MaxView

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

Metrics: main (574ad3fb9) vs feat/nnx-post-train-fixes (574ad3fb9)

Metricmain  574ad3fb9feat/nnx-post-train-fixes  574ad3fb9Diff (feat/nnx-post-train-fixes − main)
Parameters1.104 billion1.104 billion—
Final loss7.18107.18100
TFLOP/s92.05492.235+0.181
Tok/s13875.313902.7+27.315
Avg s/step7.7077.729+0.022
Memory %1.381.380
JAX0.9.20.9.2—

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

main  ·  574ad3fb9  ·  main_20260420_060028  ·  full log 
XPK Start: Mon Apr 20 07:16:14 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-20 07:16:38.242202: 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)
I0420 07:16:38.773853 137542338553664 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 07:16:47.816251 137542338553664 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 07:16:47.818670 137542338553664 distributed.py:172] Connecting to JAX distributed service on mt-16-zarr3-true-sv-vluci-slice-job-0-0.mt-16-zarr3-true-sv-vluci:8482
I0420 07:16:48.926547 137542338553664 max_utils.py:284] Jax distributed system initialized!
I0420 07:16:55.242302 137542338553664 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 07:16:55.242430 137542338553664 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 07:16:55.242474 137542338553664 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
I0420 07:16:55.242511 137542338553664 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 07:16:55.942656 137542338553664 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 07:16:55.942933 137542338553664 checkpointing.py:677] Setting up checkpoint logger...
I0420 07:16:55.942986 137542338553664 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 07:16:55.943032 137542338553664 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 07:16:55.943374 137542338553664 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 0x7d178410a6c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 07:16:58.806046 137542338553664 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0420 07:16:58.806276 137542338553664 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d02f0643050>}, handler_registry=None
I0420 07:16:58.806525 137542338553664 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d02f0643050>` 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`.
I0420 07:16:58.806586 137542338553664 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d02f0644050>` 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`.
I0420 07:16:58.806628 137542338553664 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 0x7d02f0643050>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d02f0643050>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d02f0644050>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d02f0644050>}).
I0420 07:16:58.806951 137542338553664 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 07:16:58.807021 137542338553664 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d02f03db880> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 07:16:59.976365 137542338553664 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints
I0420 07:17:00.276830 137542338553664 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=5), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d02f079f8c0>
I0420 07:17:00.276997 137542338553664 checkpointing.py:301] Checkpoint manager created!
I0420 07:17:01.196409 137542338553664 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 07:17:01.196519 137542338553664 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 07:17:01.571112 137542338553664 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 07:17:01.571200 137542338553664 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 07:17:01.587372 137542338553664 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 07:17:01.587427 137542338553664 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 07:17:01.611468 137542338553664 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 07:17:01.611536 137542338553664 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 07:17:01.627722 137542338553664 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 07:17:01.627784 137542338553664 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 07:17:01.644080 137542338553664 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 07:17:01.644138 137542338553664 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 07:17:01.668555 137542338553664 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 07:17:01.668625 137542338553664 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 07:17:01.693949 137542338553664 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 07:17:01.694046 137542338553664 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 07:17:01.908438 137542338553664 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 07:17:01.908540 137542338553664 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 07:17:03.342322 137542338553664 maxtext_utils.py:1654]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 07:17:03.342477 137542338553664 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 07:17:03.342529 137542338553664 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 07:17:03.342589 137542338553664 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0420 07:17:03.342643 137542338553664 maxtext_utils.py:1654]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 07:17:03.342680 137542338553664 maxtext_utils.py:1654]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 07:17:03.342733 137542338553664 maxtext_utils.py:1654]  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)
I0420 07:17:03.342785 137542338553664 maxtext_utils.py:1654]  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')
I0420 07:17:03.342824 137542338553664 maxtext_utils.py:1654]  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)
I0420 07:17:03.342859 137542338553664 maxtext_utils.py:1654]  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)
I0420 07:17:03.342905 137542338553664 maxtext_utils.py:1654]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0420 07:17:03.342952 137542338553664 maxtext_utils.py:1654]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0420 07:17:03.833580 137542338553664 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 07:17:03.833679 137542338553664 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 07:17:03.848945 137542338553664 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 07:17:03.849003 137542338553664 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 07:17:14.566355 137542338553664 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.
