MaxView

Timing Test Reference Updates
‹ 10_shardy_falseCase: 10_shardy_true11_optimizer_offload_false ›

Metrics: main (e27fc1e97) vs feat/nnx-trainstate-and-training-loop (e27fc1e97)

Metricmain  e27fc1e97feat/nnx-trainstate-and-training-loop  e27fc1e97Diff (feat/nnx-trainstate-and-training-loop − main)
Parameters1.104 billion1.104 billion—
Final loss8.17908.17900
TFLOP/s92.21592.307+0.092
Tok/s13899.613913.5+13.881
Avg s/step2.8212.644-0.177
Memory %1.381.380
JAX0.9.20.9.2—

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

main  ·  e27fc1e97  ·  main_20260421_144122  ·  full log 
XPK Start: Tue Apr 21 15:14:59 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. 
2026-04-21 15:15:24.109045: 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)
I0421 15:15:24.319963 135414068074304 max_utils.py:273] Attempting to initialize the jax distributed system...
I0421 15:15:33.359740 135414068074304 distributed.py:149] Starting JAX distributed service on [::]:8482
I0421 15:15:33.361909 135414068074304 distributed.py:172] Connecting to JAX distributed service on mt-10-shardy-true-j1nvr-slice-job-0-0.mt-10-shardy-true-j1nvr:8482
I0421 15:15:34.466329 135414068074304 max_utils.py:284] Jax distributed system initialized!
I0421 15:15:40.464592 135414068074304 max_utils.py:800] System Information: Jax Version: 0.9.2
I0421 15:15:40.464712 135414068074304 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0421 15:15:40.464757 135414068074304 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
I0421 15:15:40.464795 135414068074304 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0421 15:15:41.166008 135414068074304 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 15:15:41.166290 135414068074304 checkpointing.py:677] Setting up checkpoint logger...
I0421 15:15:41.166342 135414068074304 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0421 15:15:41.166386 135414068074304 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0421 15:15:41.166742 135414068074304 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 0x7b27f9bb1520>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0421 15:15:44.198276 135414068074304 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0421 15:15:44.198523 135414068074304 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b12c0685d30>}, handler_registry=None
I0421 15:15:44.198776 135414068074304 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b12c0685d30>` 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`.
I0421 15:15:44.198824 135414068074304 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b12c0685fa0>` 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`.
I0421 15:15:44.198859 135414068074304 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 0x7b12c0685d30>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b12c0685d30>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b12c0685fa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b12c0685fa0>}).
I0421 15:15:44.199175 135414068074304 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0421 15:15:44.199244 135414068074304 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b12a0205580> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0421 15:15:45.280979 135414068074304 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints
I0421 15:15:45.705391 135414068074304 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b26c71486b0>
I0421 15:15:45.705572 135414068074304 checkpointing.py:301] Checkpoint manager created!
I0421 15:15:46.637388 135414068074304 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0421 15:15:46.637511 135414068074304 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 15:15:47.020640 135414068074304 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 15:15:47.020747 135414068074304 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 15:15:47.037297 135414068074304 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 15:15:47.037358 135414068074304 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 15:15:47.061428 135414068074304 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 15:15:47.061500 135414068074304 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 15:15:47.078124 135414068074304 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 15:15:47.078193 135414068074304 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 15:15:47.094721 135414068074304 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 15:15:47.094783 135414068074304 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 15:15:47.119684 135414068074304 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0421 15:15:47.119766 135414068074304 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 15:15:47.141033 135414068074304 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0421 15:15:47.141100 135414068074304 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0421 15:15:47.359120 135414068074304 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0421 15:15:47.359229 135414068074304 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0421 15:15:48.799915 135414068074304 maxtext_utils.py:1668]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0421 15:15:48.800041 135414068074304 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 15:15:48.800096 135414068074304 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 15:15:48.800154 135414068074304 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0421 15:15:48.800206 135414068074304 maxtext_utils.py:1668]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0421 15:15:48.800246 135414068074304 maxtext_utils.py:1668]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0421 15:15:48.800297 135414068074304 maxtext_utils.py:1668]  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)
I0421 15:15:48.800349 135414068074304 maxtext_utils.py:1668]  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')
I0421 15:15:48.800390 135414068074304 maxtext_utils.py:1668]  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)
I0421 15:15:48.800427 135414068074304 maxtext_utils.py:1668]  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)
I0421 15:15:48.800481 135414068074304 maxtext_utils.py:1668]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0421 15:15:48.800531 135414068074304 maxtext_utils.py:1668]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0421 15:15:49.293173 135414068074304 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0421 15:15:49.293269 135414068074304 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0421 15:15:49.308933 135414068074304 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0421 15:15:49.308991 135414068074304 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0421 15:16:00.177336 135414068074304 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0421 15:16:00.178191 135414068074304 metric_logger.py:301] number parameters: 1.104 billion
I0421 15:16:11.747688 135414068074304 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0421 15:16:12.420924 135414068074304 checkpointing.py:776] Waited 0.6732115745544434 seconds for step 0 to finish before starting checkpointing.
