MaxView

Timing Test Reference Updates
‹ 13_scan_layers_falseCase: 13_scan_layers_true14_async_ckpt_false_resume ›

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 loss8.17908.17900
TFLOP/s92.03492.237+0.203
Tok/s13872.413903.0+30.607
Avg s/step2.6782.816+0.138
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 06:44:32 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 06:44:56.604711: 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 06:44:56.783913 132559014876992 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 06:45:05.825188 132559014876992 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 06:45:05.827557 132559014876992 distributed.py:172] Connecting to JAX distributed service on mt-13-scan-layers-true-c8ewh-slice-job-0-0.mt-13-scan-layers-true-c8ewh:8482
I0420 06:45:07.072643 132559014876992 max_utils.py:284] Jax distributed system initialized!
I0420 06:45:12.173720 132559014876992 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 06:45:12.173822 132559014876992 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 06:45:12.173862 132559014876992 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 06:45:12.173897 132559014876992 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 06:45:12.867314 132559014876992 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 06:45:12.867597 132559014876992 checkpointing.py:677] Setting up checkpoint logger...
I0420 06:45:12.867657 132559014876992 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 06:45:12.867703 132559014876992 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 06:45:12.868046 132559014876992 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 0x788f4591c350>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 06:45:15.707861 132559014876992 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0420 06:45:15.708099 132559014876992 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 0x788e096e6630>}, handler_registry=None
I0420 06:45:15.708353 132559014876992 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x788e096e6630>` 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 06:45:15.708402 132559014876992 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x787b702ecaa0>` 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 06:45:15.708437 132559014876992 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 0x788e096e6630>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x788e096e6630>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x787b702ecaa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x787b702ecaa0>}).
I0420 06:45:15.708814 132559014876992 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 06:45:15.708888 132559014876992 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x787a806d7d80> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 06:45:16.457064 132559014876992 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_13_scan_layers_true/checkpoints
I0420 06:45:16.601490 132559014876992 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=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_13_scan_layers_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x787b702eddf0>
I0420 06:45:16.601671 132559014876992 checkpointing.py:301] Checkpoint manager created!
I0420 06:45:17.545019 132559014876992 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 06:45:17.545130 132559014876992 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 06:45:17.924985 132559014876992 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 06:45:17.925076 132559014876992 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 06:45:17.941872 132559014876992 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 06:45:17.941937 132559014876992 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 06:45:17.965631 132559014876992 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 06:45:17.965692 132559014876992 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 06:45:17.982176 132559014876992 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 06:45:17.982236 132559014876992 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 06:45:17.999102 132559014876992 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 06:45:17.999169 132559014876992 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 06:45:18.024164 132559014876992 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 06:45:18.024230 132559014876992 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 06:45:18.046307 132559014876992 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 06:45:18.046378 132559014876992 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 06:45:18.262600 132559014876992 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 06:45:18.262708 132559014876992 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 06:45:19.688031 132559014876992 maxtext_utils.py:1654]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 06:45:19.688159 132559014876992 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 06:45:19.688213 132559014876992 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 06:45:19.688272 132559014876992 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 06:45:19.688326 132559014876992 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 06:45:19.688380 132559014876992 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 06:45:19.688436 132559014876992 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 06:45:19.688491 132559014876992 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 06:45:19.688532 132559014876992 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 06:45:19.688575 132559014876992 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 06:45:19.688884 132559014876992 maxtext_utils.py:1654]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0420 06:45:19.689028 132559014876992 maxtext_utils.py:1654]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0420 06:45:20.180001 132559014876992 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 06:45:20.180099 132559014876992 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 06:45:20.195593 132559014876992 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 06:45:20.195653 132559014876992 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 06:45:31.027102 132559014876992 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 06:45:31.027934 132559014876992 metric_logger.py:301] number parameters: 1.104 billion
I0420 06:45:42.158329 132559014876992 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0420 06:45:42.393008 132559014876992 checkpointing.py:776] Waited 0.2346630096435547 seconds for step 0 to finish before starting checkpointing.
I0420 06:45:42.395500 132559014876992 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 06:45:42.397157 132559014876992 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0
I0420 06:45:42.398583 132559014876992 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_13_scan_layers_true/checkpoints/0.
I0420 06:45:42.693160 132559014876992 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 06:45:42.694186 132559014876992 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 06:45:42.694243 132559014876992 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 06:45:42.966320 132559014876992 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273237s
I0420 06:45:42.966510 132559014876992 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.535 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2787172794342041 s) (per-host)
I0420 06:45:42.966562 132559014876992 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.278780s (batch_requests_ready=0.002293s, total_serialization_initiated=0.276417s, others=0.000071s)
I0420 06:45:42.966654 132559014876992 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.282999s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.282982)
I0420 06:45:42.967406 132559014876992 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.57 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_13_scan_layers_true/checkpoints/0.
