MaxView

Timing Test Reference Updates
‹ 03_dropoutCase: 04_int805_fp8 ›

Metrics: main (574ad3fb9) vs feat/nnx-trainstate-and-training-loop (574ad3fb9)

Metricmain  574ad3fb9feat/nnx-trainstate-and-training-loop  574ad3fb9Diff (feat/nnx-trainstate-and-training-loop − main)
Parameters1.104 billion1.104 billion—
Final loss8.17508.17500
TFLOP/s59.28459.232-0.052
Tok/s8935.98928.1-7.83
Avg s/step3.6803.700+0.02
Memory %1.381.380
JAX0.9.20.9.2—

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

main  ·  574ad3fb9  ·  main_20260419_180002  ·  full log 
XPK Start: Sun Apr 19 18:17:28 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-19 18:17:52.753285: 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)
I0419 18:17:53.293206 135611359127360 max_utils.py:273] Attempting to initialize the jax distributed system...
I0419 18:18:02.335359 135611359127360 distributed.py:149] Starting JAX distributed service on [::]:8482
I0419 18:18:02.337694 135611359127360 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-f32cx-slice-job-0-0.mt-04-int8-f32cx:8482
I0419 18:18:04.213380 135611359127360 max_utils.py:284] Jax distributed system initialized!
I0419 18:18:10.285667 135611359127360 max_utils.py:800] System Information: Jax Version: 0.9.2
I0419 18:18:10.285769 135611359127360 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0419 18:18:10.285807 135611359127360 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
I0419 18:18:10.285841 135611359127360 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0419 18:18:11.022398 135611359127360 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 18:18:11.022812 135611359127360 checkpointing.py:677] Setting up checkpoint logger...
I0419 18:18:11.022868 135611359127360 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0419 18:18:11.022912 135611359127360 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0419 18:18:11.023247 135611359127360 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 0x7b55d9f14230>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0419 18:18:14.063397 135611359127360 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0419 18:18:14.063781 135611359127360 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b417c463b00>}, handler_registry=None
I0419 18:18:14.064023 135611359127360 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b417c463b00>` 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`.
I0419 18:18:14.064071 135611359127360 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b417c3c61b0>` 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`.
I0419 18:18:14.064105 135611359127360 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 0x7b417c463b00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b417c463b00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b417c3c61b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b417c3c61b0>}).
I0419 18:18:14.064431 135611359127360 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0419 18:18:14.064506 135611359127360 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b40a47f77e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0419 18:18:15.121534 135611359127360 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints
I0419 18:18:15.585783 135611359127360 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=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_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b417c3c4440>
I0419 18:18:15.585953 135611359127360 checkpointing.py:301] Checkpoint manager created!
I0419 18:18:16.722642 135611359127360 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 18:18:16.722757 135611359127360 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 18:18:17.106274 135611359127360 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 18:18:17.106366 135611359127360 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 18:18:17.122425 135611359127360 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 18:18:17.122498 135611359127360 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 18:18:17.189075 135611359127360 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 18:18:17.189163 135611359127360 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 18:18:17.205303 135611359127360 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 18:18:17.205366 135611359127360 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 18:18:17.221619 135611359127360 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 18:18:17.221678 135611359127360 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 18:18:17.246205 135611359127360 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0419 18:18:17.246274 135611359127360 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 18:18:17.314240 135611359127360 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0419 18:18:17.314324 135611359127360 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0419 18:18:17.834338 135611359127360 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0419 18:18:17.834459 135611359127360 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0419 18:18:19.873654 135611359127360 maxtext_utils.py:1654]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0419 18:18:19.873779 135611359127360 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)
I0419 18:18:19.873832 135611359127360 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)
I0419 18:18:19.873890 135611359127360 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')
I0419 18:18:19.873946 135611359127360 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)
I0419 18:18:19.873986 135611359127360 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)
I0419 18:18:19.874041 135611359127360 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)
I0419 18:18:19.874095 135611359127360 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')
I0419 18:18:19.874136 135611359127360 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)
I0419 18:18:19.874174 135611359127360 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)
I0419 18:18:19.874221 135611359127360 maxtext_utils.py:1654]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0419 18:18:19.874267 135611359127360 maxtext_utils.py:1654]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0419 18:18:21.717385 135611359127360 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0419 18:18:21.717495 135611359127360 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0419 18:18:21.733727 135611359127360 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0419 18:18:21.733790 135611359127360 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0419 18:18:35.245478 135611359127360 max_utils.py:791] Total memory size: 1.8 GB, Output size: 0.4 GB, Temp size: 1.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0419 18:18:35.246288 135611359127360 metric_logger.py:301] number parameters: 1.104 billion
I0419 18:18:50.460363 135611359127360 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0419 18:18:50.699476 135611359127360 checkpointing.py:776] Waited 0.23905563354492188 seconds for step 0 to finish before starting checkpointing.
