MaxView

Timing Test Reference Updates
‹ 05_fp8Case: 06_grad_accum07_eval ›

Metrics: main (b117f50cf) vs feat/nnx-set-defaults-true (73213e044)

Metricmain  b117f50cffeat/nnx-set-defaults-true  73213e044Diff (feat/nnx-set-defaults-true − main)
Parameters1.104 billion1.104 billion—
Final loss9.45709.45700
TFLOP/s94.00494.001-0.003
Tok/s14169.314168.8-0.465
Avg s/step6.9455.941-1.004
Memory %1.381.380
JAX0.9.20.8.1—

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

main  ·  b117f50cf  ·  main_20260424_070227  ·  full log 
XPK Start: Fri Apr 24 07:18:45 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-24 07:19:11.244584: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0424 07:19:11.454986 136837314082624 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 07:19:20.496431 136837314082624 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 07:19:20.498794 136837314082624 distributed.py:172] Connecting to JAX distributed service on mt-06-grad-accum-0qdjx-slice-job-0-0.mt-06-grad-accum-0qdjx:8482
I0424 07:19:21.359517 136837314082624 max_utils.py:284] Jax distributed system initialized!
I0424 07:19:27.359252 136837314082624 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 07:19:27.359357 136837314082624 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 07:19:27.359398 136837314082624 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0424 07:19:27.359436 136837314082624 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 07:19:28.048606 136837314082624 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 07:19:28.048912 136837314082624 checkpointing.py:677] Setting up checkpoint logger...
I0424 07:19:28.048967 136837314082624 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 07:19:28.049019 136837314082624 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 07:19:28.049352 136837314082624 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 0x7c734d918ec0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 07:19:31.088770 136837314082624 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0424 07:19:31.089003 136837314082624 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c5eec265130>}, handler_registry=None
I0424 07:19:31.089243 136837314082624 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c5eec265130>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0424 07:19:31.089292 136837314082624 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c5eec26cce0>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0424 07:19:31.089329 136837314082624 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 0x7c5eec265130>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c5eec265130>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c5eec26cce0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c5eec26cce0>}).
I0424 07:19:31.089673 136837314082624 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 07:19:31.089743 136837314082624 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c5ecc5ed800> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 07:19:32.009289 136837314082624 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints
I0424 07:19:32.014694 136837314082624 checkpoint_manager.py:929] [process=5][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_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c5eec267050>
I0424 07:19:32.014808 136837314082624 checkpointing.py:301] Checkpoint manager created!
I0424 07:19:33.714348 136837314082624 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 07:19:33.714462 136837314082624 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:19:34.095364 136837314082624 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 07:19:34.095458 136837314082624 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:19:34.112244 136837314082624 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 07:19:34.112307 136837314082624 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:19:34.136965 136837314082624 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 07:19:34.137034 136837314082624 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:19:34.153767 136837314082624 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 07:19:34.153826 136837314082624 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:19:34.170581 136837314082624 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 07:19:34.170642 136837314082624 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:19:34.195411 136837314082624 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0424 07:19:34.195493 136837314082624 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:19:34.216834 136837314082624 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 07:19:34.216899 136837314082624 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 07:19:34.429572 136837314082624 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 07:19:34.429694 136837314082624 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 07:19:35.874355 136837314082624 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 07:19:35.874487 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 07:19:35.874540 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 07:19:35.874599 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0424 07:19:35.874662 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 07:19:35.874807 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 07:19:35.875104 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0424 07:19:35.875189 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0424 07:19:35.875249 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0424 07:19:35.875294 136837314082624 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0424 07:19:35.875353 136837314082624 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0424 07:19:35.875416 136837314082624 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0424 07:19:35.902098 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown.
I0424 07:19:35.902163 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,).
I0424 07:19:35.917338 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown.
I0424 07:19:35.917392 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None).
I0424 07:19:35.947381 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown.
I0424 07:19:35.947451 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp').
I0424 07:19:35.962607 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown.
I0424 07:19:35.962678 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None).
I0424 07:19:35.992775 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown.
I0424 07:19:35.992836 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None).
I0424 07:19:36.007949 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown.
I0424 07:19:36.008005 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp').
I0424 07:19:36.052964 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown.
I0424 07:19:36.053038 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None).
I0424 07:19:36.068126 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown.
I0424 07:19:36.068180 136837314082624 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp').
