MaxView

Timing Test Reference Updates
‹ 02_syntheticCase: 03_dropout04_int8 ›

Metrics: main (586e69205) vs feat/nnx-set-defaults-true (73213e044)

Metricmain  586e69205feat/nnx-set-defaults-true  73213e044Diff (feat/nnx-set-defaults-true − main)
Parameters1.104 billion1.104 billion—
Final loss6.98106.98100
TFLOP/s11.31211.111-0.201
Tok/s1807.91775.8-32.155
Avg s/step1.4651.290-0.175
Memory %1.311.310
JAX0.9.20.8.1—

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

main  ·  586e69205  ·  main_20260423_071538  ·  full log 
XPK Start: Thu Apr 23 07:23:38 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-23 07:24:12.112707: 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)
I0423 07:24:12.327553 137695362717504 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 07:24:21.369910 137695362717504 distributed.py:149] Starting JAX distributed service on [::]:8482
I0423 07:24:21.372273 137695362717504 distributed.py:172] Connecting to JAX distributed service on mt-03-dropout-1qzac-slice-job-0-0.mt-03-dropout-1qzac:8482
I0423 07:27:08.319933 137695362717504 max_utils.py:284] Jax distributed system initialized!
I0423 07:27:14.538051 137695362717504 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 07:27:14.538174 137695362717504 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 07:27:14.538217 137695362717504 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
I0423 07:27:14.538252 137695362717504 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 07:27:15.233934 137695362717504 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 07:27:15.234220 137695362717504 checkpointing.py:677] Setting up checkpoint logger...
I0423 07:27:15.234276 137695362717504 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 07:27:15.234320 137695362717504 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0423 07:27:15.234661 137695362717504 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 0x7d3b1a31ffb0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 07:27:18.081327 137695362717504 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0423 07:27:18.081564 137695362717504 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 0x7d25a872b5f0>}, handler_registry=None
I0423 07:27:18.081799 137695362717504 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d25a872b5f0>` 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`.
I0423 07:27:18.081849 137695362717504 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d25a872f020>` 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`.
I0423 07:27:18.081883 137695362717504 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 0x7d25a872b5f0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d25a872b5f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d25a872f020>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d25a872f020>}).
I0423 07:27:18.082204 137695362717504 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0423 07:27:18.082275 137695362717504 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d25a86d5b20> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0423 07:27:19.690536 137695362717504 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints
I0423 07:27:20.128229 137695362717504 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_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d25a872d4f0>
I0423 07:27:20.128398 137695362717504 checkpointing.py:301] Checkpoint manager created!
I0423 07:27:20.477376 137695362717504 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 07:27:20.477490 137695362717504 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0423 07:27:20.863691 137695362717504 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,128,2048]....................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0423 07:27:20.863788 137695362717504 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0423 07:27:20.880234 137695362717504 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,128,2048]....................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0423 07:27:20.880295 137695362717504 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0423 07:27:20.903953 137695362717504 attentions.py:1154] attentions/query Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 07:27:20.904021 137695362717504 attentions.py:1154] attentions/query Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0423 07:27:20.920228 137695362717504 attentions.py:1155] attentions/key Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 07:27:20.920288 137695362717504 attentions.py:1155] attentions/key Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0423 07:27:20.936294 137695362717504 attentions.py:1156] attentions/value Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 07:27:20.936355 137695362717504 attentions.py:1156] attentions/value Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0423 07:27:20.960750 137695362717504 attentions.py:1198] attentions/out Logical: bfloat16[32,128,16,128]..................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0423 07:27:20.960824 137695362717504 attentions.py:1198] attentions/out Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0423 07:27:20.987699 137695362717504 linears.py:525] linears/x Logical: bfloat16[32,128,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 07:27:20.987769 137695362717504 linears.py:525] linears/x Physical: bfloat16[32,128,7168]....................................... ('fsdp', None, None).
