MaxView

Timing Test Reference Updates
‹ 16_zarr3_true_resumeCase: 16_zarr3_true_save— ›

Metrics: main (b117f50cf) vs feat/nnx-post-train-fixes (28b1e4a16)

Metricmain  b117f50cffeat/nnx-post-train-fixes  28b1e4a16Diff (feat/nnx-post-train-fixes − main)
Parameters1.104 billion1.104 billion—
Final loss7.18107.18100
TFLOP/s92.06392.028-0.035
Tok/s13876.713871.5-5.263
Avg s/step7.6567.771+0.115
Memory %1.381.380
JAX0.9.20.9.2—

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

main  ·  b117f50cf  ·  main_20260424_070227  ·  full log 
XPK Start: Fri Apr 24 08:16:22 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. 
2026-04-24 08:16:47.427435: 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)
I0424 08:16:47.632542 137695244932928 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 08:16:56.673437 137695244932928 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 08:16:56.676238 137695244932928 distributed.py:172] Connecting to JAX distributed service on mt-16-zarr3-true-sv-zax37-slice-job-0-0.mt-16-zarr3-true-sv-zax37:8482
I0424 08:16:57.691249 137695244932928 max_utils.py:284] Jax distributed system initialized!
I0424 08:17:03.930434 137695244932928 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 08:17:03.930544 137695244932928 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 08:17:03.930586 137695244932928 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
I0424 08:17:03.930624 137695244932928 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 08:17:04.628371 137695244932928 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 08:17:04.628655 137695244932928 checkpointing.py:677] Setting up checkpoint logger...
I0424 08:17:04.628710 137695244932928 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 08:17:04.628756 137695244932928 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 08:17:04.629118 137695244932928 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7d3af4650080>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 08:17:07.958366 137695244932928 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0424 08:17:07.958607 137695244932928 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 0x7d25d8208920>}, handler_registry=None
I0424 08:17:07.958844 137695244932928 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d25d8208920>` 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`.
I0424 08:17:07.958892 137695244932928 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d26b87e8710>` 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`.
I0424 08:17:07.958928 137695244932928 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 0x7d25d8208920>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d25d8208920>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d26b87e8710>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d26b87e8710>}).
I0424 08:17:07.959264 137695244932928 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 08:17:07.959337 137695244932928 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d26b81c1a80> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 08:17:08.665066 137695244932928 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints
I0424 08:17:08.672685 137695244932928 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_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d25d8208410>
I0424 08:17:08.672802 137695244932928 checkpointing.py:301] Checkpoint manager created!
I0424 08:17:09.603585 137695244932928 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 08:17:09.603703 137695244932928 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:17:09.986272 137695244932928 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 08:17:09.986365 137695244932928 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:17:10.003666 137695244932928 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 08:17:10.003726 137695244932928 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:17:10.028550 137695244932928 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 08:17:10.028620 137695244932928 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:17:10.045938 137695244932928 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 08:17:10.045999 137695244932928 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:17:10.063263 137695244932928 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 08:17:10.063329 137695244932928 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:17:10.089174 137695244932928 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 08:17:10.089251 137695244932928 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:17:10.110493 137695244932928 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 08:17:10.110558 137695244932928 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 08:17:10.324801 137695244932928 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 08:17:10.324910 137695244932928 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 08:17:11.754591 137695244932928 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 08:17:11.754717 137695244932928 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 08:17:11.754772 137695244932928 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 08:17:11.754831 137695244932928 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0424 08:17:11.754884 137695244932928 maxtext_utils.py:1707]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 08:17:11.754925 137695244932928 maxtext_utils.py:1707]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 08:17:11.754985 137695244932928 maxtext_utils.py:1707]  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)
I0424 08:17:11.755039 137695244932928 maxtext_utils.py:1707]  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')
I0424 08:17:11.755080 137695244932928 maxtext_utils.py:1707]  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)
I0424 08:17:11.755131 137695244932928 maxtext_utils.py:1707]  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)
I0424 08:17:11.755181 137695244932928 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0424 08:17:11.755235 137695244932928 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0424 08:17:12.260963 137695244932928 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 08:17:12.261058 137695244932928 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 08:17:12.277005 137695244932928 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 08:17:12.277065 137695244932928 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 08:17:23.330502 137695244932928 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.