I0420 07:17:14.567132 137542338553664 metric_logger.py:301] number parameters: 1.104 billion
I0420 07:17:26.014483 137542338553664 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0420 07:17:26.168104 137542338553664 checkpointing.py:776] Waited 0.15360283851623535 seconds for step 0 to finish before starting checkpointing.
I0420 07:17:26.170780 137542338553664 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 07:17:26.172412 137542338553664 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0
I0420 07:17:26.173855 137542338553664 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/0.
I0420 07:17:26.504199 137542338553664 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 07:17:26.505139 137542338553664 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 07:17:26.505197 137542338553664 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
I0420 07:17:26.777873 137542338553664 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273744s
I0420 07:17:26.778046 137542338553664 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.525 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2792205810546875 s) (per-host)
I0420 07:17:26.778098 137542338553664 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279282s (batch_requests_ready=0.002289s, total_serialization_initiated=0.276924s, others=0.000069s)
I0420 07:17:26.778188 137542338553664 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283413s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.283395)
I0420 07:17:26.779043 137542338553664 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/0.
I0420 07:17:26.779329 137412158699264 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 07:37:26.779293
I0420 07:17:26.790814 137542338553664 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 07:17:26.791093 137410552416000 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0420 07:17:26.791249 137542338553664 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776669446.1707616, 'wait_for_prev_duration_secs': 5.936622619628906e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776669446.172449, 'checkpointer_blocking_duration_secs': 0.6070399284362793, 'get_old_steps_start_time': 1776669446.7795155, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1776669446.1686616, 'checkpoint_manager_blocking_duration_secs': 0.6225507259368896}
I0420 07:17:26.791373 137542338553664 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0420 07:17:26.791424 137542338553664 max_utils.py:750] 
Memstats: After params initialized:
I0420 07:17:26.791473 137542338553664 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0))
I0420 07:17:26.791506 137542338553664 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0))
I0420 07:17:26.791534 137542338553664 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0))
I0420 07:17:26.791559 137542338553664 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0))
I0420 07:17:27.106599 137542338553664 metric_logger.py:196] completed step: 0, seconds: 11.447, TFLOP/s/device: 1.187, Tokens/s/device: 178.908, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 07:17:27.278324 137542338553664 metric_logger.py:196] completed step: 1, seconds: 1.091, TFLOP/s/device: 12.458, Tokens/s/device: 1877.876, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 07:17:27.688173 137542338553664 metric_logger.py:196] completed step: 2, seconds: 0.024, TFLOP/s/device: 558.245, Tokens/s/device: 84144.788, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719
I0420 07:17:27.835178 137542338553664 metric_logger.py:196] completed step: 3, seconds: 0.410, TFLOP/s/device: 33.173, Tokens/s/device: 5000.256, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506
I0420 07:17:27.839765 137542338553664 checkpointing.py:772] Waiting for step 5 to finish before checkpoint...
I0420 07:17:28.129288 137542338553664 checkpointing.py:776] Waited 0.2894906997680664 seconds for step 5 to finish before starting checkpointing.
I0420 07:17:28.132366 137542338553664 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 07:17:29.739420    2865 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 07:17:32.252832 137410560808704 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/0/items/array_metadatas/process_4
I0420 07:18:03.264580 137412158699264 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 42.964 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.765721559524536 s) (per-host)
I0420 07:18:03.264697 137412158699264 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.485245s.
I0420 07:18:12.005557 137412158699264 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 45.226089s.
I0420 07:18:12.005851 137410552416000 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 07:18:12.005975 137410552416000 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0420 07:18:12.006026 137410552416000 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 07:18:12.008569 137410552416000 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 07:18:12.008754 137542338553664 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 07:18:12.008895 137542338553664 checkpoint_manager.py:1452] Waiting for previous save to complete took 43.876545 seconds. If this number is high, consider checkpointing less frequently.
I0420 07:18:12.010880 137542338553664 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 5
I0420 07:18:12.013251 137542338553664 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/5.
