MaxView

‹ —Case: 01_base02_synthetic ›

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

MetricLinen  e27fc1e97NNX  e27fc1e97Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s92.16492.517+0.353
Tok/s13891.913945.2+53.255
Avg s/step2.6202.405-0.215
Memory %1.381.380
JAX0.8.10.8.1

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

Linen  ·  e27fc1e97  ·  feat_nnx_set_defaults_true_20260420_190404  ·  full log
XPK Start: Mon Apr 20 19:11:25 UTC 2026
2026-04-20 19:11:29.800465: 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 00:00:1776712289.813022      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776712289.816914      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776712289.828058      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776712289.828081      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776712289.828084      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776712289.828086      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-20 19:12:04.111430: 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)
I0420 19:12:04.635776 137584981997376 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-20 19:12:13,675:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0420 19:12:13.675369 137584981997376 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-20 19:12:13,677:jax._src.distributed:157: Connecting to JAX distributed service on mt-01-base-wckj3-slice-job-0-0.mt-01-base-wckj3:8482
I0420 19:12:13.677783 137584981997376 distributed.py:157] Connecting to JAX distributed service on mt-01-base-wckj3-slice-job-0-0.mt-01-base-wckj3:8482
I0420 19:12:16.966220 137584981997376 max_utils.py:284] Jax distributed system initialized!
I0420 19:12:23.999302 137584981997376 max_utils.py:800] System Information: Jax Version: 0.8.1
I0420 19:12:23.999410 137584981997376 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0420 19:12:23.999451 137584981997376 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
I0420 19:12:23.999485 137584981997376 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 19:12:24.629558 137584981997376 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 19:12:24.630171 137584981997376 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 19:12:24.630356 137584981997376 checkpointing.py:688] Setting up checkpoint logger...
I0420 19:12:24.630405 137584981997376 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 19:12:24.630449 137584981997376 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0420 19:12:24.630802 137584981997376 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 0x7d21769792b0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 19:12:28.149949 137584981997376 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 19:12:28.150208 137584981997376 checkpoint_manager.py:709] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d0d1456d8b0>}, handler_registry=None
I0420 19:12:28.150445 137584981997376 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d0d1456d8b0>` 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`.
I0420 19:12:28.150493 137584981997376 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d0d14572840>` 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`.
I0420 19:12:28.150529 137584981997376 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 0x7d0d1456d8b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d0d1456d8b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d0d14572840>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d0d14572840>}).
I0420 19:12:28.151057 137584981997376 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0420 19:12:28.151137 137584981997376 async_checkpointer.py:177] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d0cfc32c180> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0420 19:12:29.404478 137584981997376 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints
I0420 19:12:29.449644 137584981997376 checkpoint_manager.py:929] [process=7][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_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d0d146d30b0>
I0420 19:12:29.449789 137584981997376 checkpointing.py:302] Checkpoint manager created!
I0420 19:12:30.346245 137584981997376 nnx_wrappers.py:455] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 19:12:30.346347 137584981997376 nnx_wrappers.py:455] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 19:12:30.724965 137584981997376 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 19:12:30.725070 137584981997376 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 19:12:30.740977 137584981997376 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 19:12:30.741036 137584981997376 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 19:12:30.763773 137584981997376 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 19:12:30.763840 137584981997376 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 19:12:30.779728 137584981997376 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 19:12:30.779792 137584981997376 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 19:12:30.795457 137584981997376 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 19:12:30.795522 137584981997376 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 19:12:30.820237 137584981997376 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 19:12:30.820306 137584981997376 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 19:12:30.839723 137584981997376 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 19:12:30.839787 137584981997376 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 19:12:31.033786 137584981997376 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 19:12:31.033905 137584981997376 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 19:12:32.460570 137584981997376 maxtext_utils.py:1740]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   PartitionSpec('norm',)
    Physical:  (None,)
I0420 19:12:32.460708 137584981997376 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)
I0420 19:12:32.460768 137584981997376 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)
I0420 19:12:32.460831 137584981997376 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')
I0420 19:12:32.460887 137584981997376 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)
I0420 19:12:32.460927 137584981997376 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)
I0420 19:12:32.460983 137584981997376 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)
I0420 19:12:32.461039 137584981997376 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')
I0420 19:12:32.461083 137584981997376 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)
I0420 19:12:32.461122 137584981997376 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)
I0420 19:12:32.461199 137584981997376 maxtext_utils.py:1740]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   PartitionSpec('embed', 'vocab')
    Physical:  ('fsdp', None)
I0420 19:12:32.461256 137584981997376 maxtext_utils.py:1740]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   PartitionSpec('vocab', 'embed')
    Physical:  (None, 'fsdp')

I0420 19:12:32.959867 137584981997376 train.py:158] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 19:12:32.959961 137584981997376 train.py:158] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 19:12:32.974756 137584981997376 train.py:165] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 19:12:32.974819 137584981997376 train.py:165] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 19:12:43.713973 137584981997376 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.
