MaxView

‹ 05_fp8Case: 06_grad_accum07_eval ›

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

MetricLinen  574ad3fb9NNX  574ad3fb9Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss9.45709.45700
TFLOP/s93.94692.611-1.335
Tok/s14160.613959.3-201.22
Avg s/step5.9956.040+0.045
Memory %1.381.380
JAX0.8.10.8.1

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

Linen  ·  574ad3fb9  ·  feat_nnx_set_defaults_true_20260419_025828  ·  full log
XPK Start: Sun Apr 19 03:30:34 UTC 2026
2026-04-19 03:30:38.441578: 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 00:00:1776569438.454199       9 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776569438.457911       9 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776569438.469347       9 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776569438.469364       9 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776569438.469366       9 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776569438.469368       9 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-19 03:30:57.737811: 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)
I0419 03:30:58.263913 137565018941248 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-19 03:31:07,305:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0419 03:31:07.305313 137565018941248 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-19 03:31:07,307:jax._src.distributed:157: Connecting to JAX distributed service on mt-06-grad-accum-30h2s-slice-job-0-0.mt-06-grad-accum-30h2s:8482
I0419 03:31:07.307692 137565018941248 distributed.py:157] Connecting to JAX distributed service on mt-06-grad-accum-30h2s-slice-job-0-0.mt-06-grad-accum-30h2s:8482
I0419 03:31:09.277634 137565018941248 max_utils.py:284] Jax distributed system initialized!
I0419 03:31:16.092916 137565018941248 max_utils.py:800] System Information: Jax Version: 0.8.1
I0419 03:31:16.093024 137565018941248 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0419 03:31:16.093065 137565018941248 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
I0419 03:31:16.093101 137565018941248 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0419 03:31:16.711641 137565018941248 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 03:31:16.712268 137565018941248 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 03:31:16.712453 137565018941248 checkpointing.py:688] Setting up checkpoint logger...
I0419 03:31:16.712502 137565018941248 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0419 03:31:16.712545 137565018941248 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0419 03:31:16.712891 137565018941248 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 0x7d1cc9371310>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0419 03:31:20.920358 137565018941248 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0419 03:31:20.920593 137565018941248 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 0x7d0967c21a00>}, handler_registry=None
I0419 03:31:20.920826 137565018941248 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d0967c21a00>` 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`.
I0419 03:31:20.920875 137565018941248 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d0967c26e70>` 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`.
I0419 03:31:20.920911 137565018941248 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 0x7d0967c21a00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d0967c21a00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d0967c26e70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d0967c26e70>}).
I0419 03:31:20.921431 137565018941248 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0419 03:31:20.921509 137565018941248 async_checkpointer.py:177] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d09667dc180> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0419 03:31:23.282353 137565018941248 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints
I0419 03:31:23.719413 137565018941248 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_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d0967b8d670>
I0419 03:31:23.719589 137565018941248 checkpointing.py:302] Checkpoint manager created!
I0419 03:31:25.406169 137565018941248 nnx_wrappers.py:455] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 03:31:25.406291 137565018941248 nnx_wrappers.py:455] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 03:31:25.787695 137565018941248 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 03:31:25.787792 137565018941248 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 03:31:25.803815 137565018941248 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 03:31:25.803879 137565018941248 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 03:31:25.826701 137565018941248 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 03:31:25.826771 137565018941248 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 03:31:25.842770 137565018941248 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 03:31:25.842832 137565018941248 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 03:31:25.858689 137565018941248 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 03:31:25.858748 137565018941248 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 03:31:25.883632 137565018941248 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0419 03:31:25.883703 137565018941248 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 03:31:25.903590 137565018941248 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0419 03:31:25.903660 137565018941248 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0419 03:31:26.096900 137565018941248 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0419 03:31:26.097014 137565018941248 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0419 03:31:27.525939 137565018941248 maxtext_utils.py:1740]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0419 03:31:27.526071 137565018941248 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)
I0419 03:31:27.526127 137565018941248 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)
I0419 03:31:27.526187 137565018941248 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')
I0419 03:31:27.526267 137565018941248 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)
I0419 03:31:27.526320 137565018941248 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)
I0419 03:31:27.526380 137565018941248 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)
I0419 03:31:27.526434 137565018941248 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')
I0419 03:31:27.526482 137565018941248 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)
I0419 03:31:27.526522 137565018941248 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)
I0419 03:31:27.526571 137565018941248 maxtext_utils.py:1740]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed', 'vocab')
    Physical:  ('fsdp', None)
I0419 03:31:27.526621 137565018941248 maxtext_utils.py:1740]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed')
    Physical:  (None, 'fsdp')

I0419 03:31:27.548681 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown.