I0421 15:16:12.423309 135414068074304 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 15:16:12.424983 135414068074304 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0
I0421 15:16:12.426536 135414068074304 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints/0.
I0421 15:16:13.182392 135414068074304 signaling_client.py:364] Using JaxDistributedSignalingClient
I0421 15:16:13.183336 135414068074304 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 15:16:13.183393 135414068074304 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
I0421 15:16:13.462216 135414068074304 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279897s
I0421 15:16:13.462395 135414068074304 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.406 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2853226661682129 s) (per-host)
I0421 15:16:13.462456 135414068074304 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.285393s (batch_requests_ready=0.002202s, total_serialization_initiated=0.283113s, others=0.000078s)
I0421 15:16:13.462550 135414068074304 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.289804s (all_items=0.000020s, per_item={'items': '0.00002003'}, temp_paths=0.289784)
I0421 15:16:13.463392 135414068074304 event_tracking.py:125] [process=4] [async] Finished blocking save in 1.04 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints/0.
I0421 15:16:13.463733 135285858518784 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 15:36:13.463688
I0421 15:16:13.468339 135414068074304 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 15:16:13.468617 135285181425408 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0421 15:16:13.468759 135414068074304 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776784572.42329, 'wait_for_prev_duration_secs': 6.031990051269531e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776784572.4250233, 'checkpointer_blocking_duration_secs': 1.0388622283935547, 'get_old_steps_start_time': 1776784573.4639115, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776784572.421439, 'checkpoint_manager_blocking_duration_secs': 1.0472822189331055}
I0421 15:16:13.468913 135414068074304 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0421 15:16:13.468962 135414068074304 max_utils.py:750] 
Memstats: After params initialized:
I0421 15:16:13.469013 135414068074304 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0))
I0421 15:16:13.469048 135414068074304 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0))
I0421 15:16:13.469074 135414068074304 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0))
I0421 15:16:13.469100 135414068074304 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0))
I0421 15:16:13.784295 135414068074304 metric_logger.py:196] completed step: 0, seconds: 11.569, TFLOP/s/device: 1.174, Tokens/s/device: 177.019, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0421 15:16:13.962409 135414068074304 metric_logger.py:196] completed step: 1, seconds: 2.035, TFLOP/s/device: 6.676, Tokens/s/device: 1006.326, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0421 15:16:14.367958 135414068074304 metric_logger.py:196] completed step: 2, seconds: 0.029, TFLOP/s/device: 461.316, Tokens/s/device: 69534.513, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0421 15:16:14.515337 135414068074304 metric_logger.py:196] completed step: 3, seconds: 0.407, TFLOP/s/device: 33.383, Tokens/s/device: 5031.867, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0421 15:16:14.810586 135414068074304 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.602, Tokens/s/device: 13354.983, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0421 15:16:14.816484 135414068074304 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.338, Tokens/s/device: 13918.244, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0421 15:16:16.796802    2798 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0421 15:16:19.006842 135285189818112 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints/0/items/array_metadatas/process_4
I0421 15:16:36.986803 135414068074304 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.942, Tokens/s/device: 6924.861, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0421 15:16:37.134178 135414068074304 metric_logger.py:196] completed step: 7, seconds: 22.023, TFLOP/s/device: 0.617, Tokens/s/device: 92.992, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0421 15:16:37.281547 135414068074304 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.279, Tokens/s/device: 13457.040, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0421 15:16:37.428175 135414068074304 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0421 15:16:37.428861 135414068074304 checkpointing.py:776] Waited 0.0007059574127197266 seconds for step 9 to finish before starting checkpointing.
I0421 15:16:37.430902 135414068074304 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 15:16:49.697159 135285858518784 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 43.253 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.52005410194397 s) (per-host)
I0421 15:16:49.697273 135285858518784 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.233426s.