I0424 07:19:36.752560 136837314082624 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 07:19:36.752664 136837314082624 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 07:19:36.768505 136837314082624 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 07:19:36.768564 136837314082624 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 07:19:48.080011 136837314082624 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0424 07:19:48.080759 136837314082624 metric_logger.py:301] number parameters: 1.104 billion
I0424 07:19:50.010352 136837314082624 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0424 07:20:01.217697 136837314082624 checkpointing.py:776] Waited 11.20729374885559 seconds for step 0 to finish before starting checkpointing.
I0424 07:20:01.219984 136837314082624 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:20:01.221640 136837314082624 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0
I0424 07:20:01.223084 136837314082624 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints/0.
I0424 07:20:01.837968 136837314082624 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 07:20:01.838947 136837314082624 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 07:20:01.839003 136837314082624 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 07:20:02.110103 136837314082624 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.272207s
I0424 07:20:02.110280 136837314082624 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.558 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27755045890808105 s) (per-host)
I0424 07:20:02.110333 136837314082624 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.277615s (batch_requests_ready=0.002177s, total_serialization_initiated=0.275365s, others=0.000072s)
I0424 07:20:02.110427 136837314082624 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.281822s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.281804)
I0424 07:20:02.111225 136837314082624 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.89 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints/0.
I0424 07:20:02.111557 136708374304512 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 07:40:02.111518
I0424 07:20:02.123533 136837314082624 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 07:20:02.123841 136705692657408 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 07:20:02.124005 136837314082624 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777015201.219966, 'wait_for_prev_duration_secs': 6.008148193359375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777015201.2216926, 'checkpointer_blocking_duration_secs': 0.8899738788604736, 'get_old_steps_start_time': 1777015202.1116915, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1777015201.2182217, 'checkpoint_manager_blocking_duration_secs': 0.9057421684265137}
I0424 07:20:02.124112 136837314082624 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0424 07:20:02.124161 136837314082624 max_utils.py:750] 
Memstats: After params initialized:
I0424 07:20:02.124207 136837314082624 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0))
I0424 07:20:02.124239 136837314082624 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0))
I0424 07:20:02.124266 136837314082624 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0))
I0424 07:20:02.124291 136837314082624 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0))
I0424 07:20:02.438288 136837314082624 metric_logger.py:196] completed step: 0, seconds: 1.929, TFLOP/s/device: 28.167, Tokens/s/device: 4245.686, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059
I0424 07:20:03.057926 136837314082624 metric_logger.py:196] completed step: 1, seconds: 12.426, TFLOP/s/device: 4.374, Tokens/s/device: 659.239, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059
I0424 07:20:03.636484 136837314082624 metric_logger.py:196] completed step: 2, seconds: 0.041, TFLOP/s/device: 1323.314, Tokens/s/device: 199464.329, total_weights: 262144, loss: 10.563, lm_loss: 10.563, perplexity: 38662.707
I0424 07:20:04.214736 136837314082624 metric_logger.py:196] completed step: 3, seconds: 0.585, TFLOP/s/device: 92.969, Tokens/s/device: 14013.240, total_weights: 262144, loss: 10.272, lm_loss: 10.272, perplexity: 28909.668
I0424 07:20:05.108313    2657 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 07:20:05.371414 136837314082624 metric_logger.py:196] completed step: 4, seconds: 0.578, TFLOP/s/device: 93.949, Tokens/s/device: 14161.079, total_weights: 262144, loss: 10.022, lm_loss: 10.022, perplexity: 22524.992
I0424 07:20:05.377807 136837314082624 metric_logger.py:196] completed step: 5, seconds: 0.579, TFLOP/s/device: 93.910, Tokens/s/device: 14155.206, total_weights: 262144, loss: 9.820, lm_loss: 9.820, perplexity: 18401.865
I0424 07:20:07.230983 136706737465088 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints/0/items/array_metadatas/process_5
I0424 07:20:39.498754 136708374304512 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 41.937 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.665985345840454 s) (per-host)
I0424 07:20:39.498878 136708374304512 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.386669s.
I0424 07:20:48.623524 136708374304512 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 46.511299s.
