MaxView

‹ 01_baseCase: 02_synthetic03_dropout ›

Metrics: Linen vs NNX  ·  feat/nnx-post-train-fixes

MetricLinen  2fbc8c2f1NNX  2fbc8c2f1Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s92.16791.963-0.204
Tok/s13892.513861.6-30.842
Avg s/step2.9062.802-0.104
Memory %1.381.380
JAX0.9.20.9.2

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

Linen  ·  2fbc8c2f1  ·  feat_nnx_post_train_fixes_20260417_023710  ·  full log
XPK Start: Fri Apr 17 02:41:54 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-17 02:42:18.457817: 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)
I0417 02:42:18.633515 136950310844224 max_utils.py:273] Attempting to initialize the jax distributed system...
I0417 02:42:27.674007 136950310844224 distributed.py:149] Starting JAX distributed service on [::]:8482
I0417 02:42:27.676299 136950310844224 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-ssz2z-slice-job-0-0.mt-02-synthetic-ssz2z:8482
I0417 02:42:28.911741 136950310844224 max_utils.py:284] Jax distributed system initialized!
I0417 02:42:35.215311 136950310844224 max_utils.py:800] System Information: Jax Version: 0.9.2
I0417 02:42:35.215414 136950310844224 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0417 02:42:35.215454 136950310844224 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0417 02:42:35.215487 136950310844224 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0417 02:42:35.911715 136950310844224 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 02:42:35.912303 136950310844224 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 02:42:35.912481 136950310844224 checkpointing.py:688] Setting up checkpoint logger...
I0417 02:42:35.912531 136950310844224 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0417 02:42:35.912579 136950310844224 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0417 02:42:35.912918 136950310844224 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7c8d99d49f10>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0417 02:42:38.840716 136950310844224 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0417 02:42:38.840974 136950310844224 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c79383f8770>}, handler_registry=None
I0417 02:42:38.841209 136950310844224 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c79383f8770>` 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`.
I0417 02:42:38.841258 136950310844224 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c79383f94f0>` 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`.
I0417 02:42:38.841292 136950310844224 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 0x7c79383f8770>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c79383f8770>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c79383f94f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c79383f94f0>}).
I0417 02:42:38.841665 136950310844224 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0417 02:42:38.841738 136950310844224 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c79187ec7c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0417 02:42:41.228450 136950310844224 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints
I0417 02:42:41.230720 136950310844224 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_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c79383fa210>
I0417 02:42:41.230854 136950310844224 checkpointing.py:302] Checkpoint manager created!
I0417 02:42:42.162304 136950310844224 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0417 02:42:42.162406 136950310844224 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 02:42:42.545507 136950310844224 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 02:42:42.545595 136950310844224 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 02:42:42.561870 136950310844224 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 02:42:42.561927 136950310844224 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 02:42:42.585494 136950310844224 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 02:42:42.585561 136950310844224 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 02:42:42.601871 136950310844224 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 02:42:42.601932 136950310844224 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 02:42:42.618306 136950310844224 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 02:42:42.618367 136950310844224 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 02:42:42.643138 136950310844224 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0417 02:42:42.643207 136950310844224 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 02:42:42.664617 136950310844224 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0417 02:42:42.664679 136950310844224 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0417 02:42:42.875921 136950310844224 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0417 02:42:42.876029 136950310844224 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0417 02:42:44.289425 136950310844224 maxtext_utils.py:1796]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0417 02:42:44.289559 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 02:42:44.289611 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 02:42:44.289668 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0417 02:42:44.289719 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 02:42:44.289757 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 02:42:44.289808 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0417 02:42:44.289882 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0417 02:42:44.289926 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0417 02:42:44.289963 136950310844224 maxtext_utils.py:1796]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0417 02:42:44.290013 136950310844224 maxtext_utils.py:1796]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0417 02:42:44.290064 136950310844224 maxtext_utils.py:1796]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0417 02:42:44.776075 136950310844224 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 02:42:44.776171 136950310844224 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 02:42:44.791888 136950310844224 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 02:42:44.791952 136950310844224 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 02:42:55.775884 136950310844224 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.
I0417 02:42:55.776743 136950310844224 metric_logger.py:289] number parameters: 1.104 billion
I0417 02:43:07.002786 136950310844224 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0417 02:43:07.159849 136950310844224 checkpointing.py:798] Waited 0.15703177452087402 seconds for step 0 to finish before starting checkpointing.
I0417 02:43:07.163487 136950310844224 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 02:43:07.165246 136950310844224 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0417 02:43:07.166594 136950310844224 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/0.
I0417 02:43:07.783733 136950310844224 signaling_client.py:364] Using JaxDistributedSignalingClient
I0417 02:43:07.784759 136950310844224 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 02:43:07.784816 136950310844224 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
I0417 02:43:08.059220 136950310844224 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.275562s