I0419 03:31:27.548747 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,).
I0419 03:31:27.562978 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown.
I0419 03:31:27.563037 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None).
I0419 03:31:27.591219 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown.
I0419 03:31:27.591295 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp').
I0419 03:31:27.605441 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown.
I0419 03:31:27.605500 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None).
I0419 03:31:27.633669 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown.
I0419 03:31:27.633739 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None).
I0419 03:31:27.647993 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown.
I0419 03:31:27.648050 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp').
I0419 03:31:27.695859 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown.
I0419 03:31:27.695953 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None).
I0419 03:31:27.710168 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown.
I0419 03:31:27.710226 137565018941248 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp').
I0419 03:31:28.397489 137565018941248 train.py:158] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0419 03:31:28.397591 137565018941248 train.py:158] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0419 03:31:28.412992 137565018941248 train.py:165] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0419 03:31:28.413055 137565018941248 train.py:165] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0419 03:31:39.726966 137565018941248 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0419 03:31:39.727890 137565018941248 metric_logger.py:289] number parameters: 1.104 billion
I0419 03:31:41.858049 137565018941248 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0419 03:31:52.046471 137565018941248 checkpointing.py:798] Waited 10.188400745391846 seconds for step 0 to finish before starting checkpointing.
I0419 03:31:52.049207 137565018941248 checkpoint_manager.py:2013] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 03:31:52.051320 137565018941248 checkpoint_manager.py:1518] [process=6] Saving checkpoint at step 0
I0419 03:31:52.054208 137565018941248 async_checkpointer.py:452] [process=6] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/0.
I0419 03:31:53.869287 137565018941248 signaling_client.py:364] Using JaxDistributedSignalingClient
I0419 03:31:53.870517 137565018941248 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0419 03:31:53.870574 137565018941248 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
I0419 03:31:54.114387 137565018941248 base_pytree_checkpoint_handler.py:153] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.245196s
I0419 03:31:54.114569 137565018941248 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.135 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 1.3585307598114014 s) (per-host)
I0419 03:31:54.114624 137565018941248 base_pytree_checkpoint_handler.py:737] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 1.358599s (batch_requests_ready=1.109794s, total_serialization_initiated=0.248729s, others=0.000076s)
I0419 03:31:54.114718 137565018941248 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.362732s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=1.362713)
I0419 03:31:54.115655 137440698300160 async_checkpointer.py:79] [process=6][thread=async_save] Background save thread started.
I0419 03:31:54.115789 137565018941248 async_checkpointer.py:561] Finished blocking save. Time taken: 2.064395s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/0.
I0419 03:31:54.117650 137565018941248 checkpoint_manager.py:1566] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 03:31:54.117930 137436889958144 async_checkpointer.py:265] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0419 03:31:54.118083 137565018941248 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776569512.049189, 'wait_for_prev_duration_secs': 8.392333984375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776569512.0513606, 'checkpointer_blocking_duration_secs': 2.0646071434020996, 'get_old_steps_start_time': 1776569514.1159878, 'get_old_steps_duration_secs': 2.384185791015625e-05, 'checkpoint_manager_blocking_start_time': 1776569512.046948, 'checkpoint_manager_blocking_duration_secs': 2.071099042892456}
I0419 03:31:54.118189 137565018941248 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0419 03:31:54.118255 137565018941248 max_utils.py:750] 
Memstats: After params initialized:
I0419 03:31:54.118304 137565018941248 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0419 03:31:54.118338 137565018941248 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0419 03:31:54.118365 137565018941248 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0419 03:31:54.118391 137565018941248 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0419 03:31:54.375028 137565018941248 metric_logger.py:185] completed step: 0, seconds: 2.130, TFLOP/s/device: 25.515, Tokens/s/device: 3845.948, total_weights: 262144, loss: 10.877