I0421 15:16:57.526410 135285858518784 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 44.062548s.
I0421 15:16:57.526721 135285181425408 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 15:16:57.526850 135285181425408 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0421 15:16:57.526903 135285181425408 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0421 15:16:57.529105 135285181425408 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0421 15:16:57.529285 135414068074304 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0421 15:16:57.529412 135414068074304 checkpoint_manager.py:1452] Waiting for previous save to complete took 20.098511 seconds. If this number is high, consider checkpointing less frequently.
I0421 15:16:57.531214 135414068074304 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9
I0421 15:16:57.533527 135414068074304 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints/9.
I0421 15:16:57.828074 135414068074304 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 15:16:57.828165 135414068074304 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
I0421 15:16:57.863198 135414068074304 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036226s
I0421 15:16:57.863341 135414068074304 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 38.316 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.040259599685668945 s) (per-host)
I0421 15:16:57.863394 135414068074304 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040320s (batch_requests_ready=0.002415s, total_serialization_initiated=0.037837s, others=0.000068s)
I0421 15:16:57.863481 135414068074304 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044461s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.044444)
I0421 15:16:57.864125 135414068074304 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints/9.
I0421 15:16:57.864463 135285181425408 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 15:36:57.864426
I0421 15:16:57.866338 135414068074304 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 15:16:57.866611 135285189818112 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0421 15:16:57.866756 135414068074304 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776784597.4308727, 'wait_for_prev_duration_secs': 20.09851098060608, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776784617.5312521, 'checkpointer_blocking_duration_secs': 0.3333573341369629, 'get_old_steps_start_time': 1776784617.8646314, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776784597.429095, 'checkpoint_manager_blocking_duration_secs': 20.43762731552124}
I0421 15:16:57.866908 135414068074304 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0421 15:16:57.866952 135414068074304 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 15:17:03.009168 135260373239552 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_144122/linen_xpk_main_20260421_144122_10_shardy_true/checkpoints/9/items/array_metadatas/process_4
I0421 15:17:39.873821 135285181425408 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.564 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.0507025718689 s) (per-host)
I0421 15:17:39.873933 135285181425408 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.009354s.
I0421 15:17:46.670607 135285181425408 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 48.806015s.
I0421 15:17:46.670900 135285189818112 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 15:17:46.671017 135285189818112 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0421 15:17:46.671065 135285189818112 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0421 15:17:46.679972 135285189818112 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0421 15:17:46.680176 135414068074304 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0421 15:17:46.680330 135414068074304 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 15:17:46.681320 135414068074304 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.215, Tokens/s/device: 13899.635, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Tue Apr 21 15:17:59 UTC 2026
EXIT_CODE=0
feat/nnx-trainstate-and-training-loop  ·  e27fc1e97  ·  feat_nnx_trainstate_and_training_loop_20260421_161900  ·  full log 
XPK Start: Tue Apr 21 17:28:45 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-21 17:29:09.463581: 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)
I0421 17:29:09.642325 138718042560320 max_utils.py:273] Attempting to initialize the jax distributed system...
I0421 17:29:18.682833 138718042560320 distributed.py:149] Starting JAX distributed service on [::]:8482
I0421 17:29:18.685259 138718042560320 distributed.py:172] Connecting to JAX distributed service on mt-10-shardy-true-7u3ng-slice-job-0-0.mt-10-shardy-true-7u3ng:8482
I0421 17:29:19.268162 138718042560320 max_utils.py:284] Jax distributed system initialized!
I0421 17:29:25.332860 138718042560320 max_utils.py:800] System Information: Jax Version: 0.9.2
I0421 17:29:25.332975 138718042560320 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0421 17:29:25.333016 138718042560320 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
I0421 17:29:25.333051 138718042560320 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0421 17:29:26.051792 138718042560320 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 17:29:26.052388 138718042560320 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0421 17:29:26.052570 138718042560320 checkpointing.py:688] Setting up checkpoint logger...
I0421 17:29:26.052618 138718042560320 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0421 17:29:26.052660 138718042560320 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0421 17:29:26.053015 138718042560320 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 0x7e2917e50260>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0421 17:29:29.342660 138718042560320 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0421 17:29:29.342849 138718042560320 checkpoint_manager.py:708] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e14f8212780>}, handler_registry=None
I0421 17:29:29.343096 138718042560320 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e14f8212780>` 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`.
I0421 17:29:29.343146 138718042560320 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e14f80b75c0>` 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`.