I0423 07:27:21.204517 137695362717504 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 07:27:21.204629 137695362717504 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0423 07:27:22.662479 137695362717504 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0423 07:27:22.662636 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 07:27:22.662710 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 07:27:22.662805 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 07:27:22.662897 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 07:27:22.662977 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 07:27:22.663074 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 07:27:22.663192 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0423 07:27:22.663272 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0423 07:27:22.663344 137695362717504 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 07:27:22.663438 137695362717504 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0423 07:27:22.663538 137695362717504 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0423 07:27:23.181473 137695362717504 train.py:155] train/xent Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0423 07:27:23.181568 137695362717504 train.py:155] train/xent Physical: float32[32,128]............................................. ('fsdp', None).
I0423 07:27:23.196985 137695362717504 train.py:162] train/z_loss Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0423 07:27:23.197047 137695362717504 train.py:162] train/z_loss Physical: float32[32,128]............................................. ('fsdp', None).
I0423 07:27:26.828001 137695362717504 max_utils.py:791] Total memory size: 0.8 GB, Output size: 0.4 GB, Temp size: 0.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0423 07:27:26.828735 137695362717504 metric_logger.py:301] number parameters: 1.104 billion
I0423 07:27:31.389015 137695362717504 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0423 07:27:31.463856 137695362717504 checkpointing.py:776] Waited 0.07482481002807617 seconds for step 0 to finish before starting checkpointing.
I0423 07:27:31.466519 137695362717504 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 07:27:31.468332 137695362717504 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0423 07:27:31.469736 137695362717504 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints/0.
I0423 07:27:32.184012 137695362717504 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 07:27:32.185111 137695362717504 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 07:27:32.185169 137695362717504 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
I0423 07:27:32.461966 137695362717504 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278025s
I0423 07:27:32.462155 137695362717504 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.442 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2834818363189697 s) (per-host)
I0423 07:27:32.462211 137695362717504 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.283548s (batch_requests_ready=0.002177s, total_serialization_initiated=0.281297s, others=0.000074s)
I0423 07:27:32.462304 137695362717504 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287669s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.287652)
I0423 07:27:32.463088 137695362717504 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.99 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints/0.
I0423 07:27:32.463353 137565923440384 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 07:47:32.463316
I0423 07:27:32.911831 137695362717504 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 07:27:32.912237 137564801902336 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 07:27:32.912406 137695362717504 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776929251.4665015, 'wait_for_prev_duration_secs': 5.9604644775390625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776929251.4683702, 'checkpointer_blocking_duration_secs': 0.9951303005218506, 'get_old_steps_start_time': 1776929252.463522, 'get_old_steps_duration_secs': 2.6464462280273438e-05, 'checkpoint_manager_blocking_start_time': 1776929251.4644022, 'checkpoint_manager_blocking_duration_secs': 1.447960376739502}
I0423 07:27:32.912519 137695362717504 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0423 07:27:32.912581 137695362717504 max_utils.py:750] 
Memstats: After params initialized:
I0423 07:27:32.912642 137695362717504 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_24(process=6,(0,6,0,0))
I0423 07:27:32.912677 137695362717504 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_25(process=6,(1,6,0,0))
I0423 07:27:32.912706 137695362717504 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_28(process=6,(0,7,0,0))
I0423 07:27:32.912731 137695362717504 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_29(process=6,(1,7,0,0))
I0423 07:27:33.225879 137695362717504 metric_logger.py:196] completed step: 0, seconds: 4.560, TFLOP/s/device: 0.176, Tokens/s/device: 28.069, total_weights: 4096, loss: 10.889, lm_loss: 10.889, perplexity: 53560.699
I0423 07:27:33.323977 137695362717504 metric_logger.py:196] completed step: 1, seconds: 1.835, TFLOP/s/device: 0.436, Tokens/s/device: 69.748, total_weights: 4096, loss: 10.902, lm_loss: 10.902, perplexity: 54298.523
