MaxView

‹ 06_grad_accumCase: 07_eval08_checkpoint_async_true_resume ›

Metrics: Linen vs NNX  ·  feat/nnx-set-defaults-true

MetricLinen  73213e044NNX  73213e044Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s92.08492.507+0.423
Tok/s13879.913943.7+63.882
Avg s/step2.4082.392-0.016
Memory %1.381.380
JAX0.8.10.8.1

Diff = NNX value − Linen value. Green = NNX improved. Red = NNX regressed.

Linen  ·  73213e044  ·  feat_nnx_set_defaults_true_20260423_155242  ·  full log
XPK Start: Thu Apr 23 16:10:51 UTC 2026
2026-04-23 16:10:55.586224: 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 00:00:1776960655.599069      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776960655.603077      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776960655.614756      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776960655.614775      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776960655.614778      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776960655.614779      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-23 16:11:14.894353: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0423 16:11:15.423039 137995844216640 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-23 16:11:24,465:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0423 16:11:24.465535 137995844216640 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-23 16:11:24,467:jax._src.distributed:157: Connecting to JAX distributed service on mt-07-eval-0cwq8-slice-job-0-0.mt-07-eval-0cwq8:8482
I0423 16:11:24.467865 137995844216640 distributed.py:157] Connecting to JAX distributed service on mt-07-eval-0cwq8-slice-job-0-0.mt-07-eval-0cwq8:8482
I0423 16:11:25.858409 137995844216640 max_utils.py:284] Jax distributed system initialized!
I0423 16:11:32.480914 137995844216640 max_utils.py:800] System Information: Jax Version: 0.8.1
I0423 16:11:32.481020 137995844216640 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0423 16:11:32.481060 137995844216640 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709
I0423 16:11:32.481108 137995844216640 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 16:11:33.107884 137995844216640 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 16:11:33.108494 137995844216640 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 16:11:33.108674 137995844216640 checkpointing.py:688] Setting up checkpoint logger...
I0423 16:11:33.108723 137995844216640 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 16:11:33.108766 137995844216640 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0423 16:11:33.109143 137995844216640 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 0x7d811b74f860>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 16:11:36.041274 137995844216640 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0423 16:11:36.041522 137995844216640 checkpoint_manager.py:709] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d6eb62dd4c0>}, handler_registry=None
I0423 16:11:36.041753 137995844216640 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d6eb62dd4c0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0423 16:11:36.041803 137995844216640 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d7fe3f526c0>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0423 16:11:36.041838 137995844216640 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 0x7d6eb62dd4c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d6eb62dd4c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d7fe3f526c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d7fe3f526c0>}).
I0423 16:11:36.042372 137995844216640 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0423 16:11:36.042454 137995844216640 async_checkpointer.py:177] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d6eb58a8180> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0423 16:11:36.784880 137995844216640 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints
I0423 16:11:36.919411 137995844216640 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d7fe43ea6f0>
I0423 16:11:36.919599 137995844216640 checkpointing.py:302] Checkpoint manager created!