I0419 03:31:54.989553 137565018941248 metric_logger.py:185] completed step: 1, seconds: 12.516, TFLOP/s/device: 4.342, Tokens/s/device: 654.532, total_weights: 262144, loss: 10.877
I0419 03:31:55.568495 137565018941248 metric_logger.py:185] completed step: 2, seconds: 0.036, TFLOP/s/device: 1509.555, Tokens/s/device: 227536.594, total_weights: 262144, loss: 10.563
I0419 03:31:56.147137 137565018941248 metric_logger.py:185] completed step: 3, seconds: 0.584, TFLOP/s/device: 93.065, Tokens/s/device: 14027.806, total_weights: 262144, loss: 10.272
I0419 03:31:56.265930    2481 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0419 03:31:57.304738 137565018941248 metric_logger.py:185] completed step: 4, seconds: 0.579, TFLOP/s/device: 93.898, Tokens/s/device: 14153.299, total_weights: 262144, loss: 10.022
I0419 03:31:57.310156 137565018941248 metric_logger.py:185] completed step: 5, seconds: 0.579, TFLOP/s/device: 93.920, Tokens/s/device: 14156.600, total_weights: 262144, loss: 9.820
I0419 03:31:58.567621 137441184814848 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_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/0/items/array_metadatas/process_6
I0419 03:32:30.359037 137440698300160 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/gbytes_per_sec: 42.007 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.60296392440796 s) (per-host)
I0419 03:32:30.359160 137440698300160 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.243294s.
I0419 03:32:35.232564 137565018941248 metric_logger.py:185] completed step: 6, seconds: 1.158, TFLOP/s/device: 46.929, Tokens/s/device: 7073.661, total_weights: 262144, loss: 9.667
I0419 03:32:35.811192 137565018941248 metric_logger.py:185] completed step: 7, seconds: 37.344, TFLOP/s/device: 1.455, Tokens/s/device: 219.367, total_weights: 262144, loss: 9.561
I0419 03:32:36.389670 137565018941248 metric_logger.py:185] completed step: 8, seconds: 0.584, TFLOP/s/device: 93.128, Tokens/s/device: 14037.300, total_weights: 262144, loss: 9.496
I0419 03:32:36.968018 137565018941248 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0419 03:32:36.968687 137565018941248 checkpointing.py:798] Waited 0.0006909370422363281 seconds for step 9 to finish before starting checkpointing.
I0419 03:32:36.970645 137565018941248 checkpoint_manager.py:2024] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 03:33:15.863502 137440698300160 async_checkpointer.py:144] [process=6][thread=async_save] Background save thread done. Time taken: 81.747619s.
I0419 03:33:15.863796 137436889958144 async_checkpointer.py:273] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 03:33:15.863922 137436889958144 async_checkpointer.py:283] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0419 03:33:15.863986 137436889958144 checkpoint_manager.py:2133] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0419 03:33:15.865615 137436889958144 checkpoint_manager.py:2142] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0419 03:33:15.865806 137565018941248 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.
W0419 03:33:15.865886 137565018941248 checkpoint_manager.py:1458] Waiting for previous save to complete took 38.895243 seconds. If this number is high, consider checkpointing less frequently.
I0419 03:33:15.867645 137565018941248 checkpoint_manager.py:1518] [process=6] Saving checkpoint at step 9
I0419 03:33:15.870959 137565018941248 async_checkpointer.py:452] [process=6] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/9.
I0419 03:33:16.855583 137565018941248 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0419 03:33:16.855683 137565018941248 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
I0419 03:33:16.900646 137565018941248 base_pytree_checkpoint_handler.py:153] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.046093s
I0419 03:33:16.900816 137565018941248 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.003 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.30832338333129883 s) (per-host)
I0419 03:33:16.900872 137565018941248 base_pytree_checkpoint_handler.py:737] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.308388s (batch_requests_ready=0.260320s, total_serialization_initiated=0.047995s, others=0.000074s)
I0419 03:33:16.900962 137565018941248 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.312200s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=0.312184)
I0419 03:33:16.901829 137436881565440 async_checkpointer.py:79] [process=6][thread=async_save] Background save thread started.
I0419 03:33:16.901938 137565018941248 async_checkpointer.py:561] Finished blocking save. Time taken: 1.034218s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/9.