I0424 08:17:23.331327 137695244932928 metric_logger.py:301] number parameters: 1.104 billion
I0424 08:17:35.182658 137695244932928 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0424 08:17:35.335709 137695244932928 checkpointing.py:776] Waited 0.15303254127502441 seconds for step 0 to finish before starting checkpointing.
I0424 08:17:35.338352 137695244932928 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 08:17:35.340042 137695244932928 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0424 08:17:35.341477 137695244932928 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/0.
I0424 08:17:35.665466 137695244932928 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 08:17:35.666465 137695244932928 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:17:35.666527 137695244932928 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
I0424 08:17:35.941931 137695244932928 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.276539s
I0424 08:17:35.942131 137695244932928 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.465 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28226757049560547 s) (per-host)
I0424 08:17:35.942189 137695244932928 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.282354s (batch_requests_ready=0.002399s, total_serialization_initiated=0.279860s, others=0.000094s)
I0424 08:17:35.942289 137695244932928 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.286440s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.286422)
I0424 08:17:35.943114 137695244932928 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.60 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/0.
I0424 08:17:35.943446 137566232258304 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:37:35.943407
I0424 08:17:35.949406 137695244932928 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:17:35.949643 137564645316352 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:17:35.949767 137695244932928 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777018655.3383327, 'wait_for_prev_duration_secs': 5.984306335449219e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777018655.3400822, 'checkpointer_blocking_duration_secs': 0.6035282611846924, 'get_old_steps_start_time': 1777018655.9436362, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1777018655.3362746, 'checkpoint_manager_blocking_duration_secs': 0.6134557723999023}
I0424 08:17:35.949873 137695244932928 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0424 08:17:35.949919 137695244932928 max_utils.py:750] 
Memstats: After params initialized:
I0424 08:17:35.949965 137695244932928 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0424 08:17:35.949996 137695244932928 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0424 08:17:35.950019 137695244932928 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0424 08:17:35.950043 137695244932928 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0424 08:17:36.263813 137695244932928 metric_logger.py:196] completed step: 0, seconds: 11.851, TFLOP/s/device: 1.146, Tokens/s/device: 172.809, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 08:17:36.452279 137695244932928 metric_logger.py:196] completed step: 1, seconds: 1.080, TFLOP/s/device: 12.585, Tokens/s/device: 1896.999, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 08:17:36.864760 137695244932928 metric_logger.py:196] completed step: 2, seconds: 0.041, TFLOP/s/device: 329.146, Tokens/s/device: 49612.403, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719
I0424 08:17:37.011946 137695244932928 metric_logger.py:196] completed step: 3, seconds: 0.413, TFLOP/s/device: 32.922, Tokens/s/device: 4962.406, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506
I0424 08:17:37.017131 137695244932928 checkpointing.py:772] Waiting for step 5 to finish before checkpoint...
I0424 08:17:37.305899 137695244932928 checkpointing.py:776] Waited 0.28873229026794434 seconds for step 5 to finish before starting checkpointing.
I0424 08:17:37.308892 137695244932928 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:17:39.189072    2786 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 08:17:41.343055 137565155878656 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/0/items/array_metadatas/process_6
I0424 08:18:13.550377 137566232258304 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 41.688 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.8904914855957 s) (per-host)
I0424 08:18:13.550498 137566232258304 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.606931s.
I0424 08:18:21.306421 137566232258304 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 45.362838s.
I0424 08:18:21.306650 137564645316352 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:18:21.306725 137564645316352 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:18:21.306785 137564645316352 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:18:21.308459 137564645316352 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 08:18:21.308651 137695244932928 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.
W0424 08:18:21.308786 137695244932928 checkpoint_manager.py:1452] Waiting for previous save to complete took 43.999893 seconds. If this number is high, consider checkpointing less frequently.
I0424 08:18:21.310889 137695244932928 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5
I0424 08:18:21.313463 137695244932928 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/5.
I0424 08:18:21.607299 137695244932928 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:18:21.607393 137695244932928 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
I0424 08:18:21.644263 137695244932928 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038080s
I0424 08:18:21.644405 137695244932928 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 37.037 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04164910316467285 s) (per-host)
I0424 08:18:21.644453 137695244932928 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.041705s (batch_requests_ready=0.001931s, total_serialization_initiated=0.039711s, others=0.000063s)
I0424 08:18:21.644532 137695244932928 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.046622s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.046607)
I0424 08:18:21.645179 137695244932928 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/5.