I0423 07:27:33.731970 137695362717504 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 28.335, Tokens/s/device: 4528.569, total_weights: 4096, loss: 9.956, lm_loss: 9.956, perplexity: 21082.051
I0423 07:27:33.802659 137695362717504 metric_logger.py:196] completed step: 3, seconds: 0.408, TFLOP/s/device: 1.962, Tokens/s/device: 313.644, total_weights: 4096, loss: 9.139, lm_loss: 9.139, perplexity: 9314.691
I0423 07:27:33.944547 137695362717504 metric_logger.py:196] completed step: 4, seconds: 0.076, TFLOP/s/device: 10.546, Tokens/s/device: 1685.586, total_weights: 4096, loss: 8.459, lm_loss: 8.459, perplexity: 4715.493
I0423 07:27:33.950472 137695362717504 metric_logger.py:196] completed step: 5, seconds: 0.071, TFLOP/s/device: 11.329, Tokens/s/device: 1810.595, total_weights: 4096, loss: 7.924, lm_loss: 7.924, perplexity: 2762.373
I0423 07:27:36.865986    2857 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 07:27:38.889534 137564810295040 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints/0/items/array_metadatas/process_6
I0423 07:27:44.499533 137695362717504 metric_logger.py:196] completed step: 6, seconds: 0.142, TFLOP/s/device: 5.629, Tokens/s/device: 899.622, total_weights: 4096, loss: 7.553, lm_loss: 7.553, perplexity: 1906.617
I0423 07:27:44.570355 137695362717504 metric_logger.py:196] completed step: 7, seconds: 10.480, TFLOP/s/device: 0.076, Tokens/s/device: 12.214, total_weights: 4096, loss: 7.267, lm_loss: 7.267, perplexity: 1431.653
I0423 07:27:44.640942 137695362717504 metric_logger.py:196] completed step: 8, seconds: 0.075, TFLOP/s/device: 10.710, Tokens/s/device: 1711.733, total_weights: 4096, loss: 7.111, lm_loss: 7.111, perplexity: 1225.983
I0423 07:27:44.712214 137695362717504 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0423 07:27:44.712855 137695362717504 checkpointing.py:776] Waited 0.0006546974182128906 seconds for step 9 to finish before starting checkpointing.
I0423 07:27:44.714940 137695362717504 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 07:28:06.127242 137565923440384 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 46.529 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 33.948530197143555 s) (per-host)
I0423 07:28:06.127369 137565923440384 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.663904s.
I0423 07:28:16.901794 137565923440384 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 44.438312s.
I0423 07:28:16.902066 137564801902336 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 07:28:16.902198 137564801902336 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 07:28:16.902256 137564801902336 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 07:28:16.905006 137564801902336 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 07:28:16.905195 137695362717504 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.
W0423 07:28:16.905321 137695362717504 checkpoint_manager.py:1452] Waiting for previous save to complete took 32.190382 seconds. If this number is high, consider checkpointing less frequently.
I0423 07:28:16.907179 137695362717504 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0423 07:28:16.910282 137695362717504 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints/9.
I0423 07:28:17.654159 137695362717504 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 07:28:17.654251 137695362717504 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
I0423 07:28:17.687867 137695362717504 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034898s
I0423 07:28:17.688027 137695362717504 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 40.223 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.038350582122802734 s) (per-host)
I0423 07:28:17.688076 137695362717504 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038410s (batch_requests_ready=0.001794s, total_serialization_initiated=0.036550s, others=0.000066s)
I0423 07:28:17.688176 137695362717504 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042612s (all_items=0.000015s, per_item={'items': '0.00001550'}, temp_paths=0.042597)
I0423 07:28:17.688875 137695362717504 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints/9.
I0423 07:28:17.689182 137564801902336 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 07:48:17.689150
I0423 07:28:17.691946 137695362717504 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 07:28:17.692243 137564810295040 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 07:28:17.692413 137695362717504 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776929264.7149117, 'wait_for_prev_duration_secs': 32.19038224220276, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776929296.9072185, 'checkpointer_blocking_duration_secs': 0.7821056842803955, 'get_old_steps_start_time': 1776929297.6893463, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776929264.7131162, 'checkpoint_manager_blocking_duration_secs': 32.979262351989746}
I0423 07:28:17.692530 137695362717504 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0423 07:28:17.692574 137695362717504 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 07:28:24.520203 137542808508160 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_03_dropout/checkpoints/9/items/array_metadatas/process_6
I0423 07:28:59.779388 137564801902336 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.494 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.12967276573181 s) (per-host)
I0423 07:28:59.779528 137564801902336 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.090234s.
