MaxView

‹ 01_baseCase: 02_synthetic03_dropout ›

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/s92.11792.555+0.438
Tok/s13884.813951.0+66.114
Avg s/step2.3562.392+0.036
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_20260424_145744  ·  full log
XPK Start: Fri Apr 24 15:01:41 UTC 2026
2026-04-24 15:01:45.830206: 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 00:00:1777042905.842935      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1777042905.846602      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1777042905.857848      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777042905.857867      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777042905.857869      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777042905.857871      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-24 15:02:05.182952: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0424 15:02:05.714050 136773766911808 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-24 15:02:14,755:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0424 15:02:14.755385 136773766911808 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-24 15:02:14,757:jax._src.distributed:157: Connecting to JAX distributed service on mt-02-synthetic-n2tsv-slice-job-0-0.mt-02-synthetic-n2tsv:8482
I0424 15:02:14.757887 136773766911808 distributed.py:157] Connecting to JAX distributed service on mt-02-synthetic-n2tsv-slice-job-0-0.mt-02-synthetic-n2tsv:8482
I0424 15:02:15.834186 136773766911808 max_utils.py:284] Jax distributed system initialized!
I0424 15:02:22.798540 136773766911808 max_utils.py:800] System Information: Jax Version: 0.8.1
I0424 15:02:22.798646 136773766911808 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0424 15:02:22.798709 136773766911808 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709
I0424 15:02:22.798745 136773766911808 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 15:02:23.420390 136773766911808 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 15:02:23.420990 136773766911808 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 15:02:23.421174 136773766911808 checkpointing.py:688] Setting up checkpoint logger...
I0424 15:02:23.421230 136773766911808 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 15:02:23.421274 136773766911808 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0424 15:02:23.421629 136773766911808 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 0x7c64aff793d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 15:02:26.564719 136773766911808 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 15:02:26.564961 136773766911808 checkpoint_manager.py:709] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c4f9840db80>}, handler_registry=None
I0424 15:02:26.565195 136773766911808 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c4f9840db80>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0424 15:02:26.565249 136773766911808 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c4f984126c0>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0424 15:02:26.565287 136773766911808 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 0x7c4f9840db80>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c4f9840db80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c4f984126c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c4f984126c0>}).
I0424 15:02:26.565804 136773766911808 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0424 15:02:26.565883 136773766911808 async_checkpointer.py:177] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c4f80564180> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0424 15:02:27.253732 136773766911808 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints
I0424 15:02:27.291681 136773766911808 checkpoint_manager.py:929] [process=5][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c4f9840e0f0>
I0424 15:02:27.292137 136773766911808 checkpointing.py:302] Checkpoint manager created!
I0424 15:02:28.198142 136773766911808 nnx_wrappers.py:455] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 15:02:28.198245 136773766911808 nnx_wrappers.py:455] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 15:02:28.570533 136773766911808 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 15:02:28.570630 136773766911808 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 15:02:28.586498 136773766911808 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 15:02:28.586555 136773766911808 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 15:02:28.609379 136773766911808 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 15:02:28.609445 136773766911808 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 15:02:28.625348 136773766911808 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 15:02:28.625407 136773766911808 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 15:02:28.641099 136773766911808 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 15:02:28.641162 136773766911808 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 15:02:28.665973 136773766911808 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0424 15:02:28.666043 136773766911808 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 15:02:28.685760 136773766911808 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 15:02:28.685824 136773766911808 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 15:02:28.875072 136773766911808 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 15:02:28.875181 136773766911808 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 15:02:30.306215 136773766911808 maxtext_utils.py:1740]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0424 15:02:30.306344 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 15:02:30.306399 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   PartitionSpec('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 15:02:30.306457 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   PartitionSpec('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0424 15:02:30.306512 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0424 15:02:30.306552 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   PartitionSpec('norm', 'layers')
    Physical:  (None, None)
I0424 15:02:30.306606 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0424 15:02:30.306671 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   PartitionSpec('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0424 15:02:30.306716 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0424 15:02:30.306756 136773766911808 maxtext_utils.py:1740]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0424 15:02:30.306806 136773766911808 maxtext_utils.py:1740]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed', 'vocab')
    Physical:  ('fsdp', None)
I0424 15:02:30.306855 136773766911808 maxtext_utils.py:1740]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed')

    Physical:  (None, 'fsdp')
I0424 15:02:30.812424 136773766911808 train.py:158] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 15:02:30.812521 136773766911808 train.py:158] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 15:02:30.827348 136773766911808 train.py:165] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 15:02:30.827410 136773766911808 train.py:165] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 15:02:41.441673 136773766911808 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.