I0421 17:29:29.343182 138718042560320 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 0x7e14f8212780>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e14f8212780>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e14f80b75c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e14f80b75c0>}).
I0421 17:29:29.343501 138718042560320 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0421 17:29:29.343572 138718042560320 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e14b857c2c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0421 17:29:30.613445 138718042560320 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints
I0421 17:29:30.902566 138718042560320 checkpoint_manager.py:929] [process=3][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e14f80b6de0>
I0421 17:29:30.902757 138718042560320 checkpointing.py:302] Checkpoint manager created!
I0421 17:29:31.840537 138718042560320 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0421 17:29:31.840646 138718042560320 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 17:29:32.222719 138718042560320 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 17:29:32.222816 138718042560320 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 17:29:32.239173 138718042560320 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0421 17:29:32.239233 138718042560320 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0421 17:29:32.263034 138718042560320 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 17:29:32.263101 138718042560320 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 17:29:32.279545 138718042560320 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 17:29:32.279604 138718042560320 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 17:29:32.296008 138718042560320 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0421 17:29:32.296066 138718042560320 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 17:29:32.320773 138718042560320 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0421 17:29:32.320839 138718042560320 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0421 17:29:32.341985 138718042560320 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0421 17:29:32.342047 138718042560320 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0421 17:29:32.560333 138718042560320 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0421 17:29:32.560436 138718042560320 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0421 17:29:33.983901 138718042560320 maxtext_utils.py:1821]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0421 17:29:33.984044 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 17:29:33.984372 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0421 17:29:33.984500 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0421 17:29:33.984574 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0421 17:29:33.984618 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0421 17:29:33.984682 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 17:29:33.984741 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0421 17:29:33.984787 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0421 17:29:33.984826 138718042560320 maxtext_utils.py:1821]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0421 17:29:33.984879 138718042560320 maxtext_utils.py:1821]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0421 17:29:33.984953 138718042560320 maxtext_utils.py:1821]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0421 17:29:34.488640 138718042560320 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0421 17:29:34.488739 138718042560320 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0421 17:29:34.504283 138718042560320 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0421 17:29:34.504346 138718042560320 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0421 17:29:45.404716 138718042560320 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0421 17:29:45.405606 138718042560320 metric_logger.py:301] number parameters: 1.104 billion
I0421 17:29:56.870869 138718042560320 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0421 17:29:57.026061 138718042560320 checkpointing.py:798] Waited 0.15516996383666992 seconds for step 0 to finish before starting checkpointing.
I0421 17:29:57.028534 138718042560320 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 17:29:57.030333 138718042560320 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0
I0421 17:29:57.032075 138718042560320 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints/0.
I0421 17:29:57.358899 138718042560320 signaling_client.py:364] Using JaxDistributedSignalingClient
I0421 17:29:57.359879 138718042560320 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 17:29:57.359955 138718042560320 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
I0421 17:29:57.641590 138718042560320 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.282763s
I0421 17:29:57.641770 138718042560320 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 5.351 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2882697582244873 s) (per-host)
I0421 17:29:57.641826 138718042560320 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.288336s (batch_requests_ready=0.002296s, total_serialization_initiated=0.285964s, others=0.000076s)
I0421 17:29:57.641921 138718042560320 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.292487s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.292470)
I0421 17:29:57.642782 138718042560320 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints/0.