I0424 07:20:48.623833 136705692657408 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 07:20:48.623965 136705692657408 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 07:20:48.624016 136705692657408 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 07:20:48.626894 136705692657408 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 07:20:51.936732 136837314082624 metric_logger.py:196] completed step: 6, seconds: 1.157, TFLOP/s/device: 46.972, Tokens/s/device: 7080.178, total_weights: 262144, loss: 9.667, lm_loss: 9.667, perplexity: 15787.604
I0424 07:20:52.514712 136837314082624 metric_logger.py:196] completed step: 7, seconds: 45.981, TFLOP/s/device: 1.182, Tokens/s/device: 178.159, total_weights: 262144, loss: 9.561, lm_loss: 9.561, perplexity: 14203.827
I0424 07:20:53.092909 136837314082624 metric_logger.py:196] completed step: 8, seconds: 0.583, TFLOP/s/device: 93.195, Tokens/s/device: 14047.386, total_weights: 262144, loss: 9.496, lm_loss: 9.496, perplexity: 13302.920
I0424 07:20:53.670548 136837314082624 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0424 07:20:53.671236 136837314082624 checkpointing.py:776] Waited 0.0007088184356689453 seconds for step 9 to finish before starting checkpointing.
I0424 07:20:53.673281 136837314082624 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:20:53.674978 136837314082624 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9
I0424 07:20:53.676365 136837314082624 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints/9.
I0424 07:20:53.963263 136837314082624 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 07:20:53.963353 136837314082624 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 07:20:53.999104 136837314082624 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036648s
I0424 07:20:53.999268 136837314082624 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 37.885 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.040717124938964844 s) (per-host)
I0424 07:20:53.999319 136837314082624 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040779s (batch_requests_ready=0.001641s, total_serialization_initiated=0.039071s, others=0.000067s)
I0424 07:20:53.999407 136837314082624 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044884s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.044868)
I0424 07:20:54.000066 136837314082624 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints/9.
I0424 07:20:54.000412 136707869505280 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 07:40:54.000372
I0424 07:20:54.002400 136837314082624 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 07:20:54.002689 136701963925248 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 07:20:54.002817 136837314082624 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777015253.6732478, 'wait_for_prev_duration_secs': 7.271766662597656e-05, 'time_between_consecutive_saves_sec': 5.046308279037476, 'checkpointer_blocking_start_time': 1777015253.675019, 'checkpointer_blocking_duration_secs': 0.32553744316101074, 'get_old_steps_start_time': 1777015254.00058, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1777015253.6714602, 'checkpoint_manager_blocking_duration_secs': 0.33132505416870117}
I0424 07:20:54.002977 136837314082624 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0424 07:20:54.003029 136837314082624 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 07:20:59.414666 136705692657408 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_06_grad_accum/checkpoints/9/items/array_metadatas/process_5
I0424 07:21:36.414484 136707869505280 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.206 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.45591449737549 s) (per-host)
I0424 07:21:36.414589 136707869505280 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.414068s.
I0424 07:21:44.204477 136707869505280 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.203939s.
I0424 07:21:44.204817 136701963925248 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 07:21:44.204941 136701963925248 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 07:21:44.204990 136701963925248 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 07:21:44.207411 136701963925248 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 07:21:44.207587 136837314082624 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 07:21:44.207747 136837314082624 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:21:44.208695 136837314082624 metric_logger.py:196] completed step: 9, seconds: 0.578, TFLOP/s/device: 94.004, Tokens/s/device: 14169.284, total_weights: 262144, loss: 9.457, lm_loss: 9.457, perplexity: 12802.546
Per train step:
 Total TFLOPs: 54.35 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 24 07:21:56 UTC 2026
EXIT_CODE=0
feat/nnx-set-defaults-true  ·  73213e044  ·  feat_nnx_set_defaults_true_20260424_145744  ·  full log 
XPK Start: Fri Apr 24 15:12:43 UTC 2026
2026-04-24 15:12:47.662492: 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 00:00:1777043567.675423      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1777043567.679198      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1777043567.690950      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777043567.690971      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777043567.690974      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777043567.690976      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-24 15:13:06.859986: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0424 15:13:07.389020 133694480537408 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-24 15:13:16,430:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0424 15:13:16.430072 133694480537408 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-24 15:13:16,432:jax._src.distributed:157: Connecting to JAX distributed service on mt-06-grad-accum-wmzf8-slice-job-0-0.mt-06-grad-accum-wmzf8:8482
I0424 15:13:16.432368 133694480537408 distributed.py:157] Connecting to JAX distributed service on mt-06-grad-accum-wmzf8-slice-job-0-0.mt-06-grad-accum-wmzf8:8482
I0424 15:13:17.786119 133694480537408 max_utils.py:284] Jax distributed system initialized!