I0420 07:18:12.741195 137542338553664 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 07:18:12.741293 137542338553664 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
I0420 07:18:12.774454 137542338553664 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034340s
I0420 07:18:12.774657 137542338553664 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 40.703 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03789854049682617 s) (per-host)
I0420 07:18:12.774709 137542338553664 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.037960s (batch_requests_ready=0.001872s, total_serialization_initiated=0.036018s, others=0.000069s)
I0420 07:18:12.774796 137542338553664 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043129s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.043113)
I0420 07:18:12.775530 137542338553664 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/5.
I0420 07:18:12.775805 137410552416000 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 07:38:12.775782
I0420 07:18:12.785679 137542338553664 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 07:18:12.785980 137388747613952 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0420 07:18:12.786143 137542338553664 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776669448.1323204, 'wait_for_prev_duration_secs': 43.876544713974, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776669492.0109177, 'checkpointer_blocking_duration_secs': 0.7649953365325928, 'get_old_steps_start_time': 1776669492.7759364, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776669448.129948, 'checkpoint_manager_blocking_duration_secs': 44.65616154670715}
I0420 07:18:12.786257 137542338553664 checkpointing.py:408] Started an asynchronous checkpoint save for step 5
I0420 07:18:12.787040 137542338553664 metric_logger.py:196] completed step: 4, seconds: 0.154, TFLOP/s/device: 88.438, Tokens/s/device: 13330.296, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355
I0420 07:18:12.792789 137542338553664 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.421, Tokens/s/device: 13930.741, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467
I0420 07:18:17.971905 137410560808704 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/5/items/array_metadatas/process_4
I0420 07:18:35.226711 137542338553664 metric_logger.py:196] completed step: 6, seconds: 44.952, TFLOP/s/device: 0.302, Tokens/s/device: 45.560, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740
I0420 07:18:35.374139 137542338553664 metric_logger.py:196] completed step: 7, seconds: 22.288, TFLOP/s/device: 0.610, Tokens/s/device: 91.889, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151
I0420 07:18:35.521697 137542338553664 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.411, Tokens/s/device: 13476.965, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098
I0420 07:18:35.668426 137542338553664 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0420 07:18:35.669105 137542338553664 checkpointing.py:776] Waited 0.0006990432739257812 seconds for step 9 to finish before starting checkpointing.
I0420 07:18:35.671230 137542338553664 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 07:18:55.064854 137410552416000 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.318 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.328054666519165 s) (per-host)
I0420 07:18:55.064978 137410552416000 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.289090s.
I0420 07:19:02.974133 137410552416000 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 50.198228s.
I0420 07:19:02.974433 137388747613952 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 07:19:02.974554 137388747613952 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0420 07:19:02.974602 137388747613952 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts...
I0420 07:19:02.976805 137388747613952 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts.
I0420 07:19:02.976962 137542338553664 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5.
W0420 07:19:02.977081 137542338553664 checkpoint_manager.py:1452] Waiting for previous save to complete took 27.305858 seconds. If this number is high, consider checkpointing less frequently.
I0420 07:19:02.981086 137542338553664 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9
I0420 07:19:02.983196 137542338553664 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/9.
I0420 07:19:03.294789 137542338553664 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 07:19:03.294882 137542338553664 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
I0420 07:19:03.327592 137542338553664 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.033633s
I0420 07:19:03.327753 137542338553664 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 41.559 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.037117958068847656 s) (per-host)
I0420 07:19:03.327803 137542338553664 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.037177s (batch_requests_ready=0.001746s, total_serialization_initiated=0.035367s, others=0.000064s)
I0420 07:19:03.327901 137542338553664 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.041232s (all_items=0.000010s, per_item={'items': '0.00001025'}, temp_paths=0.041222)
I0420 07:19:03.328608 137542338553664 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/9.