I0421 17:29:57.643094 138589678253824 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 17:49:57.643061
I0421 17:29:57.645122 138718042560320 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 17:29:57.645405 138588624770816 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0421 17:29:57.645538 138718042560320 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776792597.028515, 'wait_for_prev_duration_secs': 6.198883056640625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776792597.0304177, 'checkpointer_blocking_duration_secs': 0.612830638885498, 'get_old_steps_start_time': 1776792597.6432781, 'get_old_steps_duration_secs': 3.337860107421875e-05, 'checkpoint_manager_blocking_start_time': 1776792597.0264502, 'checkpoint_manager_blocking_duration_secs': 0.6190471649169922}
I0421 17:29:57.645710 138718042560320 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0421 17:29:57.645774 138718042560320 max_utils.py:750] 
Memstats: After params initialized:
I0421 17:29:57.645831 138718042560320 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0))
I0421 17:29:57.645879 138718042560320 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0))
I0421 17:29:57.645919 138718042560320 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0))
I0421 17:29:57.645964 138718042560320 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0))
I0421 17:29:57.954435 138718042560320 metric_logger.py:196] completed step: 0, seconds: 11.465, TFLOP/s/device: 1.185, Tokens/s/device: 178.628, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0421 17:29:58.129860 138718042560320 metric_logger.py:196] completed step: 1, seconds: 1.082, TFLOP/s/device: 12.557, Tokens/s/device: 1892.749, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0421 17:29:58.550608 138718042560320 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 484.044, Tokens/s/device: 72960.456, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0421 17:29:58.697969 138718042560320 metric_logger.py:196] completed step: 3, seconds: 0.421, TFLOP/s/device: 32.279, Tokens/s/device: 4865.371, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0421 17:29:58.993102 138718042560320 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.611, Tokens/s/device: 13356.464, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0421 17:29:58.998874 138718042560320 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.335, Tokens/s/device: 13917.676, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0421 17:30:00.591645    2787 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0421 17:30:02.646905 138589147494144 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints/0/items/array_metadatas/process_3
I0421 17:30:20.514900 138718042560320 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.978, Tokens/s/device: 6930.298, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0421 17:30:20.662004 138718042560320 metric_logger.py:196] completed step: 7, seconds: 21.369, TFLOP/s/device: 0.636, Tokens/s/device: 95.839, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0421 17:30:20.809265 138718042560320 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.194, Tokens/s/device: 13444.319, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0421 17:30:20.955849 138718042560320 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0421 17:30:20.956532 138718042560320 checkpointing.py:798] Waited 0.0006947517395019531 seconds for step 9 to finish before starting checkpointing.
I0421 17:30:20.958552 138718042560320 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 17:30:33.825026 138589678253824 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 43.310 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.4714879989624 s) (per-host)
I0421 17:30:33.825152 138589678253824 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.181945s.
I0421 17:30:42.883198 138589678253824 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 45.239974s.
I0421 17:30:42.883494 138588624770816 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 17:30:42.883630 138588624770816 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0421 17:30:42.883682 138588624770816 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0421 17:30:42.885772 138588624770816 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0421 17:30:42.885972 138718042560320 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0421 17:30:42.886108 138718042560320 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.927556 seconds. If this number is high, consider checkpointing less frequently.
I0421 17:30:42.887981 138718042560320 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 9
I0421 17:30:42.890121 138718042560320 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints/9.
I0421 17:30:43.184727 138718042560320 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0421 17:30:43.184822 138718042560320 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
I0421 17:30:43.224730 138718042560320 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041085s
I0421 17:30:43.224907 138718042560320 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 34.562 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.044632673263549805 s) (per-host)
I0421 17:30:43.224983 138718042560320 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044719s (batch_requests_ready=0.001897s, total_serialization_initiated=0.042727s, others=0.000095s)
I0421 17:30:43.225079 138718042560320 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048793s (all_items=0.000016s, per_item={'items': '0.00001645'}, temp_paths=0.048777)
I0421 17:30:43.225784 138718042560320 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints/9.
I0421 17:30:43.226076 138588624770816 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 17:50:43.226052
I0421 17:30:43.228041 138718042560320 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0421 17:30:43.228307 138567141529344 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0421 17:30:43.228440 138718042560320 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776792620.9585223, 'wait_for_prev_duration_secs': 21.927555561065674, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776792642.8880222, 'checkpointer_blocking_duration_secs': 0.338153600692749, 'get_old_steps_start_time': 1776792643.2262006, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776792620.9567618, 'checkpoint_manager_blocking_duration_secs': 22.271645545959473}
I0421 17:30:43.228609 138718042560320 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0421 17:30:43.228654 138718042560320 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0421 17:30:48.760573 138589147494144 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_161900_10_shardy_true/checkpoints/9/items/array_metadatas/process_3
I0421 17:31:25.261991 138588624770816 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.536 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.08165955543518 s) (per-host)
I0421 17:31:25.262121 138588624770816 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.035969s.
I0421 17:31:34.653883 138588624770816 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 51.427715s.
I0421 17:31:34.654193 138567141529344 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0421 17:31:34.654309 138567141529344 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0421 17:31:34.654380 138567141529344 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0421 17:31:34.655520 138567141529344 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0421 17:31:34.655736 138718042560320 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0421 17:31:34.655911 138718042560320 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0421 17:31:34.657120 138718042560320 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.307, Tokens/s/device: 13913.516, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Tue Apr 21 17:31:46 UTC 2026
EXIT_CODE=0