I0424 08:18:21.645509 137564645316352 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:38:21.645471
I0424 08:18:21.650725 137695244932928 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:18:21.651010 137564591904512 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:18:21.651162 137695244932928 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777018657.3088632, 'wait_for_prev_duration_secs': 43.99989342689514, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777018701.310928, 'checkpointer_blocking_duration_secs': 0.334730863571167, 'get_old_steps_start_time': 1777018701.6456826, 'get_old_steps_duration_secs': 3.075599670410156e-05, 'checkpoint_manager_blocking_start_time': 1777018657.306568, 'checkpoint_manager_blocking_duration_secs': 44.344560623168945}
I0424 08:18:21.651326 137695244932928 checkpointing.py:408] Started an asynchronous checkpoint save for step 5
I0424 08:18:21.652152 137695244932928 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.833, Tokens/s/device: 13389.822, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355
I0424 08:18:21.658349 137695244932928 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 91.919, Tokens/s/device: 13855.063, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467
I0424 08:18:26.861124 137565155878656 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/5/items/array_metadatas/process_6
I0424 08:18:43.929287 137695244932928 metric_logger.py:196] completed step: 6, seconds: 44.640, TFLOP/s/device: 0.304, Tokens/s/device: 45.878, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740
I0424 08:18:44.076942 137695244932928 metric_logger.py:196] completed step: 7, seconds: 22.125, TFLOP/s/device: 0.614, Tokens/s/device: 92.565, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151
I0424 08:18:44.224131 137695244932928 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.224, Tokens/s/device: 13448.822, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098
I0424 08:18:44.370766 137695244932928 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0424 08:18:44.371489 137695244932928 checkpointing.py:776] Waited 0.0007452964782714844 seconds for step 9 to finish before starting checkpointing.
I0424 08:18:44.373587 137695244932928 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:19:04.039371 137564645316352 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.223 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.43657398223877 s) (per-host)
I0424 08:19:04.039499 137564645316352 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.393872s.
I0424 08:19:10.670994 137564645316352 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.025350s.
I0424 08:19:10.671291 137564591904512 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:19:10.671409 137564591904512 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:19:10.671458 137564591904512 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:19:10.673514 137564591904512 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts.
I0424 08:19:10.673695 137695244932928 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5.
W0424 08:19:10.673817 137695244932928 checkpoint_manager.py:1452] Waiting for previous save to complete took 26.300234 seconds. If this number is high, consider checkpointing less frequently.
I0424 08:19:10.675322 137695244932928 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0424 08:19:10.677398 137695244932928 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/9.
I0424 08:19:10.969194 137695244932928 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:19:10.969286 137695244932928 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
I0424 08:19:11.003648 137695244932928 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035270s
I0424 08:19:11.003820 137695244932928 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 39.606 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03894805908203125 s) (per-host)
I0424 08:19:11.003873 137695244932928 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.039011s (batch_requests_ready=0.001811s, total_serialization_initiated=0.037133s, others=0.000068s)
I0424 08:19:11.003966 137695244932928 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043061s (all_items=0.000010s, per_item={'items': '0.00001049'}, temp_paths=0.043051)
I0424 08:19:11.004662 137695244932928 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/9.
I0424 08:19:11.004983 137564069537536 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:39:11.004946
I0424 08:19:11.006902 137695244932928 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:19:11.007141 137564591904512 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:19:11.007309 137695244932928 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777018724.3735573, 'wait_for_prev_duration_secs': 26.300233602523804, 'time_between_consecutive_saves_sec': 23.065059185028076, 'checkpointer_blocking_start_time': 1777018750.6753623, 'checkpointer_blocking_duration_secs': 0.3297874927520752, 'get_old_steps_start_time': 1777018751.005167, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1777018724.371722, 'checkpoint_manager_blocking_duration_secs': 26.63555121421814}
I0424 08:19:11.007411 137695244932928 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0424 08:19:11.007452 137695244932928 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:19:16.297291 137565155878656 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/9/items/array_metadatas/process_6
I0424 08:19:53.348717 137564069537536 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.269 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.38380527496338 s) (per-host)
I0424 08:19:53.348842 137564069537536 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.343728s.
I0424 08:19:59.607160 137564069537536 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.602029s.
I0424 08:19:59.607479 137564591904512 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:19:59.607604 137564591904512 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:19:59.607646 137564591904512 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:19:59.608998 137564591904512 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 08:19:59.609203 137695244932928 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.