I0424 15:02:41.442536 136773766911808 metric_logger.py:289] number parameters: 1.104 billion
I0424 15:02:52.772444 136773766911808 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 15:02:52.927095 136773766911808 checkpointing.py:798] Waited 0.15463805198669434 seconds for step 0 to finish before starting checkpointing.
I0424 15:02:52.929545 136773766911808 checkpoint_manager.py:2013] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 15:02:52.931158 136773766911808 checkpoint_manager.py:1518] [process=5] Saving checkpoint at step 0
I0424 15:02:52.933870 136773766911808 async_checkpointer.py:452] [process=5] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/0.
I0424 15:02:53.475091 136773766911808 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 15:02:53.476048 136773766911808 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0424 15:02:53.476105 136773766911808 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 15:02:53.726582 136773766911808 base_pytree_checkpoint_handler.py:153] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.251571s
I0424 15:02:53.726764 136773766911808 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/blocking_gbytes_per_sec: 3.002 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5139126777648926 s) (per-host)
I0424 15:02:53.726818 136773766911808 base_pytree_checkpoint_handler.py:737] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.513978s (batch_requests_ready=0.258694s, total_serialization_initiated=0.255211s, others=0.000072s)
I0424 15:02:53.726907 136773766911808 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.517964s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.517945)
I0424 15:02:53.727809 136648293586688 async_checkpointer.py:79] [process=5][thread=async_save] Background save thread started.
I0424 15:02:53.727974 136773766911808 async_checkpointer.py:561] Finished blocking save. Time taken: 0.796746s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/0.
I0424 15:02:54.026104 136773766911808 checkpoint_manager.py:1566] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 15:02:54.026505 136646716544768 async_checkpointer.py:265] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 15:02:54.026684 136773766911808 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777042972.9295282, 'wait_for_prev_duration_secs': 6.341934204101562e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777042972.9311957, 'checkpointer_blocking_duration_secs': 0.7968857288360596, 'get_old_steps_start_time': 1777042973.7280998, 'get_old_steps_duration_secs': 2.7418136596679688e-05, 'checkpoint_manager_blocking_start_time': 1777042972.9276803, 'checkpoint_manager_blocking_duration_secs': 1.0989494323730469}
I0424 15:02:54.026789 136773766911808 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 15:02:54.026844 136773766911808 max_utils.py:750] 
Memstats: After params initialized:
I0424 15:02:54.026903 136773766911808 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0))
I0424 15:02:54.026946 136773766911808 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0))
I0424 15:02:54.026976 136773766911808 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0))
I0424 15:02:54.027000 136773766911808 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0))
I0424 15:02:54.286839 136773766911808 metric_logger.py:185] completed step: 0, seconds: 11.330, TFLOP/s/device: 1.199, Tokens/s/device: 180.762, total_weights: 65536, loss: 10.874
I0424 15:02:54.460808 136773766911808 metric_logger.py:185] completed step: 1, seconds: 1.513, TFLOP/s/device: 8.980, Tokens/s/device: 1353.545, total_weights: 65536, loss: 10.874
I0424 15:02:54.849718 136773766911808 metric_logger.py:185] completed step: 2, seconds: 0.027, TFLOP/s/device: 505.549, Tokens/s/device: 76201.816, total_weights: 65536, loss: 10.263
I0424 15:02:54.997383 136773766911808 metric_logger.py:185] completed step: 3, seconds: 0.389, TFLOP/s/device: 34.892, Tokens/s/device: 5259.373, total_weights: 65536, loss: 9.734
I0424 15:02:55.292496 136773766911808 metric_logger.py:185] completed step: 4, seconds: 0.153, TFLOP/s/device: 89.078, Tokens/s/device: 13426.867, total_weights: 65536, loss: 9.278
I0424 15:02:55.298263 136773766911808 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.139, Tokens/s/device: 13888.230, total_weights: 65536, loss: 8.892
I0424 15:03:13.823352 136773766911808 metric_logger.py:185] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.943, Tokens/s/device: 6925.002, total_weights: 65536, loss: 8.592
I0424 15:03:13.971040 136773766911808 metric_logger.py:185] completed step: 7, seconds: 18.378, TFLOP/s/device: 0.739, Tokens/s/device: 111.435, total_weights: 65536, loss: 8.384
I0424 15:03:14.118211 136773766911808 metric_logger.py:185] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.416, Tokens/s/device: 13477.763, total_weights: 65536, loss: 8.255
I0424 15:03:14.265363 136773766911808 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 15:03:14.266039 136773766911808 checkpointing.py:798] Waited 0.0006961822509765625 seconds for step 9 to finish before starting checkpointing.