I0420 19:12:43.714859 137584981997376 metric_logger.py:289] number parameters: 1.104 billion
I0420 19:12:54.914634 137584981997376 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 19:12:55.072338 137584981997376 checkpointing.py:798] Waited 0.1576850414276123 seconds for step 0 to finish before starting checkpointing.
I0420 19:12:55.074975 137584981997376 checkpoint_manager.py:2013] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 19:12:55.076631 137584981997376 checkpoint_manager.py:1518] [process=7] Saving checkpoint at step 0
I0420 19:12:55.079677 137584981997376 async_checkpointer.py:452] [process=7] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/0.
I0420 19:12:56.103449 137584981997376 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 19:12:56.104483 137584981997376 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0420 19:12:56.104538 137584981997376 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
I0420 19:12:56.353411 137584981997376 base_pytree_checkpoint_handler.py:153] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.250042s
I0420 19:12:56.353587 137584981997376 base_pytree_checkpoint_handler.py:129] [process=7] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.862 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5389392375946045 s) (per-host)
I0420 19:12:56.353641 137584981997376 base_pytree_checkpoint_handler.py:737] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.539004s (batch_requests_ready=0.285319s, total_serialization_initiated=0.253611s, others=0.000074s)
I0420 19:12:56.353732 137584981997376 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.543098s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.543079)
I0420 19:12:56.354660 137462156351232 async_checkpointer.py:79] [process=7][thread=async_save] Background save thread started.
I0420 19:12:56.354830 137584981997376 async_checkpointer.py:561] Finished blocking save. Time taken: 1.278132s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/0.
I0420 19:12:56.363374 137584981997376 checkpoint_manager.py:1566] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 19:12:56.363671 137458373076736 async_checkpointer.py:265] [process=7][thread=save_finalize] Waiting for background save thread=async_save.
I0420 19:12:56.363826 137584981997376 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776712375.0749443, 'wait_for_prev_duration_secs': 0.00012421607971191406, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776712375.0766685, 'checkpointer_blocking_duration_secs': 1.2782671451568604, 'get_old_steps_start_time': 1776712376.3549554, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1776712375.0728512, 'checkpoint_manager_blocking_duration_secs': 1.2909355163574219}
I0420 19:12:56.363928 137584981997376 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 19:12:56.363980 137584981997376 max_utils.py:750] 
Memstats: After params initialized:
I0420 19:12:56.364030 137584981997376 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_26(process=7,(2,6,0,0))
I0420 19:12:56.364064 137584981997376 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0))
I0420 19:12:56.364092 137584981997376 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0))
I0420 19:12:56.364118 137584981997376 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0))
I0420 19:12:56.630865 137584981997376 metric_logger.py:185] completed step: 0, seconds: 11.200, TFLOP/s/device: 1.213, Tokens/s/device: 182.863, total_weights: 65536, loss: 10.874
I0420 19:12:56.806803 137584981997376 metric_logger.py:185] completed step: 1, seconds: 1.715, TFLOP/s/device: 7.923, Tokens/s/device: 1194.187, total_weights: 65536, loss: 10.874
I0420 19:12:57.193350 137584981997376 metric_logger.py:185] completed step: 2, seconds: 0.028, TFLOP/s/device: 477.160, Tokens/s/device: 71922.739, total_weights: 65536, loss: 10.263
I0420 19:12:57.340419 137584981997376 metric_logger.py:185] completed step: 3, seconds: 0.387, TFLOP/s/device: 35.132, Tokens/s/device: 5295.438, total_weights: 65536, loss: 9.734
I0420 19:12:57.635380 137584981997376 metric_logger.py:185] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.787, Tokens/s/device: 13382.909, total_weights: 65536, loss: 9.278
