942149, 'checkpointer_blocking_duration_secs': 0.744434118270874, 'get_old_steps_start_time': 1776399161.1386664, 'get_old_steps_duration_secs': 2.47955322265625e-05, 'checkpoint_manager_blocking_start_time': 1776399160.3907788, 'checkpoint_manager_blocking_duration_secs': 0.7562272548675537}
I0417 04:12:41.147153 138967777109824 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0417 04:12:41.147200 138967777109824 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 04:12:46.612717 138837791246080 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_6
I0417 04:12:52.244771 138839985223424 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 141.718 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.146058797836304 s) (per-host)
I0417 04:12:52.244910 138839985223424 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.106290s.
I0417 04:13:00.869072 138839985223424 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 19.730436s.
I0417 04:13:00.869366 138838330504960 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 04:13:00.869481 138838330504960 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 04:13:00.869524 138838330504960 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0417 04:13:00.871179 138838330504960 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0417 04:13:00.871367 138967777109824 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0417 04:13:00.871520 138967777109824 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 04:13:00.872255 138967777109824 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.169, Tokens/s/device: 13892.752, total_weights: 65536, loss: 7.181
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 17 04:13:09 UTC 2026
EXIT_CODE=0