I0424 15:03:14.268085 136773766911808 checkpoint_manager.py:2024] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 15:03:14.941082    2601 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 15:03:16.869388 136648301979392 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/0/items/array_metadatas/process_5
I0424 15:03:49.114164 136648293586688 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/gbytes_per_sec: 28.257 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 55.901273012161255 s) (per-host)
I0424 15:03:49.114287 136648293586688 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 55.386368s.
I0424 15:03:57.475132 136648293586688 async_checkpointer.py:144] [process=5][thread=async_save] Background save thread done. Time taken: 63.747197s.
I0424 15:03:57.475421 136646716544768 async_checkpointer.py:273] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 15:03:57.475563 136646716544768 async_checkpointer.py:283] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 15:03:57.475615 136646716544768 checkpoint_manager.py:2133] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 15:03:57.477152 136646716544768 checkpoint_manager.py:2142] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 15:03:57.477327 136773766911808 checkpoint_manager.py:2036] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0424 15:03:57.477465 136773766911808 checkpoint_manager.py:1458] Waiting for previous save to complete took 43.209385 seconds. If this number is high, consider checkpointing less frequently.
I0424 15:03:57.479412 136773766911808 checkpoint_manager.py:1518] [process=5] Saving checkpoint at step 9
I0424 15:03:57.482858 136773766911808 async_checkpointer.py:452] [process=5] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/9.
I0424 15:03:58.027760 136773766911808 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0424 15:03:58.027860 136773766911808 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 15:03:58.073064 136773766911808 base_pytree_checkpoint_handler.py:153] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.046506s
I0424 15:03:58.073228 136773766911808 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.102 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.30237603187561035 s) (per-host)
I0424 15:03:58.073286 136773766911808 base_pytree_checkpoint_handler.py:737] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.302442s (batch_requests_ready=0.253926s, total_serialization_initiated=0.048442s, others=0.000074s)
I0424 15:03:58.073375 136773766911808 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.306391s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.306375)
I0424 15:03:58.074080 136648301979392 async_checkpointer.py:79] [process=5][thread=async_save] Background save thread started.
I0424 15:03:58.074208 136773766911808 async_checkpointer.py:561] Finished blocking save. Time taken: 0.594733s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/9.
I0424 15:03:58.080673 136773766911808 checkpoint_manager.py:1566] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 15:03:58.080971 136646716544768 async_checkpointer.py:265] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 15:03:58.081130 136773766911808 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777042994.268057, 'wait_for_prev_duration_secs': 43.20938491821289, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777043037.4794521, 'checkpointer_blocking_duration_secs': 0.5948929786682129, 'get_old_steps_start_time': 1777043038.0743632, 'get_old_steps_duration_secs': 2.288818359375e-05, 'checkpoint_manager_blocking_start_time': 1777042994.266268, 'checkpoint_manager_blocking_duration_secs': 43.81483054161072}
I0424 15:03:58.081227 136773766911808 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 15:03:58.081279 136773766911808 checkpoint_manager.py:2024] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 15:04:03.040207 136648310372096 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/linen_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/9/items/array_metadatas/process_5
I0424 15:04:39.571078 136648301979392 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/gbytes_per_sec: 37.789 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.80019283294678 s) (per-host)
I0424 15:04:39.571187 136648301979392 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.496912s.
I0424 15:04:48.328086 136648301979392 async_checkpointer.py:144] [process=5][thread=async_save] Background save thread done. Time taken: 50.253796s.