I0424 08:19:59.609365 137695244932928 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 08:19:59.610217 137695244932928 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.063, Tokens/s/device: 13876.749, 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 24 08:20:12 UTC 2026
EXIT_CODE=0
feat/nnx-post-train-fixes  ·  28b1e4a16  ·  feat_nnx_post_train_fixes_20260424_120657  ·  full log 
XPK Start: Fri Apr 24 13:13:47 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. 
2026-04-24 13:14:11.613794: 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)
I0424 13:14:12.189528 135444366239552 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 13:14:21.230675 135444366239552 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 13:14:21.233128 135444366239552 distributed.py:172] Connecting to JAX distributed service on mt-16-zarr3-true-sv-mj2lf-slice-job-0-0.mt-16-zarr3-true-sv-mj2lf:8482
I0424 13:14:22.259091 135444366239552 max_utils.py:284] Jax distributed system initialized!
I0424 13:14:27.209175 135444366239552 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 13:14:27.209285 135444366239552 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 13:14:27.209328 135444366239552 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
I0424 13:14:27.209366 135444366239552 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 13:14:27.916567 135444366239552 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 13:14:27.917172 135444366239552 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 13:14:27.917363 135444366239552 checkpointing.py:688] Setting up checkpoint logger...
I0424 13:14:27.917415 135444366239552 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 13:14:27.917459 135444366239552 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 13:14:27.917793 135444366239552 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7b2ee1948a10>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 13:14:30.817314 135444366239552 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 13:14:30.817552 135444366239552 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 0x7b1b4063f800>}, handler_registry=None
I0424 13:14:30.817796 135444366239552 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b1b4063f800>` 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`.
I0424 13:14:30.817846 135444366239552 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b1b4044c800>` 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`.
I0424 13:14:30.817882 135444366239552 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 0x7b1b4063f800>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b1b4063f800>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b1b4044c800>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b1b4044c800>}).
I0424 13:14:30.818214 135444366239552 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 13:14:30.818290 135444366239552 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b1b385b9940> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 13:14:31.564633 135444366239552 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints
I0424 13:14:31.571481 135444366239552 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b1b40449a90>
I0424 13:14:31.571597 135444366239552 checkpointing.py:302] Checkpoint manager created!
I0424 13:14:32.509484 135444366239552 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 13:14:32.509590 135444366239552 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:14:32.888522 135444366239552 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 13:14:32.888614 135444366239552 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:14:32.905217 135444366239552 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 13:14:32.905282 135444366239552 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:14:32.929044 135444366239552 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:14:32.929130 135444366239552 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:14:32.945676 135444366239552 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:14:32.945737 135444366239552 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:14:32.962389 135444366239552 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:14:32.962448 135444366239552 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:14:32.987176 135444366239552 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 13:14:32.987248 135444366239552 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:14:33.007978 135444366239552 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 13:14:33.008044 135444366239552 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 13:14:33.224223 135444366239552 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 13:14:33.224335 135444366239552 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 13:14:34.660684 135444366239552 maxtext_utils.py:1880]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 13:14:34.660814 135444366239552 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 13:14:34.660871 135444366239552 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 13:14:34.660931 135444366239552 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0424 13:14:34.660984 135444366239552 maxtext_utils.py:1880]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 13:14:34.661026 135444366239552 maxtext_utils.py:1880]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 13:14:34.661079 135444366239552 maxtext_utils.py:1880]  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)
I0424 13:14:34.661166 135444366239552 maxtext_utils.py:1880]  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')
I0424 13:14:34.661216 135444366239552 maxtext_utils.py:1880]  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)
I0424 13:14:34.661253 135444366239552 maxtext_utils.py:1880]  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)
I0424 13:14:34.661301 135444366239552 maxtext_utils.py:1880]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0424 13:14:34.661352 135444366239552 maxtext_utils.py:1880]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')

    Physical:  (None, 'fsdp')
I0424 13:14:35.157723 135444366239552 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 13:14:35.157819 135444366239552 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 13:14:35.173482 135444366239552 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 13:14:35.173544 135444366239552 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 13:14:46.173603 135444366239552 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.
I0424 13:14:46.174443 135444366239552 metric_logger.py:301] number parameters: 1.104 billion
I0424 13:14:57.600003 135444366239552 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 13:14:57.752918 135444366239552 checkpointing.py:798] Waited 0.15289998054504395 seconds for step 0 to finish before starting checkpointing.