I0423 07:29:08.206967 137564801902336 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.517657s.
I0423 07:29:08.207246 137564810295040 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 07:29:08.207304 137564810295040 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 07:29:08.207349 137564810295040 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 07:29:08.209121 137564810295040 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 07:29:08.209291 137695362717504 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.
I0423 07:29:08.209454 137695362717504 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 07:29:08.210365 137695362717504 metric_logger.py:196] completed step: 9, seconds: 0.071, TFLOP/s/device: 11.312, Tokens/s/device: 1807.910, total_weights: 4096, loss: 6.981, lm_loss: 6.981, perplexity: 1075.867
Per train step:
 Total TFLOPs: 0.80 
 split as 99.60% learnable weight flops and 0.40% attention flops
XPK End: Thu Apr 23 07:29:19 UTC 2026
EXIT_CODE=0
feat/nnx-set-defaults-true  ·  73213e044  ·  feat_nnx_set_defaults_true_20260423_155242  ·  full log 
XPK Start: Thu Apr 23 16:00:44 UTC 2026
2026-04-23 16:00:48.151177: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1776960048.164236      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776960048.167996      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776960048.179652      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776960048.179670      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776960048.179673      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776960048.179675      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-23 16:01:15.177881: 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)
I0423 16:01:15.714601 134159164888896 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-23 16:01:24,756:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0423 16:01:24.756212 134159164888896 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-23 16:01:24,758:jax._src.distributed:157: Connecting to JAX distributed service on mt-03-dropout-y5zxo-slice-job-0-0.mt-03-dropout-y5zxo:8482
I0423 16:01:24.758472 134159164888896 distributed.py:157] Connecting to JAX distributed service on mt-03-dropout-y5zxo-slice-job-0-0.mt-03-dropout-y5zxo:8482
I0423 16:01:31.039340 134159164888896 max_utils.py:284] Jax distributed system initialized!
I0423 16:01:37.869545 134159164888896 max_utils.py:800] System Information: Jax Version: 0.8.1
I0423 16:01:37.869652 134159164888896 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0423 16:01:37.869692 134159164888896 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709
I0423 16:01:37.869726 134159164888896 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 16:01:38.485926 134159164888896 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 16:01:38.486544 134159164888896 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 16:01:38.486732 134159164888896 checkpointing.py:688] Setting up checkpoint logger...
I0423 16:01:38.486784 134159164888896 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 16:01:38.486828 134159164888896 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0423 16:01:38.487197 134159164888896 base_pytree_checkpoint_handler.py:415] 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 0x7a03cc7795b0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 16:01:41.467990 134159164888896 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0423 16:01:41.468259 134159164888896 checkpoint_manager.py:709] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a02981d7b60>}, handler_registry=None
I0423 16:01:41.468534 134159164888896 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a02981d7b60>` 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`.
I0423 16:01:41.468588 134159164888896 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79f15fd62e10>` 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`.
I0423 16:01:41.468633 134159164888896 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 0x7a02981d7b60>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a02981d7b60>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79f15fd62e10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79f15fd62e10>}).
I0423 16:01:41.469283 134159164888896 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0423 16:01:41.469385 134159164888896 async_checkpointer.py:177] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x79f15e934180> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0423 16:01:42.608809 134159164888896 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints
I0423 16:01:42.995692 134159164888896 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_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7a0284090da0>
I0423 16:01:42.995865 134159164888896 checkpointing.py:302] Checkpoint manager created!