I0424 15:04:48.328376 136646716544768 async_checkpointer.py:273] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 15:04:48.328518 136646716544768 async_checkpointer.py:283] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 15:04:48.328570 136646716544768 checkpoint_manager.py:2133] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 15:04:48.330098 136646716544768 checkpoint_manager.py:2142] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 15:04:48.330259 136773766911808 checkpoint_manager.py:2036] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 15:04:48.330408 136773766911808 checkpoint_manager.py:2013] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 15:04:48.331076 136773766911808 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.117, Tokens/s/device: 13884.840, 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: Fri Apr 24 15:04:57 UTC 2026
EXIT_CODE=0
NNX  ·  73213e044  ·  feat_nnx_set_defaults_true_20260424_145744  ·  full log
XPK Start: Fri Apr 24 16:32:39 UTC 2026
2026-04-24 16:32:43.602064: 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 00:00:1777048363.614471      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1777048363.618147      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1777048363.629459      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777048363.629477      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777048363.629480      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1777048363.629481      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-24 16:33:02.788531: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0424 16:33:03.318047 139682481080128 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-24 16:33:12,359:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0424 16:33:12.359210 139682481080128 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-24 16:33:12,361:jax._src.distributed:157: Connecting to JAX distributed service on mt-02-synthetic-q1hpo-slice-job-0-0.mt-02-synthetic-q1hpo:8482
I0424 16:33:12.361529 139682481080128 distributed.py:157] Connecting to JAX distributed service on mt-02-synthetic-q1hpo-slice-job-0-0.mt-02-synthetic-q1hpo:8482
I0424 16:33:13.558737 139682481080128 max_utils.py:284] Jax distributed system initialized!
I0424 16:33:20.726229 139682481080128 max_utils.py:800] System Information: Jax Version: 0.8.1
I0424 16:33:20.726339 139682481080128 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0424 16:33:20.726381 139682481080128 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709
I0424 16:33:20.726417 139682481080128 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 16:33:21.352683 139682481080128 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 16:33:21.466772 139682481080128 checkpointing.py:688] Setting up checkpoint logger...
I0424 16:33:21.466888 139682481080128 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 16:33:21.466933 139682481080128 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0424 16:33:21.467141 139682481080128 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 0x7f09d0d75340>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 16:33:24.464019 139682481080128 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 16:33:24.464365 139682481080128 checkpoint_manager.py:709] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ef6b8c30530>}, handler_registry=None
I0424 16:33:24.464598 139682481080128 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ef6b8c30530>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0424 16:33:24.464648 139682481080128 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ef524653b30>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0424 16:33:24.464696 139682481080128 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 0x7ef6b8c30530>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ef6b8c30530>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ef524653b30>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ef524653b30>}).
I0424 16:33:24.465030 139682481080128 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0424 16:33:24.465108 139682481080128 async_checkpointer.py:177] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7ef5245c4f40> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0424 16:33:25.432743 139682481080128 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints
I0424 16:33:25.635792 139682481080128 checkpoint_manager.py:929] [process=5][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ef524651040>
I0424 16:33:25.635972 139682481080128 checkpointing.py:302] Checkpoint manager created!
I0424 16:33:26.570711 139682481080128 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 16:33:26.570832 139682481080128 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 16:33:27.999383 139682481080128 maxtext_utils.py:1749]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0424 16:33:27.999514 139682481080128 maxtext_utils.py:1749]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0424 16:33:27.999567 139682481080128 maxtext_utils.py:1749]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0424 16:33:27.999613 139682481080128 maxtext_utils.py:1749]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0424 16:33:27.999662 139682481080128 maxtext_utils.py:1749]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0424 16:33:27.999700 139682481080128 maxtext_utils.py:1749]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0424 16:33:27.999739 139682481080128 maxtext_utils.py:1749]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0424 16:33:27.999777 139682481080128 maxtext_utils.py:1749]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0424 16:33:27.999813 139682481080128 maxtext_utils.py:1749]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0424 16:33:27.999849 139682481080128 maxtext_utils.py:1749]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0424 16:33:27.999883 139682481080128 maxtext_utils.py:1749]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)

I0424 16:33:27.999917 139682481080128 maxtext_utils.py:1749]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')
I0424 16:33:28.391461 139682481080128 nnx_decoders.py:616] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 16:33:28.391622 139682481080128 nnx_decoders.py:616] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 16:33:28.397036 139682481080128 nnx_decoders.py:616] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 16:33:28.397133 139682481080128 nnx_decoders.py:616] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 16:33:28.548563 139682481080128 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 16:33:28.548718 139682481080128 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 16:33:28.701582 139682481080128 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 16:33:28.701732 139682481080128 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 16:33:28.862612 139682481080128 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 16:33:28.862769 139682481080128 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 16:33:29.008620 139682481080128 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 16:33:29.008819 139682481080128 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 16:33:29.146802 139682481080128 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 16:33:29.146957 139682481080128 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 16:33:29.300861 139682481080128 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0424 16:33:29.300998 139682481080128 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 16:33:29.323964 139682481080128 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 16:33:29.324061 139682481080128 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 16:33:40.051289 139682481080128 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.