I0420 07:19:03.328880 137388747613952 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 07:39:03.328858
I0420 07:19:03.338919 137542338553664 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 07:19:03.339206 137410560808704 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0420 07:19:03.339380 137542338553664 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776669515.6711993, 'wait_for_prev_duration_secs': 27.30585789680481, 'time_between_consecutive_saves_sec': 23.662587881088257, 'checkpointer_blocking_start_time': 1776669542.981126, 'checkpointer_blocking_duration_secs': 0.34784626960754395, 'get_old_steps_start_time': 1776669543.3289897, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1776669515.6693308, 'checkpoint_manager_blocking_duration_secs': 27.670016527175903}
I0420 07:19:03.339495 137542338553664 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0420 07:19:03.339537 137542338553664 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 07:19:08.417920 137409990600448 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_16_zarr3_true/checkpoints/9/items/array_metadatas/process_4
I0420 07:19:45.205867 137388747613952 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.686 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.91518783569336 s) (per-host)
I0420 07:19:45.205987 137388747613952 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.877038s.
I0420 07:19:52.698626 137388747613952 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 49.369661s.
I0420 07:19:52.698922 137410560808704 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 07:19:52.699042 137410560808704 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0420 07:19:52.699084 137410560808704 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 07:19:52.704137 137410560808704 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 07:19:52.704298 137542338553664 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0420 07:19:52.704495 137542338553664 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 07:19:52.705415 137542338553664 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.054, Tokens/s/device: 13875.339, 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: Mon Apr 20 07:20:05 UTC 2026
EXIT_CODE=0
feat/nnx-post-train-fixes  ·  574ad3fb9  ·  feat_nnx_post_train_fixes_20260420_153543  ·  full log 
XPK Start: Mon Apr 20 17:14:35 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-20 17:14:59.520842: 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)
I0420 17:14:59.699785 139634550314816 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 17:15:08.741649 139634550314816 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 17:15:08.744018 139634550314816 distributed.py:172] Connecting to JAX distributed service on mt-16-zarr3-true-sv-9f98v-slice-job-0-0.mt-16-zarr3-true-sv-9f98v:8482
I0420 17:15:10.125970 139634550314816 max_utils.py:284] Jax distributed system initialized!
I0420 17:15:16.259038 139634550314816 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 17:15:16.259160 139634550314816 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 17:15:16.259201 139634550314816 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
I0420 17:15:16.259238 139634550314816 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 17:15:17.306485 139634550314816 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 17:15:17.307064 139634550314816 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 17:15:17.307260 139634550314816 checkpointing.py:688] Setting up checkpoint logger...
I0420 17:15:17.307310 139634550314816 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 17:15:17.307353 139634550314816 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 17:15:17.307691 139634550314816 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 0x7efe7c050080>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 17:15:20.006291 139634550314816 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 17:15:20.006522 139634550314816 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7efd6fb41f40>}, handler_registry=None
I0420 17:15:20.006764 139634550314816 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7efd6fb41f40>` 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`.
I0420 17:15:20.006814 139634550314816 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eeacc7282c0>` 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`.
I0420 17:15:20.006850 139634550314816 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 0x7efd6fb41f40>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7efd6fb41f40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eeacc7282c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eeacc7282c0>}).
I0420 17:15:20.007172 139634550314816 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 17:15:20.007242 139634550314816 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7efd6fc6b7e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 17:15:21.112446 139634550314816 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints
I0420 17:15:21.532663 139634550314816 checkpoint_manager.py:929] [process=0][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_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_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7efd6fb408c0>
I0420 17:15:21.532831 139634550314816 checkpointing.py:302] Checkpoint manager created!
I0420 17:15:22.439964 139634550314816 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 17:15:22.440069 139634550314816 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:15:22.820051 139634550314816 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 17:15:22.820152 139634550314816 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:15:22.836384 139634550314816 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 17:15:22.836441 139634550314816 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:15:22.865269 139634550314816 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:15:22.865365 139634550314816 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:15:22.881721 139634550314816 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:15:22.881780 139634550314816 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:15:22.898028 139634550314816 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:15:22.898098 139634550314816 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:15:22.922600 139634550314816 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 17:15:22.922670 139634550314816 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:15:22.943408 139634550314816 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 17:15:22.943472 139634550314816 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 17:15:23.159147 139634550314816 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 17:15:23.159258 139634550314816 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 17:15:24.566322 139634550314816 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 17:15:24.566454 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 17:15:24.566506 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 17:15:24.566564 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0420 17:15:24.566616 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 17:15:24.566653 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 17:15:24.566713 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0420 17:15:24.566771 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0420 17:15:24.566810 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0420 17:15:24.566845 139634550314816 maxtext_utils.py:1827]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0420 17:15:24.566889 139634550314816 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0420 17:15:24.566937 139634550314816 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0420 17:15:25.056308 139634550314816 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 17:15:25.056400 139634550314816 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 17:15:25.071633 139634550314816 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 17:15:25.071695 139634550314816 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 17:15:35.808651 139634550314816 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.