I0420 19:12:57.641213 137584981997376 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.367, Tokens/s/device: 13922.502, total_weights: 65536, loss: 8.892
I0420 19:13:18.344261 137584981997376 metric_logger.py:185] completed step: 6, seconds: 0.295, TFLOP/s/device: 45.980, Tokens/s/device: 6930.673, total_weights: 65536, loss: 8.592
I0420 19:13:18.491720 137584981997376 metric_logger.py:185] completed step: 7, seconds: 20.556, TFLOP/s/device: 0.661, Tokens/s/device: 99.629, total_weights: 65536, loss: 8.384
I0420 19:13:18.639048 137584981997376 metric_logger.py:185] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.347, Tokens/s/device: 13467.394, total_weights: 65536, loss: 8.255
I0420 19:13:18.786094 137584981997376 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 19:13:18.786838 137584981997376 checkpointing.py:798] Waited 0.0007631778717041016 seconds for step 9 to finish before starting checkpointing.
I0420 19:13:18.788848 137584981997376 checkpoint_manager.py:2024] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 19:13:19.462030    2644 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 19:13:21.388205 137462164743936 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/0/items/array_metadatas/process_7
I0420 19:13:53.431050 137462156351232 base_pytree_checkpoint_handler.py:129] [process=7] /jax/checkpoint/write/gbytes_per_sec: 27.416 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 57.61635756492615 s) (per-host)
I0420 19:13:53.431211 137462156351232 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 57.076434s.
I0420 19:14:01.634038 137462156351232 async_checkpointer.py:144] [process=7][thread=async_save] Background save thread done. Time taken: 65.279249s.
I0420 19:14:01.634361 137458373076736 async_checkpointer.py:273] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 19:14:01.634488 137458373076736 async_checkpointer.py:283] [process=7][thread=save_finalize] No errors found in background save thread=async_save.
I0420 19:14:01.634541 137458373076736 checkpoint_manager.py:2133] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 19:14:01.635941 137458373076736 checkpoint_manager.py:2142] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 19:14:01.636116 137584981997376 checkpoint_manager.py:2036] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 19:14:01.636276 137584981997376 checkpoint_manager.py:1458] Waiting for previous save to complete took 42.847430 seconds. If this number is high, consider checkpointing less frequently.
I0420 19:14:01.638058 137584981997376 checkpoint_manager.py:1518] [process=7] Saving checkpoint at step 9
I0420 19:14:01.641654 137584981997376 async_checkpointer.py:452] [process=7] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/9.
I0420 19:14:02.587558 137584981997376 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array.
I0420 19:14:02.587653 137584981997376 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
I0420 19:14:02.630126 137584981997376 base_pytree_checkpoint_handler.py:153] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.043770s
I0420 19:14:02.630323 137584981997376 base_pytree_checkpoint_handler.py:129] [process=7] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.181 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.29770851135253906 s) (per-host)
I0420 19:14:02.630388 137584981997376 base_pytree_checkpoint_handler.py:737] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.297778s (batch_requests_ready=0.252076s, total_serialization_initiated=0.045621s, others=0.000081s)
I0420 19:14:02.630503 137584981997376 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.301906s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.301891)
I0420 19:14:02.631338 137459440236288 async_checkpointer.py:79] [process=7][thread=async_save] Background save thread started.
I0420 19:14:02.631491 137584981997376 async_checkpointer.py:561] Finished blocking save. Time taken: 0.993367s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/9.
I0420 19:14:02.676563 137584981997376 checkpoint_manager.py:1566] [process=7][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 19:14:02.676877 137458373076736 async_checkpointer.py:265] [process=7][thread=save_finalize] Waiting for background save thread=async_save.