I0419 03:33:17.353435 137565018941248 checkpoint_manager.py:1566] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 03:33:17.353858 137436889958144 async_checkpointer.py:265] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0419 03:33:17.354037 137565018941248 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776569556.9706163, 'wait_for_prev_duration_secs': 38.895243406295776, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776569595.8676848, 'checkpointer_blocking_duration_secs': 1.034364938735962, 'get_old_steps_start_time': 1776569596.9020681, 'get_old_steps_duration_secs': 2.3126602172851562e-05, 'checkpoint_manager_blocking_start_time': 1776569556.9689028, 'checkpoint_manager_blocking_duration_secs': 40.3850953578949}
I0419 03:33:17.354146 137565018941248 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0419 03:33:17.354194 137565018941248 checkpoint_manager.py:2024] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 03:33:22.863678 137441184814848 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_20260419_025828/linen_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/9/items/array_metadatas/process_6
I0419 03:33:58.817610 137436881565440 base_pytree_checkpoint_handler.py:129] [process=6] /jax/checkpoint/write/gbytes_per_sec: 37.409 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.22506856918335 s) (per-host)
I0419 03:33:58.817736 137436881565440 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.915846s.
I0419 03:34:08.083345 137436881565440 async_checkpointer.py:144] [process=6][thread=async_save] Background save thread done. Time taken: 51.181440s.
I0419 03:34:08.083662 137436889958144 async_checkpointer.py:273] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 03:34:08.083786 137436889958144 async_checkpointer.py:283] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0419 03:34:08.083832 137436889958144 checkpoint_manager.py:2133] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0419 03:34:08.085878 137436889958144 checkpoint_manager.py:2142] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0419 03:34:08.086063 137565018941248 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.
I0419 03:34:08.086210 137565018941248 checkpoint_manager.py:2013] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 03:34:08.086846 137565018941248 metric_logger.py:185] completed step: 9, seconds: 0.579, TFLOP/s/device: 93.946, Tokens/s/device: 14160.565, total_weights: 262144, loss: 9.457
Per train step:
 Total TFLOPs: 54.35 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Sun Apr 19 03:34:19 UTC 2026
EXIT_CODE=0
NNX  ·  574ad3fb9  ·  feat_nnx_set_defaults_true_20260419_025828  ·  full log
XPK Start: Sun Apr 19 05:08:51 UTC 2026
2026-04-19 05:08:55.193178: 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 00:00:1776575335.205809      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776575335.209530      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776575335.220760      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776575335.220778      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776575335.220781      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776575335.220783      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-19 05:09:14.384943: 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)
I0419 05:09:14.913326 135306851862336 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-19 05:09:23,954:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0419 05:09:23.954134 135306851862336 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-19 05:09:23,956:jax._src.distributed:157: Connecting to JAX distributed service on mt-06-grad-accum-i7fs3-slice-job-0-0.mt-06-grad-accum-i7fs3:8482
I0419 05:09:23.956533 135306851862336 distributed.py:157] Connecting to JAX distributed service on mt-06-grad-accum-i7fs3-slice-job-0-0.mt-06-grad-accum-i7fs3:8482
I0419 05:09:25.632728 135306851862336 max_utils.py:284] Jax distributed system initialized!
I0419 05:09:32.552972 135306851862336 max_utils.py:800] System Information: Jax Version: 0.8.1
I0419 05:09:32.553079 135306851862336 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0419 05:09:32.553122 135306851862336 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
I0419 05:09:32.553158 135306851862336 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0419 05:09:33.165302 135306851862336 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 05:09:33.278962 135306851862336 checkpointing.py:688] Setting up checkpoint logger...
I0419 05:09:33.279083 135306851862336 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0419 05:09:33.279129 135306851862336 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0419 05:09:33.279355 135306851862336 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 0x7b0f08f752e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0419 05:09:37.373973 135306851862336 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0419 05:09:37.374213 135306851862336 checkpoint_manager.py:709] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7af9f47c02c0>}, handler_registry=None
I0419 05:09:37.374462 135306851862336 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7af9f47c02c0>` 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`.
I0419 05:09:37.374512 135306851862336 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7afa847d72f0>` 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`.
I0419 05:09:37.374548 135306851862336 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 0x7af9f47c02c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7af9f47c02c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7afa847d72f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7afa847d72f0>}).