I0424 13:14:57.755447 135444366239552 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 13:14:57.757123 135444366239552 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0424 13:14:57.758703 135444366239552 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/0.
I0424 13:14:58.336176 135444366239552 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 13:14:58.337279 135444366239552 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:14:58.337335 135444366239552 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
I0424 13:14:58.613226 135444366239552 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.277136s
I0424 13:14:58.613400 135444366239552 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.456 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2827157974243164 s) (per-host)
I0424 13:14:58.613452 135444366239552 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.282780s (batch_requests_ready=0.002344s, total_serialization_initiated=0.280364s, others=0.000072s)
I0424 13:14:58.613548 135444366239552 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.286836s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.286818)
I0424 13:14:58.614373 135444366239552 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.86 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/0.
I0424 13:14:58.614701 135318581802752 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:34:58.614663
I0424 13:14:59.044935 135444366239552 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:14:59.045344 135317991642880 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:14:59.045513 135444366239552 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777036497.7554283, 'wait_for_prev_duration_secs': 6.413459777832031e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777036497.757161, 'checkpointer_blocking_duration_secs': 0.8576915264129639, 'get_old_steps_start_time': 1777036498.6148777, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1777036497.7534637, 'checkpoint_manager_blocking_duration_secs': 1.292006254196167}
I0424 13:14:59.045620 135444366239552 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 13:14:59.045674 135444366239552 max_utils.py:750] 
Memstats: After params initialized:
I0424 13:14:59.045733 135444366239552 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0424 13:14:59.045767 135444366239552 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0424 13:14:59.045797 135444366239552 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0424 13:14:59.045824 135444366239552 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0424 13:14:59.359527 135444366239552 metric_logger.py:196] completed step: 0, seconds: 11.425, TFLOP/s/device: 1.189, Tokens/s/device: 179.249, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 13:14:59.541931 135444366239552 metric_logger.py:196] completed step: 1, seconds: 1.758, TFLOP/s/device: 7.729, Tokens/s/device: 1164.967, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 13:14:59.952481 135444366239552 metric_logger.py:196] completed step: 2, seconds: 0.035, TFLOP/s/device: 386.833, Tokens/s/device: 58307.710, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719
I0424 13:15:00.099824 135444366239552 metric_logger.py:196] completed step: 3, seconds: 0.411, TFLOP/s/device: 33.081, Tokens/s/device: 4986.365, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506
I0424 13:15:00.104270 135444366239552 checkpointing.py:794] Waiting for step 5 to finish before checkpoint...
I0424 13:15:00.393862 135444366239552 checkpointing.py:798] Waited 0.28956031799316406 seconds for step 5 to finish before starting checkpointing.
I0424 13:15:00.396762 135444366239552 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:15:02.227346    2740 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 13:15:04.272768 135318000035584 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/0/items/array_metadatas/process_6
I0424 13:15:36.088763 135318581802752 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 41.835 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.75804114341736 s) (per-host)
I0424 13:15:36.088889 135318581802752 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.474075s.
I0424 13:15:44.595315 135318581802752 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 45.980485s.
I0424 13:15:44.595603 135317991642880 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:15:44.595722 135317991642880 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:15:44.595774 135317991642880 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:15:44.597209 135317991642880 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 13:15:44.597389 135444366239552 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.
W0424 13:15:44.597527 135444366239552 checkpoint_manager.py:1452] Waiting for previous save to complete took 44.200759 seconds. If this number is high, consider checkpointing less frequently.
I0424 13:15:44.599553 135444366239552 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5
I0424 13:15:44.602080 135444366239552 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/5.
I0424 13:15:44.891632 135444366239552 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:15:44.891723 135444366239552 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
I0424 13:15:44.931690 135444366239552 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041127s
I0424 13:15:44.931853 135444366239552 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 34.542 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04465842247009277 s) (per-host)
I0424 13:15:44.931904 135444366239552 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044719s (batch_requests_ready=0.001867s, total_serialization_initiated=0.042785s, others=0.000067s)
I0424 13:15:44.931990 135444366239552 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049622s (all_items=0.000031s, per_item={'items': '0.00003099'}, temp_paths=0.049591)
I0424 13:15:44.932728 135444366239552 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/5.