I0420 19:14:02.677053 137584981997376 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776712398.7888193, 'wait_for_prev_duration_secs': 42.84743046760559, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776712441.6380992, 'checkpointer_blocking_duration_secs': 0.9934670925140381, 'get_old_steps_start_time': 1776712442.6315892, 'get_old_steps_duration_secs': 2.5510787963867188e-05, 'checkpoint_manager_blocking_start_time': 1776712398.787069, 'checkpoint_manager_blocking_duration_secs': 43.889947175979614}
I0420 19:14:02.677195 137584981997376 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 19:14:02.677251 137584981997376 checkpoint_manager.py:2024] [process=7][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 19:14:07.799698 137462164743936 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/linen_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/9/items/array_metadatas/process_7
I0420 19:14:44.944954 137459440236288 base_pytree_checkpoint_handler.py:129] [process=7] /jax/checkpoint/write/gbytes_per_sec: 37.069 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.61230731010437 s) (per-host)
I0420 19:14:44.945058 137459440236288 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.313610s.
I0420 19:14:52.310953 137459440236288 async_checkpointer.py:144] [process=7][thread=async_save] Background save thread done. Time taken: 49.679490s.
I0420 19:14:52.311249 137458373076736 async_checkpointer.py:273] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 19:14:52.311369 137458373076736 async_checkpointer.py:283] [process=7][thread=save_finalize] No errors found in background save thread=async_save.
I0420 19:14:52.311417 137458373076736 checkpoint_manager.py:2133] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 19:14:52.312876 137458373076736 checkpoint_manager.py:2142] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 19:14:52.313049 137584981997376 checkpoint_manager.py:2036] [process=7][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0420 19:14:52.313210 137584981997376 checkpoint_manager.py:2013] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 19:14:52.313823 137584981997376 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.164, Tokens/s/device: 13891.904, 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: Mon Apr 20 19:15:05 UTC 2026
EXIT_CODE=0
NNX  ·  e27fc1e97  ·  feat_nnx_set_defaults_true_20260420_190404  ·  full log
XPK Start: Mon Apr 20 20:55:34 UTC 2026
2026-04-20 20:55:38.702960: 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 00:00:1776718538.715505      10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1776718538.719217      10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
W0000 00:00:1776718538.730385      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776718538.730405      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776718538.730408      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
W0000 00:00:1776718538.730410      10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once.
2026-04-20 20:55:57.887249: 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)
I0420 20:55:58.407172 136718864877376 max_utils.py:273] Attempting to initialize the jax distributed system...
INFO:2026-04-20 20:56:07,447:jax._src.distributed:140: Starting JAX distributed service on [::]:8482
I0420 20:56:07.447001 136718864877376 distributed.py:140] Starting JAX distributed service on [::]:8482
INFO:2026-04-20 20:56:07,449:jax._src.distributed:157: Connecting to JAX distributed service on mt-01-base-kmqwj-slice-job-0-0.mt-01-base-kmqwj:8482
I0420 20:56:07.449248 136718864877376 distributed.py:157] Connecting to JAX distributed service on mt-01-base-kmqwj-slice-job-0-0.mt-01-base-kmqwj:8482
I0420 20:56:08.584369 136718864877376 max_utils.py:284] Jax distributed system initialized!
I0420 20:56:15.627289 136718864877376 max_utils.py:800] System Information: Jax Version: 0.8.1
I0420 20:56:15.627399 136718864877376 max_utils.py:801] System Information: Jaxlib Version: 0.8.1
I0420 20:56:15.627441 136718864877376 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
I0420 20:56:15.627478 136718864877376 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 20:56:16.242706 136718864877376 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 20:56:16.356496 136718864877376 checkpointing.py:688] Setting up checkpoint logger...
I0420 20:56:16.356614 136718864877376 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 20:56:16.356676 136718864877376 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None
I0420 20:56:16.356886 136718864877376 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 0x7c57d87fa210>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 20:56:20.247223 136718864877376 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 20:56:20.247461 136718864877376 checkpoint_manager.py:709] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c43cc675a90>}, handler_registry=None
I0420 20:56:20.247713 136718864877376 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c43cc675a90>` 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`.
I0420 20:56:20.247764 136718864877376 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c43cc3b2cc0>` 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`.