I0419 18:18:50.706314 135611359127360 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 18:18:50.707862 135611359127360 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0419 18:18:50.709259 135611359127360 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints/0.
I0419 18:18:51.477919 135611359127360 signaling_client.py:364] Using JaxDistributedSignalingClient
I0419 18:18:51.478909 135611359127360 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0419 18:18:51.478966 135611359127360 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
I0419 18:18:51.769770 135611359127360 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.291924s
I0419 18:18:51.769945 135611359127360 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.184 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2975654602050781 s) (per-host)
I0419 18:18:51.769996 135611359127360 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.297627s (batch_requests_ready=0.002389s, total_serialization_initiated=0.295169s, others=0.000069s)
I0419 18:18:51.770086 135611359127360 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.301710s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.301692)
I0419 18:18:51.770845 135611359127360 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.06 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints/0.
I0419 18:18:51.771110 135479590754048 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 18:38:51.771088
I0419 18:18:51.773141 135611359127360 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 18:18:51.773371 135478514792192 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0419 18:18:51.773487 135611359127360 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776622730.7062948, 'wait_for_prev_duration_secs': 6.532669067382812e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776622730.7079, 'checkpointer_blocking_duration_secs': 1.063309669494629, 'get_old_steps_start_time': 1776622731.7712355, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776622730.7000296, 'checkpoint_manager_blocking_duration_secs': 1.073418140411377}
I0419 18:18:51.773643 135611359127360 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0419 18:18:51.773694 135611359127360 max_utils.py:750] 
Memstats: After params initialized:
I0419 18:18:51.773743 135611359127360 max_utils.py:756] 	Using (GB) 0.44 / 31.25 (1.408000%) on TPU_24(process=6,(0,6,0,0))
I0419 18:18:51.773774 135611359127360 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0419 18:18:51.773799 135611359127360 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0419 18:18:51.773823 135611359127360 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0419 18:18:52.102863 135611359127360 metric_logger.py:196] completed step: 0, seconds: 15.214, TFLOP/s/device: 0.893, Tokens/s/device: 134.613, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277
I0419 18:18:52.346772 135611359127360 metric_logger.py:196] completed step: 1, seconds: 1.641, TFLOP/s/device: 8.281, Tokens/s/device: 1248.128, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277
I0419 18:18:52.787103 135611359127360 metric_logger.py:196] completed step: 2, seconds: 0.015, TFLOP/s/device: 913.912, Tokens/s/device: 137754.759, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820
I0419 18:18:53.016533 135611359127360 metric_logger.py:196] completed step: 3, seconds: 0.441, TFLOP/s/device: 30.815, Tokens/s/device: 4644.770, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199
I0419 18:18:53.476527 135611359127360 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.861, Tokens/s/device: 8721.463, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531
I0419 18:18:53.482656 135611359127360 metric_logger.py:196] completed step: 5, seconds: 0.230, TFLOP/s/device: 59.203, Tokens/s/device: 8923.747, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941
I0419 18:18:54.992011    2957 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0419 18:18:57.531855 135478523184896 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints/0/items/array_metadatas/process_6
I0419 18:19:23.349889 135611359127360 metric_logger.py:196] completed step: 6, seconds: 0.460, TFLOP/s/device: 29.518, Tokens/s/device: 4449.311, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468
I0419 18:19:23.579161 135611359127360 metric_logger.py:196] completed step: 7, seconds: 29.639, TFLOP/s/device: 0.458, Tokens/s/device: 69.098, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779
I0419 18:19:23.808754 135611359127360 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.062, Tokens/s/device: 8751.725, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453
I0419 18:19:24.037546 135611359127360 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0419 18:19:24.038269 135611359127360 checkpointing.py:776] Waited 0.0007441043853759766 seconds for step 9 to finish before starting checkpointing.
I0419 18:19:24.040277 135611359127360 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 18:19:25.643794 135479590754048 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 46.226 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.17137694358826 s) (per-host)
I0419 18:19:25.643918 135479590754048 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.872740s.
I0419 18:19:34.606903 135479590754048 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 42.835710s.
I0419 18:19:34.607111 135478514792192 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 18:19:34.607185 135478514792192 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0419 18:19:34.607234 135478514792192 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0419 18:19:34.609359 135478514792192 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0419 18:19:34.609536 135611359127360 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0419 18:19:34.609669 135611359127360 checkpoint_manager.py:1452] Waiting for previous save to complete took 10.569393 seconds. If this number is high, consider checkpointing less frequently.