I0423 16:11:37.843842 137995844216640 nnx_wrappers.py:455] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 16:11:37.843949 137995844216640 nnx_wrappers.py:455] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 16:11:38.222811 137995844216640 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 16:11:38.222904 137995844216640 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 16:11:38.245227 137995844216640 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 16:11:38.245338 137995844216640 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 16:11:38.268604 137995844216640 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 16:11:38.268674 137995844216640 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 16:11:38.284648 137995844216640 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 16:11:38.284712 137995844216640 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 16:11:38.300910 137995844216640 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 16:11:38.300976 137995844216640 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 16:11:38.326033 137995844216640 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 16:11:38.326117 137995844216640 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 16:11:38.345781 137995844216640 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 16:11:38.345848 137995844216640 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 16:11:38.535067 137995844216640 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 16:11:38.535187 137995844216640 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0423 16:11:39.968872 137995844216640 maxtext_utils.py:1740]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0423 16:11:39.969003 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 16:11:39.969058 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 16:11:39.969141 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   PartitionSpec('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 16:11:39.969199 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0423 16:11:39.969238 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0423 16:11:39.969290 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 16:11:39.969344 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   PartitionSpec('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0423 16:11:39.969385 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0423 16:11:39.969422 137995844216640 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 16:11:39.969469 137995844216640 maxtext_utils.py:1740]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed', 'vocab')
    Physical:  ('fsdp', None)

I0423 16:11:39.969520 137995844216640 maxtext_utils.py:1740]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed')
    Physical:  (None, 'fsdp')
I0423 16:11:40.469596 137995844216640 train.py:158] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 16:11:40.469690 137995844216640 train.py:158] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 16:11:40.485304 137995844216640 train.py:165] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 16:11:40.485369 137995844216640 train.py:165] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 16:11:51.228682 137995844216640 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.
I0423 16:11:51.229481 137995844216640 metric_logger.py:289] number parameters: 1.104 billion
I0423 16:12:02.623587 137995844216640 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0423 16:12:02.778667 137995844216640 checkpointing.py:798] Waited 0.15506553649902344 seconds for step 0 to finish before starting checkpointing.
I0423 16:12:02.781028 137995844216640 checkpoint_manager.py:2013] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 16:12:02.784385 137995844216640 checkpoint_manager.py:1518] [process=6] Saving checkpoint at step 0
I0423 16:12:02.787137 137995844216640 async_checkpointer.py:452] [process=6] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/0.
I0423 16:12:03.666864 137995844216640 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 16:12:03.667922 137995844216640 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0423 16:12:03.667980 137995844216640 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0423 16:12:03.912555 137995844216640 base_pytree_checkpoint_handler.py:153] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.245766s
I0423 16:12:03.912729 137995844216640 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.917 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5288736820220947 s) (per-host)
I0423 16:12:03.912784 137995844216640 base_pytree_checkpoint_handler.py:737] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.528938s (batch_requests_ready=0.279411s, total_serialization_initiated=0.249454s, others=0.000073s)
I0423 16:12:03.912878 137995844216640 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.533067s (all_items=0.000019s, per_item={'items': '0.00001931'}, temp_paths=0.533048)
I0423 16:12:03.913807 137875429521152 async_checkpointer.py:79] [process=6][thread=async_save] Background save thread started.
I0423 16:12:03.913980 137995844216640 async_checkpointer.py:561] Finished blocking save. Time taken: 1.129527s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/0.
I0423 16:12:03.946443 137995844216640 checkpoint_manager.py:1566] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 16:12:03.946721 137871663032064 async_checkpointer.py:265] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 16:12:03.946852 137995844216640 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776960722.781009, 'wait_for_prev_duration_secs': 6.4849853515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776960722.7844243, 'checkpointer_blocking_duration_secs': 1.12965989112854, 'get_old_steps_start_time': 1776960723.9141197, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1776960722.7792535, 'checkpoint_manager_blocking_duration_secs': 1.167557954788208}