I0420 17:15:35.816550 139634550314816 metric_logger.py:301] number parameters: 1.104 billion
I0420 17:15:47.080656 139634550314816 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 17:15:47.368865 139634550314816 checkpointing.py:798] Waited 0.28818821907043457 seconds for step 0 to finish before starting checkpointing.
I0420 17:15:47.371550 139634550314816 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 17:15:47.373418 139634550314816 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0420 17:15:47.374763 139634550314816 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0.
I0420 17:15:48.090717 139634550314816 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 17:15:48.105651 139634550314816 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 17:15:48.105719 139634550314816 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
I0420 17:15:48.377840 139634550314816 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273092s
I0420 17:15:48.378745 139634550314816 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.520 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2794311046600342 s) (per-host)
I0420 17:15:48.378821 139634550314816 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279521s (batch_requests_ready=0.002054s, total_serialization_initiated=0.276639s, others=0.000829s)
I0420 17:15:48.378905 139634550314816 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283654s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.283637)
I0420 17:15:48.379738 139634550314816 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.01 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0.
I0420 17:15:48.380051 139507825137408 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:35:48.380017
I0420 17:15:48.392252 139634550314816 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:15:48.392503 139506187798272 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:15:48.392658 139634550314816 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776705347.3715322, 'wait_for_prev_duration_secs': 5.888938903808594e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776705347.3734553, 'checkpointer_blocking_duration_secs': 1.0067596435546875, 'get_old_steps_start_time': 1776705348.3802385, 'get_old_steps_duration_secs': 2.6941299438476562e-05, 'checkpoint_manager_blocking_start_time': 1776705347.3694317, 'checkpoint_manager_blocking_duration_secs': 1.0231895446777344}
I0420 17:15:48.392758 139634550314816 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 17:15:48.392807 139634550314816 max_utils.py:750] 
Memstats: After params initialized:
I0420 17:15:48.392860 139634550314816 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0420 17:15:48.392894 139634550314816 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0420 17:15:48.392920 139634550314816 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0420 17:15:48.392945 139634550314816 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0420 17:15:48.399553 139507816744704 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0
I0420 17:15:48.734429 139634550314816 metric_logger.py:196] completed step: 0, seconds: 10.962, TFLOP/s/device: 1.239, Tokens/s/device: 186.831, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 17:15:48.737725 139634550314816 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/tensorboard/'
I0420 17:15:49.175512 139634550314816 metric_logger.py:196] completed step: 1, seconds: 1.652, TFLOP/s/device: 8.225, Tokens/s/device: 1239.787, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 17:15:49.321639 139634550314816 metric_logger.py:196] completed step: 2, seconds: 0.442, TFLOP/s/device: 30.755, Tokens/s/device: 4635.729, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719
I0420 17:15:49.468961 139634550314816 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1532.671, Tokens/s/device: 231020.869, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506
I0420 17:15:49.478042 139634550314816 checkpointing.py:794] Waiting for step 5 to finish before checkpoint...
I0420 17:15:49.762673 139634550314816 checkpointing.py:798] Waited 0.28450584411621094 seconds for step 5 to finish before starting checkpointing.
I0420 17:15:49.765541 139634550314816 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:15:50.250942 139507799959296 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0/items
I0420 17:15:50.403151 139501410187008 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776705349987669283, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776705349987669283, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0/_CHECKPOINT_METADATA
I0420 17:15:51.409990    2795 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 17:15:53.252408 139506242025216 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0/items/array_metadatas/process_0
I0420 17:16:28.445143 139506196190976 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.035922s (commit=35.769801s, array_metadata_write=1.266121s)
I0420 17:16:28.446644 139507825137408 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 39.150 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.34729623794556 s) (per-host)
I0420 17:16:28.446810 139507825137408 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.066627s.