I0423 16:01:43.313405 134159164888896 nnx_wrappers.py:455] Unknown Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 16:01:43.313519 134159164888896 nnx_wrappers.py:455] Unknown Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0423 16:01:43.686253 134159164888896 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 16:01:43.686355 134159164888896 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0423 16:01:43.702576 134159164888896 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 16:01:43.702636 134159164888896 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0423 16:01:43.726016 134159164888896 attentions.py:1150] attentions/query Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 16:01:43.726089 134159164888896 attentions.py:1150] attentions/query Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0423 16:01:43.742472 134159164888896 attentions.py:1151] attentions/key Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 16:01:43.742535 134159164888896 attentions.py:1151] attentions/key Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0423 16:01:43.758687 134159164888896 attentions.py:1152] attentions/value Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 16:01:43.758750 134159164888896 attentions.py:1152] attentions/value Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0423 16:01:43.784126 134159164888896 attentions.py:1193] attentions/out Logical: bfloat16[32,128,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 16:01:43.784201 134159164888896 attentions.py:1193] attentions/out Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0423 16:01:43.810584 134159164888896 linears.py:541] linears/x Logical: bfloat16[32,128,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 16:01:43.810654 134159164888896 linears.py:541] linears/x Physical: bfloat16[32,128,7168]....................................... ('fsdp', None, None).
I0423 16:01:44.012829 134159164888896 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 16:01:44.012955 134159164888896 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0423 16:01:45.472268 134159164888896 maxtext_utils.py:1740]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0423 16:01:45.472421 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 16:01:45.472477 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 16:01:45.472539 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   PartitionSpec('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 16:01:45.472595 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0423 16:01:45.472637 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0423 16:01:45.472694 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 16:01:45.472749 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   PartitionSpec('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0423 16:01:45.472793 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0423 16:01:45.472832 134159164888896 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 16:01:45.472882 134159164888896 maxtext_utils.py:1740]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed', 'vocab')
    Physical:  ('fsdp', None)

I0423 16:01:45.472932 134159164888896 maxtext_utils.py:1740]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed')
    Physical:  (None, 'fsdp')
I0423 16:01:46.005550 134159164888896 train.py:158] train/xent Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0423 16:01:46.005651 134159164888896 train.py:158] train/xent Physical: float32[32,128]............................................. ('fsdp', None).
I0423 16:01:46.020658 134159164888896 train.py:165] train/z_loss Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0423 16:01:46.020719 134159164888896 train.py:165] train/z_loss Physical: float32[32,128]............................................. ('fsdp', None).
I0423 16:01:49.296836 134159164888896 max_utils.py:791] Total memory size: 0.8 GB, Output size: 0.4 GB, Temp size: 0.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0423 16:01:49.297742 134159164888896 metric_logger.py:289] number parameters: 1.104 billion
I0423 16:01:53.327226 134159164888896 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0423 16:01:53.557003 134159164888896 checkpointing.py:798] Waited 0.22976255416870117 seconds for step 0 to finish before starting checkpointing.
I0423 16:01:53.560582 134159164888896 checkpoint_manager.py:2013] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 16:01:53.562588 134159164888896 checkpoint_manager.py:1518] [process=6] Saving checkpoint at step 0
I0423 16:01:53.565351 134159164888896 async_checkpointer.py:452] [process=6] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints/0.
I0423 16:01:54.322071 134159164888896 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 16:01:54.323262 134159164888896 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0423 16:01:54.323324 134159164888896 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
I0423 16:01:54.570469 134159164888896 base_pytree_checkpoint_handler.py:153] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.248480s
I0423 16:01:54.570645 134159164888896 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.139 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.721306562423706 s) (per-host)
I0423 16:01:54.570700 134159164888896 base_pytree_checkpoint_handler.py:737] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.721372s (batch_requests_ready=0.468881s, total_serialization_initiated=0.252417s, others=0.000073s)
I0423 16:01:54.570789 134159164888896 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.725496s (all_items=0.000021s, per_item={'items': '0.00002074'}, temp_paths=0.725475)
I0423 16:01:54.571824 134040988395264 async_checkpointer.py:79] [process=6][thread=async_save] Background save thread started.