I0423 16:12:03.946954 137995844216640 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0423 16:12:03.947005 137995844216640 max_utils.py:750] 
Memstats: After params initialized:
I0423 16:12:03.947061 137995844216640 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0423 16:12:03.947121 137995844216640 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0423 16:12:03.947151 137995844216640 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0423 16:12:03.947176 137995844216640 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0423 16:12:04.209355 137995844216640 metric_logger.py:185] completed step: 0, seconds: 11.394, TFLOP/s/device: 1.192, Tokens/s/device: 179.744, total_weights: 65536, loss: 10.874
I0423 16:12:04.397960 137995844216640 metric_logger.py:185] completed step: 1, seconds: 1.584, TFLOP/s/device: 8.575, Tokens/s/device: 1292.544, total_weights: 65536, loss: 10.874
I0423 16:12:04.838230 137995844216640 metric_logger.py:185] completed step: 2, seconds: 0.041, TFLOP/s/device: 328.731, Tokens/s/device: 49549.985, total_weights: 65536, loss: 10.263
I0423 16:12:04.985618 137995844216640 metric_logger.py:185] completed step: 3, seconds: 0.441, TFLOP/s/device: 30.824, Tokens/s/device: 4646.193, total_weights: 65536, loss: 9.734
I0423 16:12:05.280814 137995844216640 metric_logger.py:185] completed step: 4, seconds: 0.153, TFLOP/s/device: 89.012, Tokens/s/device: 13416.839, total_weights: 65536, loss: 9.278
I0423 16:12:05.286820 137995844216640 metric_logger.py:185] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.081, Tokens/s/device: 13879.476, total_weights: 65536, loss: 8.892
I0423 16:12:24.138144 137995844216640 metric_logger.py:185] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.963, Tokens/s/device: 6928.117, total_weights: 65536, loss: 8.592
I0423 16:12:24.285595 137995844216640 metric_logger.py:185] completed step: 7, seconds: 18.705, TFLOP/s/device: 0.726, Tokens/s/device: 109.491, total_weights: 65536, loss: 8.384
I0423 16:12:24.433084 137995844216640 metric_logger.py:185] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.435, Tokens/s/device: 13480.691, total_weights: 65536, loss: 8.255
I0423 16:12:24.580166 137995844216640 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0423 16:12:24.580810 137995844216640 checkpointing.py:798] Waited 0.0006635189056396484 seconds for step 9 to finish before starting checkpointing.
I0423 16:12:24.582800 137995844216640 checkpoint_manager.py:2024] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 16:12:25.266061    2592 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 16:12:27.222726 137875916035840 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/0/items/array_metadatas/process_6
I0423 16:12:58.766506 137875429521152 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/gbytes_per_sec: 28.522 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 55.382603883743286 s) (per-host)
I0423 16:12:58.766627 137875429521152 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 54.852706s.
I0423 16:13:07.402634 137875429521152 async_checkpointer.py:144] [process=6][thread=async_save] Background save thread done. Time taken: 63.488697s.
I0423 16:13:07.402857 137871663032064 async_checkpointer.py:273] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 16:13:07.402930 137871663032064 async_checkpointer.py:283] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 16:13:07.402977 137871663032064 checkpoint_manager.py:2133] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 16:13:07.404629 137871663032064 checkpoint_manager.py:2142] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 16:13:07.404808 137995844216640 checkpoint_manager.py:2036] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0423 16:13:07.404945 137995844216640 checkpoint_manager.py:1458] Waiting for previous save to complete took 42.822148 seconds. If this number is high, consider checkpointing less frequently.
I0423 16:13:07.406970 137995844216640 checkpoint_manager.py:1518] [process=6] Saving checkpoint at step 9
I0423 16:13:07.410482 137995844216640 async_checkpointer.py:452] [process=6] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/9.
I0423 16:13:07.974755 137995844216640 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0423 16:13:07.974855 137995844216640 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0423 16:13:08.021551 137995844216640 base_pytree_checkpoint_handler.py:153] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.047963s
I0423 16:13:08.021725 137995844216640 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.948 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.31175971031188965 s) (per-host)
I0423 16:13:08.021777 137995844216640 base_pytree_checkpoint_handler.py:737] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.311822s (batch_requests_ready=0.261949s, total_serialization_initiated=0.049803s, others=0.000069s)
I0423 16:13:08.021863 137995844216640 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.315832s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.315816)
I0423 16:13:08.022733 137872700958464 async_checkpointer.py:79] [process=6][thread=async_save] Background save thread started.
I0423 16:13:08.022901 137995844216640 async_checkpointer.py:561] Finished blocking save. Time taken: 0.615862s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/9.