I0420 17:16:29.239601 139507825137408 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776705349987669283, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0/_CHECKPOINT_METADATA
I0420 17:16:29.966215 139507825137408 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000374s.
I0420 17:16:30.017364 139501410187008 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776705349987669283, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0/_CHECKPOINT_METADATA
I0420 17:16:30.628694 139507825137408 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 17:16:31.590212 139507825137408 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.222939s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0/items
I0420 17:16:31.592418 139507825137408 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0/items
I0420 17:16:32.156462 139507825137408 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0
I0420 17:16:33.649337 139507825137408 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0`.
I0420 17:16:33.650392 139507825137408 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 46.28 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/0
I0420 17:16:33.652089 139507825137408 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 45.271891s.
I0420 17:16:33.652321 139506187798272 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:16:33.652493 139506187798272 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:16:33.652597 139506187798272 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:16:33.654216 139506187798272 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 17:16:33.654522 139634550314816 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 17:16:33.654714 139634550314816 checkpoint_manager.py:1452] Waiting for previous save to complete took 43.889174 seconds. If this number is high, consider checkpointing less frequently.
I0420 17:16:33.656667 139634550314816 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 5
I0420 17:16:33.659055 139634550314816 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5.
I0420 17:16:33.960837 139634550314816 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 17:16:33.960970 139634550314816 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
I0420 17:16:34.006214 139634550314816 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.046404s
I0420 17:16:34.007151 139634550314816 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 28.158 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.05478358268737793 s) (per-host)
I0420 17:16:34.007246 139634550314816 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.054901s (batch_requests_ready=0.003914s, total_serialization_initiated=0.050162s, others=0.000825s)
I0420 17:16:34.007366 139634550314816 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.059891s (all_items=0.000023s, per_item={'items': '0.00002337'}, temp_paths=0.059868)
I0420 17:16:34.008435 139634550314816 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5.
I0420 17:16:34.008877 139506187798272 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:36:34.008836
I0420 17:16:34.011363 139634550314816 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:16:34.011745 139507173156608 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:16:34.011980 139634550314816 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776705349.7654955, 'wait_for_prev_duration_secs': 43.889174461364746, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776705393.6567347, 'checkpointer_blocking_duration_secs': 0.35233163833618164, 'get_old_steps_start_time': 1776705394.009106, 'get_old_steps_duration_secs': 4.291534423828125e-05, 'checkpoint_manager_blocking_start_time': 1776705349.7632942, 'checkpoint_manager_blocking_duration_secs': 44.248631954193115}
I0420 17:16:34.012135 139634550314816 checkpointing.py:409] Started an asynchronous checkpoint save for step 5
I0420 17:16:34.196312 139507825137408 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5
I0420 17:16:35.998769 139500700362496 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5/items
I0420 17:16:38.994722 139507279570688 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5/items/array_metadatas/process_0
I0420 17:16:56.327829 139634550314816 metric_logger.py:196] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.347, Tokens/s/device: 13919.568, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355
I0420 17:16:56.337242 139634550314816 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.056, Tokens/s/device: 13875.715, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467
I0420 17:16:56.482749 139634550314816 metric_logger.py:196] completed step: 6, seconds: 66.858, TFLOP/s/device: 0.203, Tokens/s/device: 30.632, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740
I0420 17:16:56.630021 139634550314816 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 1841.823, Tokens/s/device: 277619.629, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151
I0420 17:16:56.777363 139634550314816 metric_logger.py:196] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.540, Tokens/s/device: 13948.673, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098
I0420 17:16:56.924132 139634550314816 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 17:16:56.924879 139634550314816 checkpointing.py:798] Waited 0.0007889270782470703 seconds for step 9 to finish before starting checkpointing.