I0420 06:45:42.967728 132432405710592 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 07:05:42.967690
I0420 06:45:42.981384 132559014876992 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 06:45:42.981694 132430796605184 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0420 06:45:42.981851 132559014876992 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_13_scan_layers_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776667542.3954813, 'wait_for_prev_duration_secs': 6.127357482910156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776667542.397197, 'checkpointer_blocking_duration_secs': 0.5706849098205566, 'get_old_steps_start_time': 1776667542.9679062, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1776667542.3934436, 'checkpoint_manager_blocking_duration_secs': 0.5883669853210449}
I0420 06:45:42.981961 132559014876992 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0420 06:45:42.982012 132559014876992 max_utils.py:750] 
Memstats: After params initialized:
I0420 06:45:42.982058 132559014876992 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0))
I0420 06:45:42.982097 132559014876992 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0))
I0420 06:45:42.982122 132559014876992 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0))
I0420 06:45:42.982146 132559014876992 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0))
I0420 06:45:43.295617 132559014876992 metric_logger.py:196] completed step: 0, seconds: 11.130, TFLOP/s/device: 1.221, Tokens/s/device: 184.003, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 06:45:43.475248 132559014876992 metric_logger.py:196] completed step: 1, seconds: 1.136, TFLOP/s/device: 11.963, Tokens/s/device: 1803.234, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 06:45:43.898794 132559014876992 metric_logger.py:196] completed step: 2, seconds: 0.032, TFLOP/s/device: 421.372, Tokens/s/device: 63513.723, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0420 06:45:44.046145 132559014876992 metric_logger.py:196] completed step: 3, seconds: 0.424, TFLOP/s/device: 32.067, Tokens/s/device: 4833.472, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0420 06:45:44.341683 132559014876992 metric_logger.py:196] completed step: 4, seconds: 0.154, TFLOP/s/device: 88.423, Tokens/s/device: 13328.127, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0420 06:45:44.347766 132559014876992 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.435, Tokens/s/device: 13932.826, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0420 06:45:45.784612    2756 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 06:45:47.807013 132430804997888 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_13_scan_layers_true/checkpoints/0/items/array_metadatas/process_4
I0420 06:46:06.104423 132559014876992 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.884, Tokens/s/device: 6916.209, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0420 06:46:06.251955 132559014876992 metric_logger.py:196] completed step: 7, seconds: 21.610, TFLOP/s/device: 0.629, Tokens/s/device: 94.772, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0420 06:46:06.399433 132559014876992 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.103, Tokens/s/device: 13430.565, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0420 06:46:06.546242 132559014876992 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0420 06:46:06.546931 132559014876992 checkpointing.py:776] Waited 0.0007100105285644531 seconds for step 9 to finish before starting checkpointing.
I0420 06:46:06.549153 132559014876992 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 06:46:18.791249 132432405710592 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 43.752 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.10341787338257 s) (per-host)
I0420 06:46:18.791393 132432405710592 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.823553s.
I0420 06:46:27.483589 132432405710592 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 44.515732s.
I0420 06:46:27.483899 132430796605184 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 06:46:27.484027 132430796605184 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0420 06:46:27.484076 132430796605184 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 06:46:27.485859 132430796605184 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 06:46:27.485988 132559014876992 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 06:46:27.486054 132559014876992 checkpoint_manager.py:1452] Waiting for previous save to complete took 20.936909 seconds. If this number is high, consider checkpointing less frequently.
I0420 06:46:27.488103 132559014876992 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9
I0420 06:46:27.490032 132559014876992 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_13_scan_layers_true/checkpoints/9.
I0420 06:46:27.780846 132559014876992 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 06:46:27.780939 132559014876992 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 06:46:27.814449 132559014876992 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034681s
I0420 06:46:27.814652 132559014876992 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 40.271 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03830528259277344 s) (per-host)
I0420 06:46:27.814707 132559014876992 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038371s (batch_requests_ready=0.001915s, total_serialization_initiated=0.036380s, others=0.000076s)
I0420 06:46:27.814798 132559014876992 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043115s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.043100)
I0420 06:46:27.815552 132559014876992 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_13_scan_layers_true/checkpoints/9.