I0423 16:13:08.042939 137995844216640 checkpoint_manager.py:1566] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 16:13:08.043220 137871663032064 async_checkpointer.py:265] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 16:13:08.043363 137995844216640 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776960744.5827727, 'wait_for_prev_duration_secs': 42.8221480846405, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776960787.4070091, 'checkpointer_blocking_duration_secs': 0.6159994602203369, 'get_old_steps_start_time': 1776960788.0230281, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1776960744.5810328, 'checkpoint_manager_blocking_duration_secs': 43.46229362487793}
I0423 16:13:08.043470 137995844216640 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0423 16:13:08.043517 137995844216640 checkpoint_manager.py:2024] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 16:13:12.864974 137875916035840 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/linen_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/9/items/array_metadatas/process_6
I0423 16:13:49.223528 137872700958464 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/gbytes_per_sec: 38.050 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.51352000236511 s) (per-host)
I0423 16:13:49.223675 137872700958464 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.200828s.
I0423 16:13:57.250714 137872700958464 async_checkpointer.py:144] [process=6][thread=async_save] Background save thread done. Time taken: 49.227853s.
I0423 16:13:57.251011 137871663032064 async_checkpointer.py:273] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 16:13:57.251146 137871663032064 async_checkpointer.py:283] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 16:13:57.251192 137871663032064 checkpoint_manager.py:2133] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 16:13:57.253676 137871663032064 checkpoint_manager.py:2142] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 16:13:57.253864 137995844216640 checkpoint_manager.py:2036] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0423 16:13:57.254007 137995844216640 checkpoint_manager.py:2013] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 16:13:57.254753 137995844216640 metric_logger.py:185] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.084, Tokens/s/device: 13879.853, total_weights: 65536, loss: 8.179
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Thu Apr 23 16:14:07 UTC 2026
EXIT_CODE=0
NNX  ·  73213e044  ·  feat_nnx_set_defaults_true_20260423_155242  ·  full log
XPK Start: Thu Apr 23 17:46:15 UTC 2026
2026-04-23 17:46:19.201429: 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 00:00:1776966379.213805      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776966379.217697      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776966379.229452      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776966379.229470      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776966379.229474      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776966379.229476      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-23 17:46:38.431901: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0423 17:46:38.953424 138377733605184 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-23 17:46:47,994:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0423 17:46:47.994849 138377733605184 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-23 17:46:47,997:jax._src.distributed:157: Connecting to JAX distributed service on mt-07-eval-dis2p-slice-job-0-0.mt-07-eval-dis2p:8482
I0423 17:46:47.997185 138377733605184 distributed.py:157] Connecting to JAX distributed service on mt-07-eval-dis2p-slice-job-0-0.mt-07-eval-dis2p:8482
I0423 17:46:49.291754 138377733605184 max_utils.py:284] Jax distributed system initialized!
I0423 17:46:55.129777 138377733605184 max_utils.py:800] System Information: Jax Version: 0.8.1
I0423 17:46:55.129882 138377733605184 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0423 17:46:55.129929 138377733605184 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709
I0423 17:46:55.129974 138377733605184 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 17:46:55.751009 138377733605184 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 17:46:55.865396 138377733605184 checkpointing.py:688] Setting up checkpoint logger...
I0423 17:46:55.865517 138377733605184 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 17:46:55.865567 138377733605184 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0423 17:46:55.865817 138377733605184 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 0x7dda243ae9f0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 17:46:58.778538 138377733605184 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0423 17:46:58.778780 138377733605184 checkpoint_manager.py:709] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7dc71b8b2240>}, handler_registry=None
I0423 17:46:58.779016 138377733605184 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7dc71b8b2240>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0423 17:46:58.779065 138377733605184 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dc71b8b0170>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0423 17:46:58.779115 138377733605184 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 0x7dc71b8b2240>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7dc71b8b2240>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dc71b8b0170>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dc71b8b0170>}).