I0423 16:01:54.572001 134159164888896 async_checkpointer.py:561] Finished blocking save. Time taken: 1.009341s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints/0.
I0423 16:01:54.573919 134159164888896 checkpoint_manager.py:1566] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 16:01:54.574168 134035099600640 async_checkpointer.py:265] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 16:01:54.574295 134159164888896 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776960113.5605643, 'wait_for_prev_duration_secs': 6.604194641113281e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776960113.5626268, 'checkpointer_blocking_duration_secs': 1.0094990730285645, 'get_old_steps_start_time': 1776960114.5721462, 'get_old_steps_duration_secs': 2.6702880859375e-05, 'checkpoint_manager_blocking_start_time': 1776960113.5576005, 'checkpoint_manager_blocking_duration_secs': 1.0166583061218262}
I0423 16:01:54.574395 134159164888896 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0423 16:01:54.574444 134159164888896 max_utils.py:750] 
Memstats: After params initialized:
I0423 16:01:54.574490 134159164888896 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_24(process=6,(0,6,0,0))
I0423 16:01:54.574523 134159164888896 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_25(process=6,(1,6,0,0))
I0423 16:01:54.574549 134159164888896 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_28(process=6,(0,7,0,0))
I0423 16:01:54.574578 134159164888896 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_29(process=6,(1,7,0,0))
I0423 16:01:54.839968 134159164888896 metric_logger.py:185] completed step: 0, seconds: 4.029, TFLOP/s/device: 0.199, Tokens/s/device: 31.767, total_weights: 4096, loss: 10.889
I0423 16:01:54.951903 134159164888896 metric_logger.py:185] completed step: 1, seconds: 1.511, TFLOP/s/device: 0.530, Tokens/s/device: 84.689, total_weights: 4096, loss: 10.902
I0423 16:01:55.355148 134159164888896 metric_logger.py:185] completed step: 2, seconds: 0.042, TFLOP/s/device: 18.997, Tokens/s/device: 3036.197, total_weights: 4096, loss: 9.956
I0423 16:01:55.426052 134159164888896 metric_logger.py:185] completed step: 3, seconds: 0.403, TFLOP/s/device: 1.986, Tokens/s/device: 317.419, total_weights: 4096, loss: 9.139
I0423 16:01:55.567731 134159164888896 metric_logger.py:185] completed step: 4, seconds: 0.076, TFLOP/s/device: 10.586, Tokens/s/device: 1691.868, total_weights: 4096, loss: 8.458
I0423 16:01:55.573287 134159164888896 metric_logger.py:185] completed step: 5, seconds: 0.071, TFLOP/s/device: 11.331, Tokens/s/device: 1810.979, total_weights: 4096, loss: 7.924
I0423 16:02:04.863271 134159164888896 metric_logger.py:185] completed step: 6, seconds: 0.142, TFLOP/s/device: 5.636, Tokens/s/device: 900.742, total_weights: 4096, loss: 7.553
I0423 16:02:04.935311 134159164888896 metric_logger.py:185] completed step: 7, seconds: 9.221, TFLOP/s/device: 0.087, Tokens/s/device: 13.882, total_weights: 4096, loss: 7.267
I0423 16:02:05.007443 134159164888896 metric_logger.py:185] completed step: 8, seconds: 0.075, TFLOP/s/device: 10.733, Tokens/s/device: 1715.335, total_weights: 4096, loss: 7.111
I0423 16:02:05.078835 134159164888896 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0423 16:02:05.079460 134159164888896 checkpointing.py:798] Waited 0.0006430149078369141 seconds for step 9 to finish before starting checkpointing.
I0423 16:02:05.081363 134159164888896 checkpoint_manager.py:2024] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 16:02:06.412570    2592 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 16:02:08.545531 134040996787968 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints/0/items/array_metadatas/process_6
I0423 16:02:36.344496 134040988395264 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/gbytes_per_sec: 37.171 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.495110750198364 s) (per-host)
I0423 16:02:36.344621 134040988395264 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.772684s.