I0420 20:56:20.247801 136718864877376 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 0x7c43cc675a90>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c43cc675a90>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c43cc3b2cc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c43cc3b2cc0>}).
I0420 20:56:20.248164 136718864877376 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33
I0420 20:56:20.248244 136718864877376 async_checkpointer.py:177] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c43cc350f40> timeout: 600 secs and primary_host=0 for async checkpoint writes
I0420 20:56:21.189648 136718864877376 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints
I0420 20:56:21.602305 136718864877376 checkpoint_manager.py:929] [process=7][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_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c43cc66d970>
I0420 20:56:21.602485 136718864877376 checkpointing.py:302] Checkpoint manager created!
I0420 20:56:22.528300 136718864877376 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 20:56:22.528422 136718864877376 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 20:56:23.974533 136718864877376 maxtext_utils.py:1749]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0420 20:56:23.974664 136718864877376 maxtext_utils.py:1749]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0420 20:56:23.974723 136718864877376 maxtext_utils.py:1749]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0420 20:56:23.974770 136718864877376 maxtext_utils.py:1749]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0420 20:56:23.974810 136718864877376 maxtext_utils.py:1749]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0420 20:56:23.974848 136718864877376 maxtext_utils.py:1749]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0420 20:56:23.974886 136718864877376 maxtext_utils.py:1749]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 20:56:23.974924 136718864877376 maxtext_utils.py:1749]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0420 20:56:23.974961 136718864877376 maxtext_utils.py:1749]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 20:56:23.974997 136718864877376 maxtext_utils.py:1749]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 20:56:23.975032 136718864877376 maxtext_utils.py:1749]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0420 20:56:23.975067 136718864877376 maxtext_utils.py:1749]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')

I0420 20:56:24.372810 136718864877376 nnx_decoders.py:616] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 20:56:24.372961 136718864877376 nnx_decoders.py:616] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 20:56:24.378346 136718864877376 nnx_decoders.py:616] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 20:56:24.378441 136718864877376 nnx_decoders.py:616] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 20:56:24.519244 136718864877376 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 20:56:24.519375 136718864877376 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 20:56:24.657686 136718864877376 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 20:56:24.657817 136718864877376 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 20:56:24.809453 136718864877376 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 20:56:24.809587 136718864877376 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 20:56:24.951459 136718864877376 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 20:56:24.951592 136718864877376 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 20:56:25.090236 136718864877376 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 20:56:25.090372 136718864877376 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 20:56:25.243637 136718864877376 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 20:56:25.243794 136718864877376 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 20:56:25.266377 136718864877376 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 20:56:25.266474 136718864877376 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 20:56:36.345876 136718864877376 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.
I0420 20:56:36.348253 136718864877376 metric_logger.py:289] number parameters: 1.104 billion
I0420 20:56:47.451182 136718864877376 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 20:56:47.603321 136718864877376 checkpointing.py:798] Waited 0.15212631225585938 seconds for step 0 to finish before starting checkpointing.
I0420 20:56:47.605636 136718864877376 checkpoint_manager.py:2013] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 20:56:47.607398 136718864877376 checkpoint_manager.py:1518] [process=7] Saving checkpoint at step 0
I0420 20:56:47.610341 136718864877376 async_checkpointer.py:452] [process=7] Started async saving checkpoint to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/0.
I0420 20:56:48.217431 136718864877376 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 20:56:48.218491 136718864877376 jax_array_handlers.py:358] Scheduling D2H of 69 prioritized jax.Array.
I0420 20:56:48.218553 136718864877376 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
I0420 20:56:48.474649 136718864877376 base_pytree_checkpoint_handler.py:153] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.258360s
I0420 20:56:48.474841 136718864877376 base_pytree_checkpoint_handler.py:129] [process=7] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.833 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5445418357849121 s) (per-host)
I0420 20:56:48.474891 136718864877376 base_pytree_checkpoint_handler.py:737] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.544601s (batch_requests_ready=0.279647s, total_serialization_initiated=0.264887s, others=0.000067s)
I0420 20:56:48.474985 136718864877376 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.548591s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.548572)
I0420 20:56:48.475943 136594472277760 async_checkpointer.py:79] [process=7][thread=async_save] Background save thread started.