I0423 17:46:58.779454 138377733605184 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0423 17:46:58.779531 138377733605184 async_checkpointer.py:177] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7dc71b410f40> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0423 17:46:59.515782 138377733605184 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints
I0423 17:46:59.532068 138377733605184 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7dc71b3a1a00>
I0423 17:46:59.532213 138377733605184 checkpointing.py:302] Checkpoint manager created!
I0423 17:47:00.484956 138377733605184 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 17:47:00.485082 138377733605184 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0423 17:47:01.961406 138377733605184 maxtext_utils.py:1749]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0423 17:47:01.961528 138377733605184 maxtext_utils.py:1749]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0423 17:47:01.961592 138377733605184 maxtext_utils.py:1749]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0423 17:47:01.961638 138377733605184 maxtext_utils.py:1749]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0423 17:47:01.961675 138377733605184 maxtext_utils.py:1749]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0423 17:47:01.961709 138377733605184 maxtext_utils.py:1749]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0423 17:47:01.961746 138377733605184 maxtext_utils.py:1749]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0423 17:47:01.961782 138377733605184 maxtext_utils.py:1749]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0423 17:47:01.961819 138377733605184 maxtext_utils.py:1749]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0423 17:47:01.961854 138377733605184 maxtext_utils.py:1749]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0423 17:47:01.961891 138377733605184 maxtext_utils.py:1749]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0423 17:47:01.961925 138377733605184 maxtext_utils.py:1749]  token_embedder/embedding/value

    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')
I0423 17:47:02.354916 138377733605184 nnx_decoders.py:616] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 17:47:02.355071 138377733605184 nnx_decoders.py:616] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 17:47:02.360446 138377733605184 nnx_decoders.py:616] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 17:47:02.360542 138377733605184 nnx_decoders.py:616] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 17:47:02.520915 138377733605184 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 17:47:02.521052 138377733605184 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 17:47:02.679245 138377733605184 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 17:47:02.679387 138377733605184 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 17:47:02.846929 138377733605184 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 17:47:02.847072 138377733605184 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 17:47:03.006274 138377733605184 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 17:47:03.006436 138377733605184 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 17:47:03.165891 138377733605184 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 17:47:03.166024 138377733605184 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 17:47:03.336242 138377733605184 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 17:47:03.336392 138377733605184 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 17:47:03.358921 138377733605184 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 17:47:03.359020 138377733605184 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 17:47:14.217751 138377733605184 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.
I0423 17:47:14.220176 138377733605184 metric_logger.py:289] number parameters: 1.104 billion
I0423 17:47:25.687629 138377733605184 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0423 17:47:25.841023 138377733605184 checkpointing.py:798] Waited 0.1533811092376709 seconds for step 0 to finish before starting checkpointing.
I0423 17:47:25.843225 138377733605184 checkpoint_manager.py:2013] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 17:47:25.845033 138377733605184 checkpoint_manager.py:1518] [process=6] Saving checkpoint at step 0
I0423 17:47:25.847844 138377733605184 async_checkpointer.py:452] [process=6] Started async saving checkpoint to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/0.
I0423 17:47:26.665574 138377733605184 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 17:47:26.666550 138377733605184 jax_array_handlers.py:358] Scheduling D2H of 69 prioritized jax.Array.
I0423 17:47:26.666616 138377733605184 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0423 17:47:26.918534 138377733605184 base_pytree_checkpoint_handler.py:153] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.254092s
I0423 17:47:26.918731 138377733605184 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.975 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5184614658355713 s) (per-host)
I0423 17:47:26.918795 138377733605184 base_pytree_checkpoint_handler.py:737] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.518534s (batch_requests_ready=0.257836s, total_serialization_initiated=0.260614s, others=0.000084s)
I0423 17:47:26.918915 138377733605184 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.522659s (all_items=0.000022s, per_item={'items': '0.00002170'}, temp_paths=0.522637)
I0423 17:47:26.919921 138258612741888 async_checkpointer.py:79] [process=6][thread=async_save] Background save thread started.