I0419 18:19:34.611444 135611359127360 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0419 18:19:34.613358 135611359127360 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints/9.
I0419 18:19:34.893423 135611359127360 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0419 18:19:34.893514 135611359127360 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
I0419 18:19:34.926793 135611359127360 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035225s
I0419 18:19:34.926958 135611359127360 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 39.690 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03886580467224121 s) (per-host)
I0419 18:19:34.927009 135611359127360 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038926s (batch_requests_ready=0.001967s, total_serialization_initiated=0.036893s, others=0.000066s)
I0419 18:19:34.927104 135611359127360 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042995s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.042980)
I0419 18:19:34.927831 135611359127360 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints/9.
I0419 18:19:34.928137 135478514792192 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 18:39:34.928105
I0419 18:19:34.930103 135611359127360 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 18:19:34.930387 135478523184896 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0419 18:19:34.930567 135611359127360 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776622764.0402467, 'wait_for_prev_duration_secs': 10.569392919540405, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776622774.6114824, 'checkpointer_blocking_duration_secs': 0.3168060779571533, 'get_old_steps_start_time': 1776622774.9283116, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776622764.0385182, 'checkpoint_manager_blocking_duration_secs': 10.892014503479004}
I0419 18:19:34.930671 135611359127360 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0419 18:19:34.930714 135611359127360 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 18:19:40.546004 135455181879040 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_04_int8/checkpoints/9/items/array_metadatas/process_6
I0419 18:20:14.655169 135478514792192 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 39.721 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 39.76702833175659 s) (per-host)
I0419 18:20:14.655297 135478514792192 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.727041s.
I0419 18:20:26.440520 135478514792192 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 51.512249s.
I0419 18:20:26.440808 135478523184896 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 18:20:26.440961 135478523184896 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0419 18:20:26.441008 135478523184896 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0419 18:20:26.443790 135478523184896 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0419 18:20:26.443972 135611359127360 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0419 18:20:26.444140 135611359127360 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 18:20:26.445184 135611359127360 metric_logger.py:196] completed step: 9, seconds: 0.229, TFLOP/s/device: 59.284, Tokens/s/device: 8935.895, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3552.333
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Sun Apr 19 18:20:38 UTC 2026
EXIT_CODE=0
feat/nnx-trainstate-and-training-loop  ·  574ad3fb9  ·  feat_nnx_trainstate_and_training_loop_20260419_214525  ·  full log 
XPK Start: Sun Apr 19 22:01:49 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-19 22:02:13.891538: 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)
I0419 22:02:14.069380 133408327132992 max_utils.py:273] Attempting to initialize the jax distributed system...
I0419 22:02:23.109035 133408327132992 distributed.py:149] Starting JAX distributed service on [::]:8482
I0419 22:02:23.111433 133408327132992 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-eaedm-slice-job-0-0.mt-04-int8-eaedm:8482
I0419 22:02:24.679980 133408327132992 max_utils.py:284] Jax distributed system initialized!
I0419 22:02:30.700030 133408327132992 max_utils.py:800] System Information: Jax Version: 0.9.2
I0419 22:02:30.700132 133408327132992 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0419 22:02:30.700171 133408327132992 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
I0419 22:02:30.700202 133408327132992 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0419 22:02:31.385178 133408327132992 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 22:02:31.385777 133408327132992 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 22:02:31.386065 133408327132992 checkpointing.py:688] Setting up checkpoint logger...
I0419 22:02:31.386116 133408327132992 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0419 22:02:31.386158 133408327132992 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0419 22:02:31.386501 133408327132992 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 0x7955025766c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0419 22:02:35.684791 133408327132992 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0419 22:02:35.685023 133408327132992 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 0x7940d46ff1a0>}, handler_registry=None
I0419 22:02:35.685258 133408327132992 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7940d46ff1a0>` 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`.
I0419 22:02:35.685305 133408327132992 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7940d4701370>` 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`.
I0419 22:02:35.685351 133408327132992 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 0x7940d46ff1a0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7940d46ff1a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7940d4701370>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7940d4701370>}).
I0419 22:02:35.685678 133408327132992 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0419 22:02:35.685910 133408327132992 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7940bc4642c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0419 22:02:36.786070 133408327132992 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints
I0419 22:02:38.097432 133408327132992 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_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7953c8d5c9b0>
I0419 22:02:38.097602 133408327132992 checkpointing.py:302] Checkpoint manager created!