I0424 13:15:44.933084 135317991642880 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:35:44.933047
I0424 13:15:44.936672 135444366239552 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:15:44.936925 135318000035584 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:15:44.937087 135444366239552 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777036500.3967311, 'wait_for_prev_duration_secs': 44.20075869560242, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777036544.5995924, 'checkpointer_blocking_duration_secs': 0.3336527347564697, 'get_old_steps_start_time': 1777036544.933269, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1777036500.3945084, 'checkpoint_manager_blocking_duration_secs': 44.54254102706909}
I0424 13:15:44.937210 135444366239552 checkpointing.py:409] Started an asynchronous checkpoint save for step 5
I0424 13:15:44.938017 135444366239552 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.797, Tokens/s/device: 13384.396, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355
I0424 13:15:44.944090 135444366239552 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.232, Tokens/s/device: 13902.277, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467
I0424 13:15:50.068932 135291176208128 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/5/items/array_metadatas/process_6
I0424 13:16:07.389900 135444366239552 metric_logger.py:196] completed step: 6, seconds: 44.838, TFLOP/s/device: 0.303, Tokens/s/device: 45.675, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740
I0424 13:16:07.537286 135444366239552 metric_logger.py:196] completed step: 7, seconds: 22.300, TFLOP/s/device: 0.609, Tokens/s/device: 91.838, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151
I0424 13:16:07.684619 135444366239552 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.511, Tokens/s/device: 13492.058, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098
I0424 13:16:07.831218 135444366239552 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 13:16:07.831896 135444366239552 checkpointing.py:798] Waited 0.0006985664367675781 seconds for step 9 to finish before starting checkpointing.
I0424 13:16:07.833902 135444366239552 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:16:27.414236 135317991642880 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.143 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.527021169662476 s) (per-host)
I0424 13:16:27.414351 135317991642880 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.481135s.
I0424 13:16:35.096208 135317991642880 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.162976s.
I0424 13:16:35.096502 135318000035584 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:16:35.096624 135318000035584 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:16:35.096673 135318000035584 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:16:35.098261 135318000035584 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts.
I0424 13:16:35.098447 135444366239552 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5.
W0424 13:16:35.098587 135444366239552 checkpoint_manager.py:1452] Waiting for previous save to complete took 27.264691 seconds. If this number is high, consider checkpointing less frequently.
I0424 13:16:35.099982 135444366239552 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0424 13:16:35.102043 135444366239552 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/9.
I0424 13:16:35.406988 135444366239552 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:16:35.407076 135444366239552 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
I0424 13:16:35.446409 135444366239552 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.040256s
I0424 13:16:35.446588 135444366239552 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 35.218 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04380083084106445 s) (per-host)
I0424 13:16:35.446642 135444366239552 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043866s (batch_requests_ready=0.001751s, total_serialization_initiated=0.042045s, others=0.000071s)
I0424 13:16:35.446735 135444366239552 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.047863s (all_items=0.000011s, per_item={'items': '0.00001121'}, temp_paths=0.047852)
I0424 13:16:35.447399 135444366239552 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/9.
I0424 13:16:35.447696 135315325458176 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:36:35.447663
I0424 13:16:35.449679 135444366239552 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:16:35.449943 135318000035584 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:16:35.450111 135444366239552 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777036567.833871, 'wait_for_prev_duration_secs': 27.26469087600708, 'time_between_consecutive_saves_sec': 23.236621618270874, 'checkpointer_blocking_start_time': 1777036595.100025, 'checkpointer_blocking_duration_secs': 0.34781360626220703, 'get_old_steps_start_time': 1777036595.4478557, 'get_old_steps_duration_secs': 2.47955322265625e-05, 'checkpoint_manager_blocking_start_time': 1777036567.8321462, 'checkpoint_manager_blocking_duration_secs': 27.617918491363525}
I0424 13:16:35.450223 135444366239552 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 13:16:35.450265 135444366239552 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:16:40.917250 135291176208128 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/9/items/array_metadatas/process_6
I0424 13:17:18.360750 135315325458176 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.771 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.95792508125305 s) (per-host)
I0424 13:17:18.360877 135315325458176 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.913073s.
I0424 13:17:25.134615 135315325458176 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.686794s.
I0424 13:17:25.134887 135318000035584 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:17:25.135005 135318000035584 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:17:25.135046 135318000035584 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:17:25.136401 135318000035584 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 13:17:25.136606 135444366239552 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.
I0424 13:17:25.136754 135444366239552 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 13:17:25.137813 135444366239552 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.028, Tokens/s/device: 13871.486, 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 24 13:17:37 UTC 2026
EXIT_CODE=0