I0420 20:56:48.476129 136718864877376 async_checkpointer.py:561] Finished blocking save. Time taken: 0.868660s. Continuing background save to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/0.
I0420 20:56:48.492628 136718864877376 checkpoint_manager.py:1566] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 20:56:48.492911 136592886843136 async_checkpointer.py:265] [process=7][thread=save_finalize] Waiting for background save thread=async_save.
I0420 20:56:48.493038 136718864877376 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776718607.6056182, 'wait_for_prev_duration_secs': 8.416175842285156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776718607.6074386, 'checkpointer_blocking_duration_secs': 0.8688051700592041, 'get_old_steps_start_time': 1776718608.4762642, 'get_old_steps_duration_secs': 2.47955322265625e-05, 'checkpoint_manager_blocking_start_time': 1776718607.6037364, 'checkpoint_manager_blocking_duration_secs': 0.8892636299133301}
I0420 20:56:48.493146 136718864877376 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 20:56:48.493224 136718864877376 max_utils.py:750] 
Memstats: After params initialized:
I0420 20:56:48.493277 136718864877376 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_26(process=7,(2,6,0,0))
I0420 20:56:48.493310 136718864877376 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0))
I0420 20:56:48.493338 136718864877376 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0))
I0420 20:56:48.493364 136718864877376 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0))
I0420 20:56:48.765954 136718864877376 metric_logger.py:185] completed step: 0, seconds: 11.102, TFLOP/s/device: 1.224, Tokens/s/device: 184.473, total_weights: 65536, loss: 10.874
I0420 20:56:48.931676 136718864877376 metric_logger.py:185] completed step: 1, seconds: 1.313, TFLOP/s/device: 10.347, Tokens/s/device: 1559.663, total_weights: 65536, loss: 10.874
I0420 20:56:49.762109 136718864877376 metric_logger.py:185] completed step: 2, seconds: 0.020, TFLOP/s/device: 686.774, Tokens/s/device: 103517.994, total_weights: 65536, loss: 10.262
I0420 20:56:49.909461 136718864877376 metric_logger.py:185] completed step: 3, seconds: 0.822, TFLOP/s/device: 16.539, Tokens/s/device: 2492.867, total_weights: 65536, loss: 9.732
I0420 20:56:50.205112 136718864877376 metric_logger.py:185] completed step: 4, seconds: 0.163, TFLOP/s/device: 83.517, Tokens/s/device: 12588.590, total_weights: 65536, loss: 9.274
I0420 20:56:50.212069 136718864877376 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.453, Tokens/s/device: 13935.575, total_weights: 65536, loss: 8.888
I0420 20:57:08.943250 136718864877376 metric_logger.py:185] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.930, Tokens/s/device: 6923.012, total_weights: 65536, loss: 8.589
I0420 20:57:09.090548 136718864877376 metric_logger.py:185] completed step: 7, seconds: 18.585, TFLOP/s/device: 0.731, Tokens/s/device: 110.196, total_weights: 65536, loss: 8.380
I0420 20:57:09.238149 136718864877376 metric_logger.py:185] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.757, Tokens/s/device: 13378.363, total_weights: 65536, loss: 8.251
I0420 20:57:09.386663 136718864877376 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 20:57:09.387261 136718864877376 checkpointing.py:798] Waited 0.0006213188171386719 seconds for step 9 to finish before starting checkpointing.
I0420 20:57:09.389369 136718864877376 checkpoint_manager.py:2024] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 20:57:10.248726    2604 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 20:57:12.532080 136594480670464 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/0/items/array_metadatas/process_7
I0420 20:57:43.856866 136594472277760 base_pytree_checkpoint_handler.py:129] [process=7] /jax/checkpoint/write/gbytes_per_sec: 28.244 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 55.92653441429138 s) (per-host)
I0420 20:57:43.856971 136594472277760 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 55.380905s.
I0420 20:57:52.066171 136594472277760 async_checkpointer.py:144] [process=7][thread=async_save] Background save thread done. Time taken: 63.590091s.