I0419 22:02:39.253994 133408327132992 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 22:02:39.254101 133408327132992 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 22:02:39.634598 133408327132992 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 22:02:39.634693 133408327132992 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 22:02:39.650911 133408327132992 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 22:02:39.650974 133408327132992 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 22:02:39.716893 133408327132992 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 22:02:39.716975 133408327132992 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 22:02:39.733181 133408327132992 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 22:02:39.733246 133408327132992 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 22:02:39.749296 133408327132992 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 22:02:39.749366 133408327132992 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 22:02:39.774178 133408327132992 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0419 22:02:39.774240 133408327132992 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 22:02:39.840007 133408327132992 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0419 22:02:39.840086 133408327132992 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0419 22:02:40.356841 133408327132992 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0419 22:02:40.356947 133408327132992 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0419 22:02:42.837694 133408327132992 maxtext_utils.py:1821]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0419 22:02:42.837824 133408327132992 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0419 22:02:42.837877 133408327132992 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0419 22:02:42.837934 133408327132992 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0419 22:02:42.837985 133408327132992 maxtext_utils.py:1821]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0419 22:02:42.838023 133408327132992 maxtext_utils.py:1821]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0419 22:02:42.838079 133408327132992 maxtext_utils.py:1821]  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)
I0419 22:02:42.838132 133408327132992 maxtext_utils.py:1821]  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')
I0419 22:02:42.838173 133408327132992 maxtext_utils.py:1821]  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)
I0419 22:02:42.838208 133408327132992 maxtext_utils.py:1821]  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)
I0419 22:02:42.838255 133408327132992 maxtext_utils.py:1821]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0419 22:02:42.838301 133408327132992 maxtext_utils.py:1821]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0419 22:02:44.234514 133408327132992 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0419 22:02:44.234616 133408327132992 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0419 22:02:44.249907 133408327132992 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0419 22:02:44.249973 133408327132992 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0419 22:02:57.583684 133408327132992 max_utils.py:791] Total memory size: 1.8 GB, Output size: 0.4 GB, Temp size: 1.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0419 22:02:57.584446 133408327132992 metric_logger.py:301] number parameters: 1.104 billion
I0419 22:03:12.781171 133408327132992 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0419 22:03:13.021297 133408327132992 checkpointing.py:798] Waited 0.24010777473449707 seconds for step 0 to finish before starting checkpointing.
I0419 22:03:13.023767 133408327132992 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 22:03:13.025557 133408327132992 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0
I0419 22:03:13.027055 133408327132992 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints/0.
I0419 22:03:13.783894 133408327132992 signaling_client.py:364] Using JaxDistributedSignalingClient
I0419 22:03:13.784799 133408327132992 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0419 22:03:13.784856 133408327132992 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
I0419 22:03:14.058114 133408327132992 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274297s
I0419 22:03:14.058289 133408327132992 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.507 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28012514114379883 s) (per-host)
I0419 22:03:14.058366 133408327132992 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.280212s (batch_requests_ready=0.002276s, total_serialization_initiated=0.277841s, others=0.000094s)
I0419 22:03:14.058461 133408327132992 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.284636s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.284618)
I0419 22:03:14.059246 133408327132992 event_tracking.py:125] [process=4] [async] Finished blocking save in 1.03 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints/0.
I0419 22:03:14.059628 133276928833280 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 22:23:14.059590
I0419 22:03:14.075078 133408327132992 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 22:03:14.075367 133276882683648 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0419 22:03:14.075525 133408327132992 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776636193.023749, 'wait_for_prev_duration_secs': 6.222724914550781e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776636193.025597, 'checkpointer_blocking_duration_secs': 1.0341908931732178, 'get_old_steps_start_time': 1776636194.059817, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776636193.0217664, 'checkpoint_manager_blocking_duration_secs': 1.0537173748016357}
I0419 22:03:14.075632 133408327132992 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0419 22:03:14.075684 133408327132992 max_utils.py:750] 
Memstats: After params initialized:
I0419 22:03:14.075735 133408327132992 max_utils.py:756] 	Using (GB) 0.44 / 31.25 (1.408000%) on TPU_16(process=4,(0,4,0,0))
I0419 22:03:14.075768 133408327132992 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0))
I0419 22:03:14.075797 133408327132992 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0))
I0419 22:03:14.075830 133408327132992 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0))
I0419 22:03:14.394725 133408327132992 metric_logger.py:196] completed step: 0, seconds: 15.197, TFLOP/s/device: 0.894, Tokens/s/device: 134.767, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277