I0424 16:33:40.053642 139682481080128 metric_logger.py:289] number parameters: 1.104 billion
I0424 16:33:50.979709 139682481080128 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 16:33:51.292346 139682481080128 checkpointing.py:798] Waited 0.3126239776611328 seconds for step 0 to finish before starting checkpointing.
I0424 16:33:51.294881 139682481080128 checkpoint_manager.py:2013] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 16:33:51.296626 139682481080128 checkpoint_manager.py:1518] [process=5] Saving checkpoint at step 0
I0424 16:33:51.299331 139682481080128 async_checkpointer.py:452] [process=5] Started async saving checkpoint to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/0.
I0424 16:33:52.033682 139682481080128 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 16:33:52.034622 139682481080128 jax_array_handlers.py:358] Scheduling D2H of 69 prioritized jax.Array.
I0424 16:33:52.034693 139682481080128 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 16:33:52.284539 139682481080128 base_pytree_checkpoint_handler.py:153] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.251979s
I0424 16:33:52.284739 139682481080128 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.911 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5298876762390137 s) (per-host)
I0424 16:33:52.284797 139682481080128 base_pytree_checkpoint_handler.py:737] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.529958s (batch_requests_ready=0.271430s, total_serialization_initiated=0.258450s, others=0.000078s)
I0424 16:33:52.284894 139682481080128 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.534077s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=0.534058)
I0424 16:33:52.285827 139558779873024 async_checkpointer.py:79] [process=5][thread=async_save] Background save thread started.
I0424 16:33:52.285995 139682481080128 async_checkpointer.py:561] Finished blocking save. Time taken: 0.989280s. Continuing background save to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/0.
I0424 16:33:52.287784 139682481080128 checkpoint_manager.py:1566] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 16:33:52.288100 139554493298432 async_checkpointer.py:265] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 16:33:52.288229 139682481080128 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777048431.2948625, 'wait_for_prev_duration_secs': 7.43865966796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777048431.2966852, 'checkpointer_blocking_duration_secs': 0.9894125461578369, 'get_old_steps_start_time': 1777048432.2861161, 'get_old_steps_duration_secs': 2.4318695068359375e-05, 'checkpoint_manager_blocking_start_time': 1777048431.2927778, 'checkpoint_manager_blocking_duration_secs': 0.9954161643981934}
I0424 16:33:52.288396 139682481080128 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 16:33:52.288468 139682481080128 max_utils.py:750] 
Memstats: After params initialized:
I0424 16:33:52.288517 139682481080128 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0))
I0424 16:33:52.288552 139682481080128 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0))
I0424 16:33:52.288579 139682481080128 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0))
I0424 16:33:52.288604 139682481080128 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0))
I0424 16:33:52.557744 139682481080128 metric_logger.py:185] completed step: 0, seconds: 10.925, TFLOP/s/device: 1.244, Tokens/s/device: 187.460, total_weights: 65536, loss: 10.874
I0424 16:33:52.722639 139682481080128 metric_logger.py:185] completed step: 1, seconds: 1.576, TFLOP/s/device: 8.619, Tokens/s/device: 1299.215, total_weights: 65536, loss: 10.874
I0424 16:33:53.138109 139682481080128 metric_logger.py:185] completed step: 2, seconds: 0.019, TFLOP/s/device: 711.443, Tokens/s/device: 107236.360, total_weights: 65536, loss: 10.262
I0424 16:33:53.285414 139682481080128 metric_logger.py:185] completed step: 3, seconds: 0.407, TFLOP/s/device: 33.399, Tokens/s/device: 5034.217, total_weights: 65536, loss: 9.732
I0424 16:33:53.581207 139682481080128 metric_logger.py:185] completed step: 4, seconds: 0.162, TFLOP/s/device: 83.806, Tokens/s/device: 12632.150, total_weights: 65536, loss: 9.274
I0424 16:33:53.588168 139682481080128 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.489, Tokens/s/device: 13940.982, total_weights: 65536, loss: 8.888
I0424 16:34:12.355370 139682481080128 metric_logger.py:185] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.862, Tokens/s/device: 6912.870, total_weights: 65536, loss: 8.589
I0424 16:34:12.502733 139682481080128 metric_logger.py:185] completed step: 7, seconds: 18.621, TFLOP/s/device: 0.730, Tokens/s/device: 109.983, total_weights: 65536, loss: 8.380
I0424 16:34:12.650110 139682481080128 metric_logger.py:185] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.820, Tokens/s/device: 13387.984, total_weights: 65536, loss: 8.251
I0424 16:34:12.798569 139682481080128 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 16:34:12.799132 139682481080128 checkpointing.py:798] Waited 0.0005738735198974609 seconds for step 9 to finish before starting checkpointing.