I0419 05:09:37.374890 135306851862336 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0419 05:09:37.374970 135306851862336 async_checkpointer.py:177] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7afa841e0f40> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0419 05:09:38.943948 135306851862336 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints
I0419 05:09:39.376986 135306851862336 checkpoint_manager.py:929] [process=3][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7afac40c5bb0>
I0419 05:09:39.377162 135306851862336 checkpointing.py:302] Checkpoint manager created!
I0419 05:09:41.065435 135306851862336 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0419 05:09:41.065560 135306851862336 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0419 05:09:42.470944 135306851862336 maxtext_utils.py:1749]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0419 05:09:42.471065 135306851862336 maxtext_utils.py:1749]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0419 05:09:42.471119 135306851862336 maxtext_utils.py:1749]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0419 05:09:42.471164 135306851862336 maxtext_utils.py:1749]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0419 05:09:42.471203 135306851862336 maxtext_utils.py:1749]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0419 05:09:42.471249 135306851862336 maxtext_utils.py:1749]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0419 05:09:42.471289 135306851862336 maxtext_utils.py:1749]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 05:09:42.471326 135306851862336 maxtext_utils.py:1749]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0419 05:09:42.471361 135306851862336 maxtext_utils.py:1749]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 05:09:42.471396 135306851862336 maxtext_utils.py:1749]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 05:09:42.471430 135306851862336 maxtext_utils.py:1749]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)

I0419 05:09:42.471463 135306851862336 maxtext_utils.py:1749]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')
I0419 05:09:42.850617 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown.
I0419 05:09:42.850713 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,).
I0419 05:09:42.865113 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown.
I0419 05:09:42.865176 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None).
I0419 05:09:42.893646 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown.
I0419 05:09:42.893715 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp').
I0419 05:09:42.907918 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown.
I0419 05:09:42.907978 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None).
I0419 05:09:42.936501 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown.
I0419 05:09:42.936570 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None).
I0419 05:09:42.950967 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown.
I0419 05:09:42.951030 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp').
I0419 05:09:42.993866 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown.
I0419 05:09:42.993947 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None).
I0419 05:09:43.008204 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown.
I0419 05:09:43.008275 135306851862336 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp').
I0419 05:09:43.226397 135306851862336 nnx_decoders.py:616] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 05:09:43.226489 135306851862336 nnx_decoders.py:616] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 05:09:43.230481 135306851862336 nnx_decoders.py:616] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 05:09:43.230535 135306851862336 nnx_decoders.py:616] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 05:09:43.246609 135306851862336 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 05:09:43.246670 135306851862336 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 05:09:43.262170 135306851862336 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 05:09:43.262227 135306851862336 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 05:09:43.285575 135306851862336 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 05:09:43.285645 135306851862336 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 05:09:43.301394 135306851862336 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 05:09:43.301459 135306851862336 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 05:09:43.317087 135306851862336 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 05:09:43.317146 135306851862336 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 05:09:43.342291 135306851862336 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0419 05:09:43.342360 135306851862336 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 05:09:43.361905 135306851862336 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0419 05:09:43.361968 135306851862336 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0419 05:09:54.964310 135306851862336 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0419 05:09:54.966683 135306851862336 metric_logger.py:289] number parameters: 1.104 billion
I0419 05:09:56.822947 135306851862336 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0419 05:10:07.043578 135306851862336 checkpointing.py:798] Waited 10.22061538696289 seconds for step 0 to finish before starting checkpointing.
I0419 05:10:07.045910 135306851862336 checkpoint_manager.py:2013] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 05:10:07.047725 135306851862336 checkpoint_manager.py:1518] [process=3] Saving checkpoint at step 0
I0419 05:10:07.050502 135306851862336 async_checkpointer.py:452] [process=3] Started async saving checkpoint to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/0.
I0419 05:10:08.028313 135306851862336 signaling_client.py:364] Using JaxDistributedSignalingClient
I0419 05:10:08.029554 135306851862336 jax_array_handlers.py:358] Scheduling D2H of 69 prioritized jax.Array.
I0419 05:10:08.029614 135306851862336 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
I0419 05:10:08.272157 135306851862336 base_pytree_checkpoint_handler.py:153] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.244975s
I0419 05:10:08.272347 135306851862336 base_pytree_checkpoint_handler.py:129] [process=3] /jax/checkpoint/write/blocking_gbytes_per_sec: 3.080 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5007822513580322 s) (per-host)
I0419 05:10:08.272408 135306851862336 base_pytree_checkpoint_handler.py:737] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.500855s (batch_requests_ready=0.249038s, total_serialization_initiated=0.251736s, others=0.000081s)
I0419 05:10:08.272516 135306851862336 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.505081s (all_items=0.000021s, per_item={'items': '0.00002050'}, temp_paths=0.505060)
I0419 05:10:08.273452 135184531830528 async_checkpointer.py:79] [process=3][thread=async_save] Background save thread started.