I0419 22:03:14.642265 133408327132992 metric_logger.py:196] completed step: 1, seconds: 1.612, TFLOP/s/device: 8.429, Tokens/s/device: 1270.510, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277
I0419 22:03:15.052067 133408327132992 metric_logger.py:196] completed step: 2, seconds: 0.019, TFLOP/s/device: 734.003, Tokens/s/device: 110636.919, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820
I0419 22:03:15.281387 133408327132992 metric_logger.py:196] completed step: 3, seconds: 0.410, TFLOP/s/device: 33.127, Tokens/s/device: 4993.307, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199
I0419 22:03:15.740664 133408327132992 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.844, Tokens/s/device: 8718.826, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531
I0419 22:03:15.746777 133408327132992 metric_logger.py:196] completed step: 5, seconds: 0.229, TFLOP/s/device: 59.286, Tokens/s/device: 8936.207, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941
I0419 22:03:18.263079    2955 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0419 22:03:23.388107 133277969512192 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints/0/items/array_metadatas/process_4
I0419 22:03:45.845268 133408327132992 metric_logger.py:196] completed step: 6, seconds: 0.460, TFLOP/s/device: 29.542, Tokens/s/device: 4452.939, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468
I0419 22:03:46.074641 133408327132992 metric_logger.py:196] completed step: 7, seconds: 29.870, TFLOP/s/device: 0.455, Tokens/s/device: 68.563, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779
I0419 22:03:46.304073 133408327132992 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.132, Tokens/s/device: 8762.285, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453
I0419 22:03:46.532740 133408327132992 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0419 22:03:46.533450 133408327132992 checkpointing.py:798] Waited 0.0007302761077880859 seconds for step 9 to finish before starting checkpointing.
I0419 22:03:46.535596 133408327132992 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 22:03:47.840695 133276928833280 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 46.373 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.062495946884155 s) (per-host)
I0419 22:03:47.840803 133276928833280 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.781060s.
I0419 22:03:59.772696 133276928833280 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 45.712937s.
I0419 22:03:59.772989 133276882683648 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 22:03:59.773105 133276882683648 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0419 22:03:59.773154 133276882683648 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0419 22:03:59.774971 133276882683648 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0419 22:03:59.775197 133408327132992 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.
W0419 22:03:59.775350 133408327132992 checkpoint_manager.py:1452] Waiting for previous save to complete took 13.239743 seconds. If this number is high, consider checkpointing less frequently.
I0419 22:03:59.777105 133408327132992 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9
I0419 22:03:59.779182 133408327132992 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints/9.
I0419 22:04:00.916782 133408327132992 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0419 22:04:00.916876 133408327132992 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
I0419 22:04:00.953487 133408327132992 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037850s
I0419 22:04:00.953654 133408327132992 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 37.310 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.041344642639160156 s) (per-host)
I0419 22:04:00.953705 133408327132992 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.041405s (batch_requests_ready=0.001803s, total_serialization_initiated=0.039535s, others=0.000067s)
I0419 22:04:00.953794 133408327132992 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.045404s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.045390)
I0419 22:04:00.954516 133408327132992 event_tracking.py:125] [process=4] [async] Finished blocking save in 1.18 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints/9.
I0419 22:04:00.954874 133276882683648 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 22:24:00.954838
I0419 22:04:00.956850 133408327132992 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 22:04:00.957087 133277969512192 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0419 22:04:00.957196 133408327132992 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776636226.5355608, 'wait_for_prev_duration_secs': 13.23974323272705, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776636239.777144, 'checkpointer_blocking_duration_secs': 1.177872896194458, 'get_old_steps_start_time': 1776636240.955041, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776636226.5336819, 'checkpoint_manager_blocking_duration_secs': 14.423478841781616}
I0419 22:04:00.957370 133408327132992 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0419 22:04:00.957417 133408327132992 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 22:04:08.661182 133252811572992 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_04_int8/checkpoints/9/items/array_metadatas/process_4
I0419 22:04:43.668704 133276882683648 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 36.944 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.75637102127075 s) (per-host)
I0419 22:04:43.668811 133276882683648 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.713823s.
I0419 22:04:55.271431 133276882683648 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 54.316428s.
I0419 22:04:55.271738 133277969512192 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 22:04:55.271879 133277969512192 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0419 22:04:55.271926 133277969512192 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0419 22:04:55.274686 133277969512192 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0419 22:04:55.274869 133408327132992 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.
I0419 22:04:55.275017 133408327132992 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 22:04:55.275976 133408327132992 metric_logger.py:196] completed step: 9, seconds: 0.229, TFLOP/s/device: 59.232, Tokens/s/device: 8928.065, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3552.333
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Sun Apr 19 22:05:04 UTC 2026
EXIT_CODE=0