I0420 20:57:52.066467 136592886843136 async_checkpointer.py:273] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 20:57:52.066601 136592886843136 async_checkpointer.py:283] [process=7][thread=save_finalize] No errors found in background save thread=async_save.
I0420 20:57:52.066668 136592886843136 checkpoint_manager.py:2133] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 20:57:52.068015 136592886843136 checkpoint_manager.py:2142] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 20:57:52.068210 136718864877376 checkpoint_manager.py:2036] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 20:57:52.068344 136718864877376 checkpoint_manager.py:1458] Waiting for previous save to complete took 42.678980 seconds. If this number is high, consider checkpointing less frequently.
I0420 20:57:52.070305 136718864877376 checkpoint_manager.py:1518] [process=7] Saving checkpoint at step 9
I0420 20:57:52.073834 136718864877376 async_checkpointer.py:452] [process=7] Started async saving checkpoint to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/9.
I0420 20:57:52.606187 136718864877376 jax_array_handlers.py:358] Scheduling D2H of 69 prioritized jax.Array.
I0420 20:57:52.606348 136718864877376 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
I0420 20:57:52.652849 136718864877376 base_pytree_checkpoint_handler.py:153] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.048769s
I0420 20:57:52.653017 136718864877376 base_pytree_checkpoint_handler.py:129] [process=7] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.032 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.3065502643585205 s) (per-host)
I0420 20:57:52.653077 136718864877376 base_pytree_checkpoint_handler.py:737] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.306619s (batch_requests_ready=0.253126s, total_serialization_initiated=0.053416s, others=0.000077s)
I0420 20:57:52.653171 136718864877376 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.310614s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.310599)
I0420 20:57:52.654126 136593947649792 async_checkpointer.py:79] [process=7][thread=async_save] Background save thread started.
I0420 20:57:52.654302 136718864877376 async_checkpointer.py:561] Finished blocking save. Time taken: 0.583921s. Continuing background save to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/9.
I0420 20:57:52.665990 136718864877376 checkpoint_manager.py:1566] [process=7][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 20:57:52.666292 136592886843136 async_checkpointer.py:265] [process=7][thread=save_finalize] Waiting for background save thread=async_save.
I0420 20:57:52.666454 136718864877376 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776718629.3893404, 'wait_for_prev_duration_secs': 42.67897963523865, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776718672.0703454, 'checkpointer_blocking_duration_secs': 0.5840644836425781, 'get_old_steps_start_time': 1776718672.6544304, 'get_old_steps_duration_secs': 2.4318695068359375e-05, 'checkpoint_manager_blocking_start_time': 1776718629.38748, 'checkpoint_manager_blocking_duration_secs': 43.27893900871277}
I0420 20:57:52.666562 136718864877376 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 20:57:52.666606 136718864877376 checkpoint_manager.py:2024] [process=7][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 20:57:57.437852 136594480670464 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_set_defaults_true_20260420_190404/nnx_xpk_feat_nnx_set_defaults_true_20260420_190404_01_base/checkpoints/9/items/array_metadatas/process_7
I0420 20:58:34.050766 136593947649792 base_pytree_checkpoint_handler.py:129] [process=7] /jax/checkpoint/write/gbytes_per_sec: 37.876 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.70427870750427 s) (per-host)
I0420 20:58:34.050871 136593947649792 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.396628s.
I0420 20:58:42.265935 136593947649792 async_checkpointer.py:144] [process=7][thread=async_save] Background save thread done. Time taken: 49.611676s.
I0420 20:58:42.266248 136592886843136 async_checkpointer.py:273] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 20:58:42.266376 136592886843136 async_checkpointer.py:283] [process=7][thread=save_finalize] No errors found in background save thread=async_save.
I0420 20:58:42.266424 136592886843136 checkpoint_manager.py:2133] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 20:58:42.267890 136592886843136 checkpoint_manager.py:2142] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 20:58:42.268077 136718864877376 checkpoint_manager.py:2036] [process=7][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0420 20:58:42.268255 136718864877376 checkpoint_manager.py:2013] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 20:58:42.268995 136718864877376 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.517, Tokens/s/device: 13945.159, 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: Mon Apr 20 20:58:54 UTC 2026
EXIT_CODE=0