I0419 05:10:08.273618 135306851862336 async_checkpointer.py:561] Finished blocking save. Time taken: 1.225821s. Continuing background save to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/0.
I0419 05:10:08.717743 135306851862336 checkpoint_manager.py:1566] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 05:10:08.718131 135181335783168 async_checkpointer.py:265] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0419 05:10:08.718320 135306851862336 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776575407.0458915, 'wait_for_prev_duration_secs': 6.723403930664062e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776575407.0477648, 'checkpointer_blocking_duration_secs': 1.225964069366455, 'get_old_steps_start_time': 1776575408.2737494, 'get_old_steps_duration_secs': 2.6702880859375e-05, 'checkpoint_manager_blocking_start_time': 1776575407.0439847, 'checkpoint_manager_blocking_duration_secs': 1.6742949485778809}
I0419 05:10:08.718432 135306851862336 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0419 05:10:08.718504 135306851862336 max_utils.py:750] 
Memstats: After params initialized:
I0419 05:10:08.718559 135306851862336 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0))
I0419 05:10:08.718593 135306851862336 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0))
I0419 05:10:08.718620 135306851862336 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0))
I0419 05:10:08.718644 135306851862336 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0))
I0419 05:10:08.987684 135306851862336 metric_logger.py:185] completed step: 0, seconds: 1.855, TFLOP/s/device: 29.295, Tokens/s/device: 4415.649, total_weights: 262144, loss: 10.876
I0419 05:10:09.679537 135306851862336 metric_logger.py:185] completed step: 1, seconds: 12.163, TFLOP/s/device: 4.468, Tokens/s/device: 673.512, total_weights: 262144, loss: 10.876
I0419 05:10:10.258964 135306851862336 metric_logger.py:185] completed step: 2, seconds: 0.115, TFLOP/s/device: 471.616, Tokens/s/device: 71087.045, total_weights: 262144, loss: 10.561
I0419 05:10:10.837725 135306851862336 metric_logger.py:185] completed step: 3, seconds: 0.584, TFLOP/s/device: 93.020, Tokens/s/device: 14021.035, total_weights: 262144, loss: 10.270
I0419 05:10:11.996185 135306851862336 metric_logger.py:185] completed step: 4, seconds: 0.580, TFLOP/s/device: 93.748, Tokens/s/device: 14130.765, total_weights: 262144, loss: 10.021
I0419 05:10:12.004565 135306851862336 metric_logger.py:185] completed step: 5, seconds: 0.578, TFLOP/s/device: 93.972, Tokens/s/device: 14164.556, total_weights: 262144, loss: 9.819
I0419 05:10:50.591493 135306851862336 metric_logger.py:185] completed step: 6, seconds: 1.160, TFLOP/s/device: 46.852, Tokens/s/device: 7062.014, total_weights: 262144, loss: 9.666
I0419 05:10:51.170619 135306851862336 metric_logger.py:185] completed step: 7, seconds: 38.009, TFLOP/s/device: 1.430, Tokens/s/device: 215.525, total_weights: 262144, loss: 9.560
I0419 05:10:51.175282    2410 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0419 05:10:51.749269 135306851862336 metric_logger.py:185] completed step: 8, seconds: 0.585, TFLOP/s/device: 92.950, Tokens/s/device: 14010.364, total_weights: 262144, loss: 9.495
I0419 05:10:52.329222 135306851862336 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0419 05:10:52.329830 135306851862336 checkpointing.py:798] Waited 0.0006198883056640625 seconds for step 9 to finish before starting checkpointing.
I0419 05:10:52.331969 135306851862336 checkpoint_manager.py:2024] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 05:10:53.857568 135184540223232 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/0/items/array_metadatas/process_3
I0419 05:11:25.125383 135184531830528 base_pytree_checkpoint_handler.py:129] [process=3] /jax/checkpoint/write/gbytes_per_sec: 20.420 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 77.35377407073975 s) (per-host)
I0419 05:11:25.125487 135184531830528 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 76.851924s.