I0423 17:47:26.920106 138377733605184 async_checkpointer.py:561] Finished blocking save. Time taken: 1.074991s. Continuing background save to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/0.
I0423 17:47:26.933927 138377733605184 checkpoint_manager.py:1566] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 17:47:26.934259 138254292596480 async_checkpointer.py:265] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 17:47:26.934421 138377733605184 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776966445.8432074, 'wait_for_prev_duration_secs': 6.580352783203125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776966445.8450713, 'checkpointer_blocking_duration_secs': 1.0751426219940186, 'get_old_steps_start_time': 1776966446.920233, 'get_old_steps_duration_secs': 2.4557113647460938e-05, 'checkpoint_manager_blocking_start_time': 1776966445.8414571, 'checkpoint_manager_blocking_duration_secs': 1.0929276943206787}
I0423 17:47:26.934530 138377733605184 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0423 17:47:26.934602 138377733605184 max_utils.py:750] 
Memstats: After params initialized:
I0423 17:47:26.934652 138377733605184 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0423 17:47:26.934685 138377733605184 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0423 17:47:26.934713 138377733605184 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0423 17:47:26.934737 138377733605184 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0423 17:47:27.205747 138377733605184 metric_logger.py:185] completed step: 0, seconds: 11.466, TFLOP/s/device: 1.185, Tokens/s/device: 178.609, total_weights: 65536, loss: 10.874
I0423 17:47:27.374483 138377733605184 metric_logger.py:185] completed step: 1, seconds: 1.516, TFLOP/s/device: 8.960, Tokens/s/device: 1350.597, total_weights: 65536, loss: 10.874
I0423 17:47:27.802597 138377733605184 metric_logger.py:185] completed step: 2, seconds: 0.023, TFLOP/s/device: 589.694, Tokens/s/device: 88885.031, total_weights: 65536, loss: 10.262
I0423 17:47:27.949958 138377733605184 metric_logger.py:185] completed step: 3, seconds: 0.419, TFLOP/s/device: 32.402, Tokens/s/device: 4883.970, total_weights: 65536, loss: 9.732
I0423 17:47:28.245687 138377733605184 metric_logger.py:185] completed step: 4, seconds: 0.163, TFLOP/s/device: 83.562, Tokens/s/device: 12595.403, total_weights: 65536, loss: 9.274
I0423 17:47:28.253119 138377733605184 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.557, Tokens/s/device: 13951.144, total_weights: 65536, loss: 8.888
I0423 17:47:47.067291 138377733605184 metric_logger.py:185] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.845, Tokens/s/device: 6910.304, total_weights: 65536, loss: 8.589
I0423 17:47:47.214620 138377733605184 metric_logger.py:185] completed step: 7, seconds: 18.668, TFLOP/s/device: 0.728, Tokens/s/device: 109.705, total_weights: 65536, loss: 8.380
I0423 17:47:47.362119 138377733605184 metric_logger.py:185] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.662, Tokens/s/device: 13364.046, total_weights: 65536, loss: 8.251
I0423 17:47:47.510571 138377733605184 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0423 17:47:47.511195 138377733605184 checkpointing.py:798] Waited 0.0006356239318847656 seconds for step 9 to finish before starting checkpointing.
I0423 17:47:47.513118 138377733605184 checkpoint_manager.py:2024] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 17:47:48.182129    2562 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 17:47:50.123223 138258621134592 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/0/items/array_metadatas/process_6
I0423 17:48:25.351716 138258612741888 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/gbytes_per_sec: 26.795 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 58.95141243934631 s) (per-host)
I0423 17:48:25.351824 138258612741888 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 58.431791s.
I0423 17:48:30.646273 138258612741888 async_checkpointer.py:144] [process=6][thread=async_save] Background save thread done. Time taken: 63.726222s.