I0424 15:13:23.587427 133694480537408 max_utils.py:800] System Information: Jax Version: 0.8.1
I0424 15:13:23.587532 133694480537408 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0424 15:13:23.587576 133694480537408 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
I0424 15:13:23.587611 133694480537408 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 15:13:24.208127 133694480537408 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 15:13:24.208792 133694480537408 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 15:13:24.208995 133694480537408 checkpointing.py:688] Setting up checkpoint logger...
I0424 15:13:24.209047 133694480537408 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 15:13:24.209090 133694480537408 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0424 15:13:24.209456 133694480537408 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 0x7997c39b9a90>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 15:13:27.132443 133694480537408 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 15:13:27.132784 133694480537408 checkpoint_manager.py:709] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x799666fee180>}, handler_registry=None
I0424 15:13:27.133031 133694480537408 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x799666fee180>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0424 15:13:27.133081 133694480537408 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7985512b6780>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0424 15:13:27.133116 133694480537408 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 0x799666fee180>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x799666fee180>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7985512b6780>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7985512b6780>}).
I0424 15:13:27.133696 133694480537408 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0424 15:13:27.133776 133694480537408 async_checkpointer.py:177] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x798532b80180> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0424 15:13:27.847161 133694480537408 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints
I0424 15:13:27.849312 133694480537408 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_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x79965d3aba70>
I0424 15:13:27.849424 133694480537408 checkpointing.py:302] Checkpoint manager created!
I0424 15:13:29.543695 133694480537408 nnx_wrappers.py:455] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 15:13:29.543802 133694480537408 nnx_wrappers.py:455] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 15:13:29.915127 133694480537408 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 15:13:29.915224 133694480537408 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 15:13:29.931131 133694480537408 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 15:13:29.931189 133694480537408 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 15:13:29.953920 133694480537408 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 15:13:29.953986 133694480537408 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 15:13:29.974950 133694480537408 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 15:13:29.975053 133694480537408 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 15:13:29.991185 133694480537408 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 15:13:29.991250 133694480537408 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 15:13:30.016500 133694480537408 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0424 15:13:30.016570 133694480537408 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 15:13:30.036143 133694480537408 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 15:13:30.036206 133694480537408 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 15:13:30.225132 133694480537408 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 15:13:30.225246 133694480537408 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 15:13:31.654314 133694480537408 maxtext_utils.py:1740]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0424 15:13:31.654442 133694480537408 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)
I0424 15:13:31.654497 133694480537408 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)
I0424 15:13:31.654554 133694480537408 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')
I0424 15:13:31.654603 133694480537408 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)
I0424 15:13:31.654641 133694480537408 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)
I0424 15:13:31.654692 133694480537408 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)
I0424 15:13:31.654745 133694480537408 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')
I0424 15:13:31.654785 133694480537408 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)
I0424 15:13:31.654824 133694480537408 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)
I0424 15:13:31.654889 133694480537408 maxtext_utils.py:1740]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed', 'vocab')
    Physical:  ('fsdp', None)

I0424 15:13:31.654953 133694480537408 maxtext_utils.py:1740]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed')
    Physical:  (None, 'fsdp')
I0424 15:13:31.678922 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown.
I0424 15:13:31.678992 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,).
I0424 15:13:31.693629 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown.
I0424 15:13:31.693685 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None).
I0424 15:13:31.722425 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown.
I0424 15:13:31.722486 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp').
I0424 15:13:31.737025 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown.
I0424 15:13:31.737084 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None).
I0424 15:13:31.766018 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown.
I0424 15:13:31.766092 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None).
I0424 15:13:31.780634 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown.
I0424 15:13:31.780688 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp').
I0424 15:13:31.823997 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown.
I0424 15:13:31.824066 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None).
I0424 15:13:31.838520 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown.
I0424 15:13:31.838576 133694480537408 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp').
I0424 15:13:32.516755 133694480537408 train.py:158] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 15:13:32.516853 133694480537408 train.py:158] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 15:13:32.532526 133694480537408 train.py:165] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 15:13:32.532584 133694480537408 train.py:165] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 15:13:43.481114 133694480537408 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0424 15:13:43.481917 133694480537408 metric_logger.py:289] number parameters: 1.104 billion
I0424 15:13:45.582777 133694480537408 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 15:13:56.442950 133694480537408 checkpointing.py:798] Waited 10.860151052474976 seconds for step 0 to finish before starting checkpointing.