I0424 16:34:12.801142 139682481080128 checkpoint_manager.py:2024] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 16:34:13.837426    2654 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 16:34:15.810332 139558788265728 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/0/items/array_metadatas/process_5
I0424 16:34:47.904824 139558779873024 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/gbytes_per_sec: 28.132 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 56.14993715286255 s) (per-host)
I0424 16:34:47.904932 139558779873024 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 55.618991s.
I0424 16:34:56.462738 139558779873024 async_checkpointer.py:144] [process=5][thread=async_save] Background save thread done. Time taken: 64.176779s.
I0424 16:34:56.463064 139554493298432 async_checkpointer.py:273] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 16:34:56.463182 139554493298432 async_checkpointer.py:283] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 16:34:56.463234 139554493298432 checkpoint_manager.py:2133] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 16:34:56.464864 139554493298432 checkpoint_manager.py:2142] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 16:34:56.465044 139682481080128 checkpoint_manager.py:2036] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0424 16:34:56.465185 139682481080128 checkpoint_manager.py:1458] Waiting for previous save to complete took 43.664038 seconds. If this number is high, consider checkpointing less frequently.
I0424 16:34:56.467260 139682481080128 checkpoint_manager.py:1518] [process=5] Saving checkpoint at step 9
I0424 16:34:56.470549 139682481080128 async_checkpointer.py:452] [process=5] Started async saving checkpoint to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/9.
I0424 16:34:57.037991 139682481080128 jax_array_handlers.py:358] Scheduling D2H of 69 prioritized jax.Array.
I0424 16:34:57.038155 139682481080128 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 16:34:57.082629 139682481080128 base_pytree_checkpoint_handler.py:153] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.046642s
I0424 16:34:57.082808 139682481080128 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.056 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.30509519577026367 s) (per-host)
I0424 16:34:57.082862 139682481080128 base_pytree_checkpoint_handler.py:737] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.305158s (batch_requests_ready=0.253784s, total_serialization_initiated=0.051303s, others=0.000071s)
I0424 16:34:57.082957 139682481080128 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.309038s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.309021)
I0424 16:34:57.083848 139556056475392 async_checkpointer.py:79] [process=5][thread=async_save] Background save thread started.
I0424 16:34:57.084002 139682481080128 async_checkpointer.py:561] Finished blocking save. Time taken: 0.616672s. Continuing background save to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/9.
I0424 16:34:57.094034 139682481080128 checkpoint_manager.py:1566] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 16:34:57.094264 139554493298432 async_checkpointer.py:265] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 16:34:57.094399 139682481080128 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777048452.8011134, 'wait_for_prev_duration_secs': 43.66403794288635, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777048496.4673011, 'checkpointer_blocking_duration_secs': 0.616809606552124, 'get_old_steps_start_time': 1777048497.0841303, 'get_old_steps_duration_secs': 2.7179718017578125e-05, 'checkpoint_manager_blocking_start_time': 1777048452.7993586, 'checkpoint_manager_blocking_duration_secs': 44.295007944107056}
I0424 16:34:57.094506 139682481080128 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 16:34:57.094551 139682481080128 checkpoint_manager.py:2024] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 16:35:02.172944 139558788265728 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260424_145744/nnx_xpk_feat_nnx_set_defaults_true_20260424_145744_02_synthetic/checkpoints/9/items/array_metadatas/process_5
I0424 16:35:40.690466 139556056475392 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/gbytes_per_sec: 35.971 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.91271948814392 s) (per-host)
I0424 16:35:40.690572 139556056475392 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.606615s.
I0424 16:35:46.385403 139556056475392 async_checkpointer.py:144] [process=5][thread=async_save] Background save thread done. Time taken: 49.301430s.
I0424 16:35:46.385625 139554493298432 async_checkpointer.py:273] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 16:35:46.385697 139554493298432 async_checkpointer.py:283] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 16:35:46.385752 139554493298432 checkpoint_manager.py:2133] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 16:35:46.387317 139554493298432 checkpoint_manager.py:2142] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 16:35:46.387474 139682481080128 checkpoint_manager.py:2036] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 16:35:46.387642 139682481080128 checkpoint_manager.py:2013] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 16:35:46.388330 139682481080128 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.555, Tokens/s/device: 13950.954, 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: Fri Apr 24 16:35:59 UTC 2026
EXIT_CODE=0