I0423 17:48:30.646581 138254292596480 async_checkpointer.py:273] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 17:48:30.646696 138254292596480 async_checkpointer.py:283] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 17:48:30.646761 138254292596480 checkpoint_manager.py:2133] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 17:48:30.648259 138254292596480 checkpoint_manager.py:2142] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 17:48:30.648389 138377733605184 checkpoint_manager.py:2036] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0423 17:48:30.648474 138377733605184 checkpoint_manager.py:1458] Waiting for previous save to complete took 43.135373 seconds. If this number is high, consider checkpointing less frequently.
I0423 17:48:30.650703 138377733605184 checkpoint_manager.py:1518] [process=6] Saving checkpoint at step 9
I0423 17:48:30.654196 138377733605184 async_checkpointer.py:452] [process=6] Started async saving checkpoint to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/9.
I0423 17:48:31.213840 138377733605184 jax_array_handlers.py:358] Scheduling D2H of 69 prioritized jax.Array.
I0423 17:48:31.214002 138377733605184 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0423 17:48:31.263141 138377733605184 base_pytree_checkpoint_handler.py:153] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.051299s
I0423 17:48:31.263314 138377733605184 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.872 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.3166055679321289 s) (per-host)
I0423 17:48:31.263363 138377733605184 base_pytree_checkpoint_handler.py:737] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.316664s (batch_requests_ready=0.260597s, total_serialization_initiated=0.056000s, others=0.000067s)
I0423 17:48:31.263455 138377733605184 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.320814s (all_items=0.000015s, per_item={'items': '0.00001550'}, temp_paths=0.320799)
I0423 17:48:31.264284 138254821881600 async_checkpointer.py:79] [process=6][thread=async_save] Background save thread started.
I0423 17:48:31.264444 138377733605184 async_checkpointer.py:561] Finished blocking save. Time taken: 0.613672s. Continuing background save to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/9.
I0423 17:48:31.266220 138377733605184 checkpoint_manager.py:1566] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 17:48:31.266426 138254292596480 async_checkpointer.py:265] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 17:48:31.266520 138377733605184 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776966467.513076, 'wait_for_prev_duration_secs': 43.13537263870239, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776966510.6507447, 'checkpointer_blocking_duration_secs': 0.613806962966919, 'get_old_steps_start_time': 1776966511.2645712, 'get_old_steps_duration_secs': 2.5033950805664062e-05, 'checkpoint_manager_blocking_start_time': 1776966467.511434, 'checkpoint_manager_blocking_duration_secs': 43.755056858062744}
I0423 17:48:31.266680 138377733605184 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0423 17:48:31.266725 138377733605184 checkpoint_manager.py:2024] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 17:48:36.884409 138258621134592 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260423_155242/nnx_xpk_feat_nnx_set_defaults_true_20260423_155242_07_eval/checkpoints/9/items/array_metadatas/process_6
I0423 17:49:13.226134 138254821881600 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/gbytes_per_sec: 37.361 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.27939295768738 s) (per-host)
I0423 17:49:13.226237 138254821881600 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.961836s.
I0423 17:49:22.145399 138254821881600 async_checkpointer.py:144] [process=6][thread=async_save] Background save thread done. Time taken: 50.880982s.
I0423 17:49:22.145778 138254292596480 async_checkpointer.py:273] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 17:49:22.145839 138254292596480 async_checkpointer.py:283] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 17:49:22.145903 138254292596480 checkpoint_manager.py:2133] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 17:49:22.147367 138254292596480 checkpoint_manager.py:2142] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 17:49:22.147523 138377733605184 checkpoint_manager.py:2036] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0423 17:49:22.147707 138377733605184 checkpoint_manager.py:2013] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 17:49:22.148374 138377733605184 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.507, Tokens/s/device: 13943.735, total_weights: 65536, loss: 8.175
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Thu Apr 23 17:49:35 UTC 2026
EXIT_CODE=0