I0424 15:13:56.445652 133694480537408 checkpoint_manager.py:2013] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 15:13:56.447259 133694480537408 checkpoint_manager.py:1518] [process=4] Saving checkpoint at step 0
I0424 15:13:56.450029 133694480537408 async_checkpointer.py:452] [process=4] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints/0.
I0424 15:13:57.080322 133694480537408 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 15:13:57.081439 133694480537408 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0424 15:13:57.081496 133694480537408 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 15:13:57.325220 133694480537408 base_pytree_checkpoint_handler.py:153] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.244981s
I0424 15:13:57.325390 133694480537408 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.789 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5530204772949219 s) (per-host)
I0424 15:13:57.325441 133694480537408 base_pytree_checkpoint_handler.py:737] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.553081s (batch_requests_ready=0.304341s, total_serialization_initiated=0.248672s, others=0.000068s)
I0424 15:13:57.325530 133694480537408 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.557190s (all_items=0.000020s, per_item={'items': '0.00002027'}, temp_paths=0.557170)
I0424 15:13:57.326498 133575001229056 async_checkpointer.py:79] [process=4][thread=async_save] Background save thread started.
I0424 15:13:57.326672 133694480537408 async_checkpointer.py:561] Finished blocking save. Time taken: 0.879339s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints/0.
I0424 15:13:57.328504 133694480537408 checkpoint_manager.py:1566] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 15:13:57.328784 133571796788992 async_checkpointer.py:265] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0424 15:13:57.328926 133694480537408 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777043636.4456348, 'wait_for_prev_duration_secs': 6.270408630371094e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777043636.4472957, 'checkpointer_blocking_duration_secs': 0.8794817924499512, 'get_old_steps_start_time': 1777043637.3267968, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1777043636.4434164, 'checkpoint_manager_blocking_duration_secs': 0.8854734897613525}
I0424 15:13:57.329085 133694480537408 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 15:13:57.329137 133694480537408 max_utils.py:750] 
Memstats: After params initialized:
I0424 15:13:57.329185 133694480537408 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0))
I0424 15:13:57.329220 133694480537408 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0))
I0424 15:13:57.329247 133694480537408 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0))
I0424 15:13:57.329273 133694480537408 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0))
I0424 15:13:57.590243 133694480537408 metric_logger.py:185] completed step: 0, seconds: 2.101, TFLOP/s/device: 25.871, Tokens/s/device: 3899.562, total_weights: 262144, loss: 10.877
I0424 15:13:58.198168 133694480537408 metric_logger.py:185] completed step: 1, seconds: 12.006, TFLOP/s/device: 4.527, Tokens/s/device: 682.313, total_weights: 262144, loss: 10.877
I0424 15:13:58.776731 133694480537408 metric_logger.py:185] completed step: 2, seconds: 0.030, TFLOP/s/device: 1829.855, Tokens/s/device: 275815.629, total_weights: 262144, loss: 10.563
I0424 15:13:59.355293 133694480537408 metric_logger.py:185] completed step: 3, seconds: 0.584, TFLOP/s/device: 93.061, Tokens/s/device: 14027.157, total_weights: 262144, loss: 10.272
I0424 15:13:59.935061    2338 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 15:14:00.512752 133694480537408 metric_logger.py:185] completed step: 4, seconds: 0.579, TFLOP/s/device: 93.946, Tokens/s/device: 14160.565, total_weights: 262144, loss: 10.022
I0424 15:14:00.518695 133694480537408 metric_logger.py:185] completed step: 5, seconds: 0.579, TFLOP/s/device: 93.932, Tokens/s/device: 14158.460, total_weights: 262144, loss: 9.820
I0424 15:14:01.846140 133575009621760 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_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints/0/items/array_metadatas/process_4
I0424 15:14:34.409517 133575001229056 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/gbytes_per_sec: 41.969 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.63710141181946 s) (per-host)
I0424 15:14:34.409634 133575001229056 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.083024s.
I0424 15:14:38.468698 133694480537408 metric_logger.py:185] completed step: 6, seconds: 1.158, TFLOP/s/device: 46.929, Tokens/s/device: 7073.576, total_weights: 262144, loss: 9.667
I0424 15:14:39.047039 133694480537408 metric_logger.py:185] completed step: 7, seconds: 37.372, TFLOP/s/device: 1.454, Tokens/s/device: 219.201, total_weights: 262144, loss: 9.561
I0424 15:14:39.625254 133694480537408 metric_logger.py:185] completed step: 8, seconds: 0.583, TFLOP/s/device: 93.149, Tokens/s/device: 14040.380, total_weights: 262144, loss: 9.496
I0424 15:14:40.203197 133694480537408 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 15:14:40.203857 133694480537408 checkpointing.py:798] Waited 0.0006816387176513672 seconds for step 9 to finish before starting checkpointing.