I0419 05:11:34.288771 135184531830528 async_checkpointer.py:144] [process=3][thread=async_save] Background save thread done. Time taken: 86.015191s.
I0419 05:11:34.289072 135181335783168 async_checkpointer.py:273] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 05:11:34.289198 135181335783168 async_checkpointer.py:283] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0419 05:11:34.289271 135181335783168 checkpoint_manager.py:2133] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0419 05:11:34.291028 135181335783168 checkpoint_manager.py:2142] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0419 05:11:34.291206 135306851862336 checkpoint_manager.py:2036] [process=3][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0419 05:11:34.291347 135306851862336 checkpoint_manager.py:1458] Waiting for previous save to complete took 41.959386 seconds. If this number is high, consider checkpointing less frequently.
I0419 05:11:34.293174 135306851862336 checkpoint_manager.py:1518] [process=3] Saving checkpoint at step 9
I0419 05:11:34.296566 135306851862336 async_checkpointer.py:452] [process=3] Started async saving checkpoint to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/9.
I0419 05:11:35.273668 135306851862336 jax_array_handlers.py:358] Scheduling D2H of 69 prioritized jax.Array.
I0419 05:11:35.273829 135306851862336 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
I0419 05:11:35.317132 135306851862336 base_pytree_checkpoint_handler.py:153] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.045525s
I0419 05:11:35.317309 135306851862336 base_pytree_checkpoint_handler.py:129] [process=3] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.021 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.30723071098327637 s) (per-host)
I0419 05:11:35.317359 135306851862336 base_pytree_checkpoint_handler.py:737] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.307290s (batch_requests_ready=0.257160s, total_serialization_initiated=0.050064s, others=0.000066s)
I0419 05:11:35.317453 135306851862336 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.311324s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.311307)
I0419 05:11:35.318302 135177534134016 async_checkpointer.py:79] [process=3][thread=async_save] Background save thread started.
I0419 05:11:35.318428 135306851862336 async_checkpointer.py:561] Finished blocking save. Time taken: 1.025183s. Continuing background save to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/9.
I0419 05:11:35.764759 135306851862336 checkpoint_manager.py:1566] [process=3][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 05:11:35.765148 135181335783168 async_checkpointer.py:265] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0419 05:11:35.765300 135306851862336 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776575452.331939, 'wait_for_prev_duration_secs': 41.95938563346863, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776575494.293216, 'checkpointer_blocking_duration_secs': 1.0253853797912598, 'get_old_steps_start_time': 1776575495.3186197, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1776575452.3300483, 'checkpoint_manager_blocking_duration_secs': 43.43521690368652}
I0419 05:11:35.765472 135306851862336 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0419 05:11:35.765519 135306851862336 checkpoint_manager.py:2024] [process=3][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 05:11:42.904599 135184540223232 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260419_025828/nnx_xpk_feat_nnx_set_defaults_true_20260419_025828_06_grad_accum/checkpoints/9/items/array_metadatas/process_3
I0419 05:12:18.567153 135177534134016 base_pytree_checkpoint_handler.py:129] [process=3] /jax/checkpoint/write/gbytes_per_sec: 36.265 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.55703258514404 s) (per-host)
I0419 05:12:18.567290 135177534134016 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.248778s.
I0419 05:12:24.978952 135177534134016 async_checkpointer.py:144] [process=3][thread=async_save] Background save thread done. Time taken: 49.660427s.
I0419 05:12:24.979180 135181335783168 async_checkpointer.py:273] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 05:12:24.979268 135181335783168 async_checkpointer.py:283] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0419 05:12:24.979315 135181335783168 checkpoint_manager.py:2133] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0419 05:12:24.981098 135181335783168 checkpoint_manager.py:2142] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0419 05:12:24.981281 135306851862336 checkpoint_manager.py:2036] [process=3][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0419 05:12:24.981444 135306851862336 checkpoint_manager.py:2013] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 05:12:24.982064 135306851862336 metric_logger.py:185] completed step: 9, seconds: 0.587, TFLOP/s/device: 92.611, Tokens/s/device: 13959.345, total_weights: 262144, loss: 9.457
Per train step:
 Total TFLOPs: 54.35 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Sun Apr 19 05:12:37 UTC 2026
EXIT_CODE=0