I0420 06:46:27.815834 132430796605184 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 07:06:27.815805
I0420 06:46:27.823452 132559014876992 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 06:46:27.823670 132405612488448 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0420 06:46:27.823791 132559014876992 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_13_scan_layers_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776667566.5491216, 'wait_for_prev_duration_secs': 20.936909198760986, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776667587.4881449, 'checkpointer_blocking_duration_secs': 0.32782554626464844, 'get_old_steps_start_time': 1776667587.8159947, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776667566.5471642, 'checkpoint_manager_blocking_duration_secs': 21.276590585708618}
I0420 06:46:27.823908 132559014876992 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0420 06:46:27.823952 132559014876992 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 06:46:33.770593 132410511402752 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_13_scan_layers_true/checkpoints/9/items/array_metadatas/process_4
I0420 06:47:08.847445 132430796605184 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 38.460 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.07106137275696 s) (per-host)
I0420 06:47:08.847571 132430796605184 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.031630s.
I0420 06:47:16.699211 132430796605184 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 48.883256s.
I0420 06:47:16.699453 132405612488448 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 06:47:16.699516 132405612488448 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0420 06:47:16.699563 132405612488448 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 06:47:16.701085 132405612488448 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 06:47:16.701247 132559014876992 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 06:47:16.701411 132559014876992 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 06:47:16.702286 132559014876992 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.034, Tokens/s/device: 13872.425, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Mon Apr 20 06:47:28 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 16:42:15 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 16:42:39.328851: 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 16:42:39.506371 133724438087488 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 16:42:48.546572 133724438087488 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 16:42:48.548967 133724438087488 distributed.py:172] Connecting to JAX distributed service on mt-13-scan-layers-true-dq635-slice-job-0-0.mt-13-scan-layers-true-dq635:8482
I0420 16:42:49.877784 133724438087488 max_utils.py:284] Jax distributed system initialized!
I0420 16:42:55.841426 133724438087488 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 16:42:55.841531 133724438087488 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 16:42:55.841571 133724438087488 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 16:42:55.841610 133724438087488 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 16:42:56.904103 133724438087488 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 16:42:56.904678 133724438087488 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 16:42:56.904865 133724438087488 checkpointing.py:688] Setting up checkpoint logger...
I0420 16:42:56.904915 133724438087488 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 16:42:56.904956 133724438087488 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 16:42:56.905307 133724438087488 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 0x799e6de48080>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 16:42:59.631812 133724438087488 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 16:42:59.632039 133724438087488 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 0x799d623b3290>}, handler_registry=None
I0420 16:42:59.632289 133724438087488 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x799d623b3290>` 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 16:42:59.632340 133724438087488 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x799d622c2930>` 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 16:42:59.632375 133724438087488 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 0x799d623b3290>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x799d623b3290>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x799d622c2930>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x799d622c2930>}).
I0420 16:42:59.632695 133724438087488 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 16:42:59.632763 133724438087488 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x798a443cec00> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 16:43:01.615686 133724438087488 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_13_scan_layers_true/checkpoints
I0420 16:43:01.617928 133724438087488 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=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, 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_13_scan_layers_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x799d623b1910>
I0420 16:43:01.618041 133724438087488 checkpointing.py:302] Checkpoint manager created!
I0420 16:43:02.541815 133724438087488 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 16:43:02.541918 133724438087488 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 16:43:02.920834 133724438087488 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 16:43:02.920924 133724438087488 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 16:43:02.937331 133724438087488 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 16:43:02.937393 133724438087488 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 16:43:02.961117 133724438087488 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 16:43:02.961182 133724438087488 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 16:43:02.977616 133724438087488 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 16:43:02.977670 133724438087488 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 16:43:02.993983 133724438087488 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 16:43:02.994041 133724438087488 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 16:43:03.018895 133724438087488 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 16:43:03.018962 133724438087488 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 16:43:03.041057 133724438087488 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 16:43:03.041132 133724438087488 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 16:43:03.257649 133724438087488 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 16:43:03.257748 133724438087488 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 16:43:04.681845 133724438087488 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 16:43:04.681967 133724438087488 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 16:43:04.682019 133724438087488 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 16:43:04.682076 133724438087488 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 16:43:04.682159 133724438087488 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 16:43:04.682199 133724438087488 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 16:43:04.682253 133724438087488 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 16:43:04.682305 133724438087488 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 16:43:04.682344 133724438087488 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 16:43:04.682379 133724438087488 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 16:43:04.682425 133724438087488 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0420 16:43:04.682472 133724438087488 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0420 16:43:05.173413 133724438087488 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 16:43:05.173504 133724438087488 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 16:43:05.189234 133724438087488 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 16:43:05.189300 133724438087488 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 16:43:16.125596 133724438087488 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 16:43:16.133961 133724438087488 metric_logger.py:301] number parameters: 1.104 billion
I0420 16:43:27.454029 133724438087488 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 16:43:27.610482 133724438087488 checkpointing.py:798] Waited 0.1564345359802246 seconds for step 0 to finish before starting checkpointing.