I0424 15:14:40.205814 133694480537408 checkpoint_manager.py:2024] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 15:15:17.949624 133575001229056 async_checkpointer.py:144] [process=4][thread=async_save] Background save thread done. Time taken: 80.622998s.
I0424 15:15:17.949927 133571796788992 async_checkpointer.py:273] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 15:15:17.950060 133571796788992 async_checkpointer.py:283] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0424 15:15:17.950109 133571796788992 checkpoint_manager.py:2133] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 15:15:17.951585 133571796788992 checkpoint_manager.py:2142] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 15:15:17.953111 133694480537408 checkpoint_manager.py:2036] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0424 15:15:17.953261 133694480537408 checkpoint_manager.py:1458] Waiting for previous save to complete took 37.747451 seconds. If this number is high, consider checkpointing less frequently.
I0424 15:15:17.955050 133694480537408 checkpoint_manager.py:1518] [process=4] Saving checkpoint at step 9
I0424 15:15:17.958513 133694480537408 async_checkpointer.py:452] [process=4] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints/9.
I0424 15:15:18.498996 133694480537408 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0424 15:15:18.499095 133694480537408 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 15:15:18.540736 133694480537408 base_pytree_checkpoint_handler.py:153] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042689s
I0424 15:15:18.540916 133694480537408 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.085 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.303386926651001 s) (per-host)
I0424 15:15:18.540969 133694480537408 base_pytree_checkpoint_handler.py:737] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.303449s (batch_requests_ready=0.258802s, total_serialization_initiated=0.044576s, others=0.000071s)
I0424 15:15:18.541052 133694480537408 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.307481s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.307466)
I0424 15:15:18.541968 133571226457856 async_checkpointer.py:79] [process=4][thread=async_save] Background save thread started.
I0424 15:15:18.542128 133694480537408 async_checkpointer.py:561] Finished blocking save. Time taken: 0.587007s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints/9.
I0424 15:15:18.546368 133694480537408 checkpoint_manager.py:1566] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 15:15:18.546697 133571796788992 async_checkpointer.py:265] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0424 15:15:18.546856 133694480537408 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777043680.205785, 'wait_for_prev_duration_secs': 37.747451305389404, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777043717.9550896, 'checkpointer_blocking_duration_secs': 0.5871477127075195, 'get_old_steps_start_time': 1777043718.5422556, 'get_old_steps_duration_secs': 2.5033950805664062e-05, 'checkpoint_manager_blocking_start_time': 1777043680.2041054, 'checkpoint_manager_blocking_duration_secs': 38.34271764755249}
I0424 15:15:18.546969 133694480537408 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 15:15:18.547016 133694480537408 checkpoint_manager.py:2024] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 15:15:23.550528 133571788396288 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_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_06_grad_accum/checkpoints/9/items/array_metadatas/process_4
I0424 15:16:01.272320 133571226457856 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/gbytes_per_sec: 36.705 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.03474235534668 s) (per-host)
I0424 15:16:01.272439 133571226457856 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.730359s.
I0424 15:16:08.643947 133571226457856 async_checkpointer.py:144] [process=4][thread=async_save] Background save thread done. Time taken: 50.101852s.
I0424 15:16:08.644238 133571796788992 async_checkpointer.py:273] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 15:16:08.644368 133571796788992 async_checkpointer.py:283] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0424 15:16:08.644419 133571796788992 checkpoint_manager.py:2133] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 15:16:08.645935 133571796788992 checkpoint_manager.py:2142] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 15:16:08.646113 133694480537408 checkpoint_manager.py:2036] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 15:16:08.646254 133694480537408 checkpoint_manager.py:2013] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 15:16:08.646864 133694480537408 metric_logger.py:185] completed step: 9, seconds: 0.578, TFLOP/s/device: 94.001, Tokens/s/device: 14168.819, total_weights: 262144, loss: 9.457
Per train step:
 Total TFLOPs: 54.35 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 24 15:16:20 UTC 2026
EXIT_CODE=0