I0420 17:16:56.926843 139634550314816 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:17:19.734411 139507181549312 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 42.570348s (commit=41.349827s, array_metadata_write=1.220521s)
I0420 17:17:19.735595 139506187798272 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.502 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 45.783180236816406 s) (per-host)
I0420 17:17:19.735720 139506187798272 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 45.726695s.
I0420 17:17:21.451274 139506187798272 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000245s.
I0420 17:17:21.538965 139506187798272 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 17:17:22.642771 139506187798272 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.073985s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5/items
I0420 17:17:22.645103 139506187798272 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5/items
I0420 17:17:23.225587 139506187798272 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5
I0420 17:17:24.694074 139506187798272 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5`.
I0420 17:17:24.694906 139506187798272 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 51.04 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/5
I0420 17:17:24.697830 139506187798272 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 50.688803s.
I0420 17:17:24.697976 139507173156608 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:17:24.698041 139507173156608 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:17:24.698121 139507173156608 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:17:24.699857 139507173156608 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts.
I0420 17:17:24.699995 139634550314816 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5.
W0420 17:17:24.700099 139634550314816 checkpoint_manager.py:1452] Waiting for previous save to complete took 27.773247 seconds. If this number is high, consider checkpointing less frequently.
I0420 17:17:24.701709 139634550314816 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0420 17:17:24.703726 139634550314816 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9.
I0420 17:17:25.430522 139634550314816 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 17:17:25.430619 139634550314816 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
I0420 17:17:25.456333 139634550314816 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026809s
I0420 17:17:25.457144 139634550314816 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 49.481 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.031174898147583008 s) (per-host)
I0420 17:17:25.457223 139634550314816 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031273s (batch_requests_ready=0.001754s, total_serialization_initiated=0.028779s, others=0.000740s)
I0420 17:17:25.457310 139634550314816 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035343s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.035327)
I0420 17:17:25.457981 139634550314816 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9.
I0420 17:17:25.458271 139507147978496 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:37:25.458236
I0420 17:17:25.466033 139634550314816 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:17:25.466303 139507173156608 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:17:25.466447 139634550314816 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776705416.926811, 'wait_for_prev_duration_secs': 27.773247003555298, 'time_between_consecutive_saves_sec': 23.272525310516357, 'checkpointer_blocking_start_time': 1776705444.7017484, 'checkpointer_blocking_duration_secs': 0.7566654682159424, 'get_old_steps_start_time': 1776705445.4584317, 'get_old_steps_duration_secs': 2.7179718017578125e-05, 'checkpoint_manager_blocking_start_time': 1776705416.9251065, 'checkpoint_manager_blocking_duration_secs': 28.541305541992188}
I0420 17:17:25.466568 139634550314816 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 17:17:25.466610 139634550314816 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:17:25.682033 139506187798272 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9
I0420 17:17:27.689650 139507297793792 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9/items
I0420 17:17:31.065525 139507279570688 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9/items/array_metadatas/process_0
I0420 17:18:11.992287 139507181549312 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 43.151654s (commit=41.917618s, array_metadata_write=1.234036s)
I0420 17:18:11.994428 139507147978496 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 33.920 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 46.568421363830566 s) (per-host)
I0420 17:18:11.994551 139507147978496 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 46.536168s.
I0420 17:18:13.505845 139507147978496 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000152s.
I0420 17:18:14.154824 139507147978496 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 17:18:15.123731 139507147978496 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.225812s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9/items
I0420 17:18:15.130349 139507147978496 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9/items
I0420 17:18:16.118632 139507147978496 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9
I0420 17:18:17.991872 139507147978496 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9`.
I0420 17:18:17.992723 139507147978496 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 53.29 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_16_zarr3_true/checkpoints/9
I0420 17:18:17.994348 139507147978496 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 52.535962s.
I0420 17:18:17.994593 139507173156608 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:18:17.994791 139507173156608 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:18:17.994843 139507173156608 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:18:17.997309 139507173156608 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 17:18:17.997508 139634550314816 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0420 17:18:17.997664 139634550314816 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 17:18:17.998762 139634550314816 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.235, Tokens/s/device: 13902.654, 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: Mon Apr 20 17:18:30 UTC 2026
EXIT_CODE=0