I0420 16:43:27.612991 133724438087488 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 16:43:27.614717 133724438087488 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0420 16:43:27.616357 133724438087488 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_13_scan_layers_true/checkpoints/0.
I0420 16:43:28.771713 133724438087488 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 16:43:28.785849 133724438087488 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 16:43:28.785914 133724438087488 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 16:43:29.054093 133595159066368 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_13_scan_layers_true/checkpoints/0
I0420 16:43:29.056145 133724438087488 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.271002s
I0420 16:43:29.057001 133724438087488 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.562 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2773439884185791 s) (per-host)
I0420 16:43:29.057090 133724438087488 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.277444s (batch_requests_ready=0.002184s, total_serialization_initiated=0.274479s, others=0.000781s)
I0420 16:43:29.057176 133724438087488 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.281478s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.281462)
I0420 16:43:29.057917 133724438087488 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.44 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_13_scan_layers_true/checkpoints/0.
I0420 16:43:29.058230 133595637921536 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:03:29.058195
I0420 16:43:29.070496 133724438087488 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 16:43:29.070747 133592973825792 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 16:43:29.070875 133724438087488 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_13_scan_layers_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776703407.6129718, 'wait_for_prev_duration_secs': 5.9604644775390625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776703407.614755, 'checkpointer_blocking_duration_secs': 1.4436261653900146, 'get_old_steps_start_time': 1776703409.0584035, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776703407.6109958, 'checkpoint_manager_blocking_duration_secs': 1.4598395824432373}
I0420 16:43:29.070980 133724438087488 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 16:43:29.071030 133724438087488 max_utils.py:750] 
Memstats: After params initialized:
I0420 16:43:29.071104 133724438087488 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0420 16:43:29.071138 133724438087488 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0420 16:43:29.071166 133724438087488 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0420 16:43:29.071190 133724438087488 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0420 16:43:29.430944 133724438087488 metric_logger.py:196] completed step: 0, seconds: 11.013, TFLOP/s/device: 1.234, Tokens/s/device: 185.965, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 16:43:29.434484 133724438087488 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_13_scan_layers_true/tensorboard/'
I0420 16:43:29.856876 133724438087488 metric_logger.py:196] completed step: 1, seconds: 1.975, TFLOP/s/device: 6.879, Tokens/s/device: 1036.941, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 16:43:30.003139 133724438087488 metric_logger.py:196] completed step: 2, seconds: 0.427, TFLOP/s/device: 31.845, Tokens/s/device: 4800.064, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0420 16:43:30.150557 133724438087488 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1480.079, Tokens/s/device: 223093.682, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0420 16:43:31.067209 133594622195456 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_13_scan_layers_true/checkpoints/0/items
I0420 16:43:31.208776 133592953210624 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776703410804511756, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776703410804511756, "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_13_scan_layers_true/checkpoints/0/_CHECKPOINT_METADATA
I0420 16:43:32.311912    2845 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 16:43:34.117203 133594042545920 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_13_scan_layers_true/checkpoints/0/items/array_metadatas/process_0
I0420 16:43:52.493572 133724438087488 metric_logger.py:196] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.345, Tokens/s/device: 13919.284, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0420 16:43:52.502321 133724438087488 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.125, Tokens/s/device: 13886.064, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0420 16:43:52.648488 133724438087488 metric_logger.py:196] completed step: 6, seconds: 22.342, TFLOP/s/device: 0.608, Tokens/s/device: 91.667, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0420 16:43:52.795806 133724438087488 metric_logger.py:196] completed step: 7, seconds: 0.006, TFLOP/s/device: 2117.695, Tokens/s/device: 319201.995, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0420 16:43:52.943117 133724438087488 metric_logger.py:196] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.385, Tokens/s/device: 13925.247, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0420 16:43:53.089743 133724438087488 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 16:43:53.090311 133724438087488 checkpointing.py:798] Waited 0.0005898475646972656 seconds for step 9 to finish before starting checkpointing.
I0420 16:43:53.092469 133724438087488 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 16:44:10.462007 133594034153216 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 38.159307s (commit=36.911680s, array_metadata_write=1.247627s)
I0420 16:44:10.463127 133595637921536 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 37.895 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.68342399597168 s) (per-host)
I0420 16:44:10.463235 133595637921536 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.404894s.