I0423 16:02:44.581556 134040988395264 async_checkpointer.py:144] [process=6][thread=async_save] Background save thread done. Time taken: 50.009602s.
I0423 16:02:44.581855 134035099600640 async_checkpointer.py:273] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 16:02:44.581976 134035099600640 async_checkpointer.py:283] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 16:02:44.582026 134035099600640 checkpoint_manager.py:2133] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 16:02:44.584271 134035099600640 checkpoint_manager.py:2142] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 16:02:44.584537 134159164888896 checkpoint_manager.py:2036] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0423 16:02:44.584791 134159164888896 checkpoint_manager.py:1458] Waiting for previous save to complete took 39.503431 seconds. If this number is high, consider checkpointing less frequently.
I0423 16:02:44.586632 134159164888896 checkpoint_manager.py:1518] [process=6] Saving checkpoint at step 9
I0423 16:02:44.590083 134159164888896 async_checkpointer.py:452] [process=6] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints/9.
I0423 16:02:45.135917 134159164888896 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0423 16:02:45.136015 134159164888896 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
I0423 16:02:45.179642 134159164888896 base_pytree_checkpoint_handler.py:153] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.044724s
I0423 16:02:45.179811 134159164888896 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.060 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.304851770401001 s) (per-host)
I0423 16:02:45.179864 134159164888896 base_pytree_checkpoint_handler.py:737] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.304914s (batch_requests_ready=0.258289s, total_serialization_initiated=0.046554s, others=0.000072s)
I0423 16:02:45.179955 134159164888896 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.308800s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.308785)
I0423 16:02:45.180823 134037247182592 async_checkpointer.py:79] [process=6][thread=async_save] Background save thread started.
I0423 16:02:45.180988 134159164888896 async_checkpointer.py:561] Finished blocking save. Time taken: 0.594287s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints/9.
I0423 16:02:45.198037 134159164888896 checkpoint_manager.py:1566] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 16:02:45.198355 134035099600640 async_checkpointer.py:265] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 16:02:45.198521 134159164888896 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776960125.0813339, 'wait_for_prev_duration_secs': 39.50343108177185, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776960164.5866718, 'checkpointer_blocking_duration_secs': 0.5944337844848633, 'get_old_steps_start_time': 1776960165.1811254, 'get_old_steps_duration_secs': 2.6941299438476562e-05, 'checkpoint_manager_blocking_start_time': 1776960125.0796912, 'checkpoint_manager_blocking_duration_secs': 40.11879253387451}
I0423 16:02:45.198631 134159164888896 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0423 16:02:45.198677 134159164888896 checkpoint_manager.py:2024] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 16:02:50.187813 134006905390848 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_03_dropout/checkpoints/9/items/array_metadatas/process_6
I0423 16:03:26.888631 134037247182592 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/gbytes_per_sec: 37.597 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.01362705230713 s) (per-host)
I0423 16:03:26.888748 134037247182592 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.707811s.
I0423 16:03:32.836924 134037247182592 async_checkpointer.py:144] [process=6][thread=async_save] Background save thread done. Time taken: 47.655970s.
I0423 16:03:32.837220 134035099600640 async_checkpointer.py:273] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 16:03:32.837339 134035099600640 async_checkpointer.py:283] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 16:03:32.837387 134035099600640 checkpoint_manager.py:2133] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 16:03:32.838714 134035099600640 checkpoint_manager.py:2142] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 16:03:32.838895 134159164888896 checkpoint_manager.py:2036] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0423 16:03:32.839033 134159164888896 checkpoint_manager.py:2013] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 16:03:32.839797 134159164888896 metric_logger.py:185] completed step: 9, seconds: 0.072, TFLOP/s/device: 11.111, Tokens/s/device: 1775.755, total_weights: 4096, loss: 6.981
Per train step:
 Total TFLOPs: 0.80 
 split as 99.60% learnable weight flops and 0.40% attention flops
XPK End: Thu Apr 23 16:03:44 UTC 2026
EXIT_CODE=0