I0420 16:44:11.259338 133595637921536 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776703410804511756, '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_13_scan_layers_true/checkpoints/0/_CHECKPOINT_METADATA
I0420 16:44:12.041274 133592953210624 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776703410804511756, '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_13_scan_layers_true/checkpoints/0/_CHECKPOINT_METADATA
I0420 16:44:12.052657 133595637921536 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000252s.
I0420 16:44:12.637841 133595637921536 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 16:44:13.601620 133595637921536 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.213321s. 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_13_scan_layers_true/checkpoints/0/items
I0420 16:44:13.603350 133595637921536 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_13_scan_layers_true/checkpoints/0/items
I0420 16:44:14.148838 133595637921536 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_13_scan_layers_true/checkpoints/0
I0420 16:44:16.067910 133595637921536 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_13_scan_layers_true/checkpoints/0`.
I0420 16:44:16.068815 133595637921536 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 48.45 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_13_scan_layers_true/checkpoints/0
I0420 16:44:16.070387 133595637921536 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 47.012043s.
I0420 16:44:16.070590 133592973825792 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 16:44:16.070714 133592973825792 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 16:44:16.070778 133592973825792 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 16:44:16.072805 133592973825792 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 16:44:16.073004 133724438087488 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 16:44:16.073153 133724438087488 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.980683 seconds. If this number is high, consider checkpointing less frequently.
I0420 16:44:16.075260 133724438087488 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0420 16:44:16.077422 133724438087488 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_13_scan_layers_true/checkpoints/9.
I0420 16:44:16.774416 133724438087488 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 16:44:16.774506 133724438087488 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 16:44:16.800669 133724438087488 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027053s
I0420 16:44:16.801522 133724438087488 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 48.715 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03166532516479492 s) (per-host)
I0420 16:44:16.801583 133724438087488 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031742s (batch_requests_ready=0.001940s, total_serialization_initiated=0.029046s, others=0.000756s)
I0420 16:44:16.801662 133724438087488 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035727s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.035710)
I0420 16:44:16.802398 133724438087488 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.73 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_13_scan_layers_true/checkpoints/9.
I0420 16:44:16.802763 133592973825792 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:04:16.802726
I0420 16:44:16.816576 133724438087488 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 16:44:16.816829 133594034153216 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 16:44:16.816960 133724438087488 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_13_scan_layers_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776703433.092438, 'wait_for_prev_duration_secs': 22.980682611465454, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776703456.075307, 'checkpointer_blocking_duration_secs': 0.7275938987731934, 'get_old_steps_start_time': 1776703456.8029234, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776703433.0905359, 'checkpoint_manager_blocking_duration_secs': 23.726391553878784}
I0420 16:44:16.817062 133724438087488 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 16:44:16.817115 133724438087488 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 16:44:17.013113 133595637921536 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_13_scan_layers_true/checkpoints/9
I0420 16:44:18.813501 133594588624640 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_13_scan_layers_true/checkpoints/9/items
I0420 16:44:21.938111 133594598676224 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_13_scan_layers_true/checkpoints/9/items/array_metadatas/process_0
I0420 16:45:03.996397 133594042545920 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 43.852819s (commit=42.032879s, array_metadata_write=1.819940s)
I0420 16:45:03.997498 133592973825792 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 33.446 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 47.22761011123657 s) (per-host)
I0420 16:45:03.997631 133592973825792 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 47.194757s.
I0420 16:45:05.883909 133592973825792 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000217s.
I0420 16:45:06.131860 133592973825792 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 16:45:07.086815 133592973825792 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.193326s. 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_13_scan_layers_true/checkpoints/9/items
I0420 16:45:07.088394 133592973825792 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_13_scan_layers_true/checkpoints/9/items
I0420 16:45:07.613582 133592973825792 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_13_scan_layers_true/checkpoints/9
I0420 16:45:09.049602 133592973825792 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_13_scan_layers_true/checkpoints/9`.
I0420 16:45:09.050335 133592973825792 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 52.98 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_13_scan_layers_true/checkpoints/9
I0420 16:45:09.051892 133592973825792 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 52.249015s.
I0420 16:45:09.052114 133594034153216 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 16:45:09.052234 133594034153216 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 16:45:09.052299 133594034153216 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 16:45:09.055008 133594034153216 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 16:45:09.055158 133724438087488 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 16:45:09.055240 133724438087488 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 16:45:09.056284 133724438087488 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.237, Tokens/s/device: 13903.032, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Mon Apr 20 16:45:21 UTC 2026
EXIT_CODE=0