I0417 02:43:08.059386 136950310844224 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.493 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2808213233947754 s) (per-host)
I0417 02:43:08.059437 136950310844224 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.280882s (batch_requests_ready=0.002137s, total_serialization_initiated=0.278676s, others=0.000069s)
I0417 02:43:08.059529 136950310844224 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.285154s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.285137)
I0417 02:43:08.060315 136950310844224 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.89 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/0.
I0417 02:43:08.060649 136820998264576 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 03:03:08.060613
I0417 02:43:08.075391 136950310844224 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 02:43:08.075670 136818867566336 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 02:43:08.075849 136950310844224 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776393787.1634688, 'wait_for_prev_duration_secs': 6.198883056640625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776393787.1652846, 'checkpointer_blocking_duration_secs': 0.8955204486846924, 'get_old_steps_start_time': 1776393788.0608482, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776393787.1603749, 'checkpoint_manager_blocking_duration_secs': 0.9154160022735596}
I0417 02:43:08.075955 136950310844224 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0417 02:43:08.076004 136950310844224 max_utils.py:750] 
Memstats: After params initialized:
I0417 02:43:08.076052 136950310844224 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0417 02:43:08.076085 136950310844224 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0417 02:43:08.076111 136950310844224 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0417 02:43:08.076135 136950310844224 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0417 02:43:08.384700 136950310844224 metric_logger.py:185] completed step: 0, seconds: 11.226, TFLOP/s/device: 1.210, Tokens/s/device: 182.435, total_weights: 65536, loss: 10.874
I0417 02:43:08.571570 136950310844224 metric_logger.py:185] completed step: 1, seconds: 1.381, TFLOP/s/device: 9.841, Tokens/s/device: 1483.346, total_weights: 65536, loss: 10.874
I0417 02:43:08.969328 136950310844224 metric_logger.py:185] completed step: 2, seconds: 0.040, TFLOP/s/device: 341.643, Tokens/s/device: 51496.103, total_weights: 65536, loss: 10.262
I0417 02:43:09.116642 136950310844224 metric_logger.py:185] completed step: 3, seconds: 0.398, TFLOP/s/device: 34.148, Tokens/s/device: 5147.087, total_weights: 65536, loss: 9.734
I0417 02:43:09.412089 136950310844224 metric_logger.py:185] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.941, Tokens/s/device: 13406.212, total_weights: 65536, loss: 9.277
I0417 02:43:09.417894 136950310844224 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.230, Tokens/s/device: 13901.899, total_weights: 65536, loss: 8.892
I0417 02:43:12.672709    2822 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0417 02:43:15.635707 136819347691264 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_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/0/items/array_metadatas/process_6
I0417 02:43:33.006459 136950310844224 metric_logger.py:185] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.914, Tokens/s/device: 6920.696, total_weights: 65536, loss: 8.592
I0417 02:43:33.153891 136950310844224 metric_logger.py:185] completed step: 7, seconds: 23.442, TFLOP/s/device: 0.580, Tokens/s/device: 87.365, total_weights: 65536, loss: 8.384
I0417 02:43:33.301036 136950310844224 metric_logger.py:185] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.390, Tokens/s/device: 13473.773, total_weights: 65536, loss: 8.255
I0417 02:43:33.448147 136950310844224 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0417 02:43:33.448807 136950310844224 checkpointing.py:798] Waited 0.0006730556488037109 seconds for step 9 to finish before starting checkpointing.
I0417 02:43:33.450908 136950310844224 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 02:43:44.177575 136820998264576 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 43.397 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.39897179603577 s) (per-host)
I0417 02:43:44.177695 136820998264576 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.116937s.
I0417 02:43:54.819126 136820998264576 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 46.758351s.
I0417 02:43:54.819407 136818867566336 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 02:43:54.819526 136818867566336 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 02:43:54.819577 136818867566336 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0417 02:43:54.821452 136818867566336 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0417 02:43:54.821639 136950310844224 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0417 02:43:54.821773 136950310844224 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.370864 seconds. If this number is high, consider checkpointing less frequently.
I0417 02:43:54.823610 136950310844224 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0417 02:43:54.825587 136950310844224 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/9.
I0417 02:43:55.535737 136950310844224 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 02:43:55.535846 136950310844224 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
I0417 02:43:55.579834 136950310844224 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.045204s
I0417 02:43:55.580001 136950310844224 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 31.737 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04860424995422363 s) (per-host)
I0417 02:43:55.580051 136950310844224 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.048664s (batch_requests_ready=0.001768s, total_serialization_initiated=0.046828s, others=0.000068s)
I0417 02:43:55.580139 136950310844224 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.052957s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.052939)
I0417 02:43:55.580765 136950310844224 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/9.
I0417 02:43:55.581127 136820989871872 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 03:03:55.581089
I0417 02:43:55.583083 136950310844224 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 02:43:55.583389 136818867566336 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 02:43:55.583526 136950310844224 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776393813.4508784, 'wait_for_prev_duration_secs': 21.370864152908325, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776393834.82365, 'checkpointer_blocking_duration_secs': 0.75762939453125, 'get_old_steps_start_time': 1776393835.5813022, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776393813.4490533, 'checkpoint_manager_blocking_duration_secs': 22.134438037872314}
I0417 02:43:55.583701 136950310844224 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0417 02:43:55.583747 136950310844224 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 02:44:02.328886 136819347691264 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_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/9/items/array_metadatas/process_6
I0417 02:44:38.511849 136820989871872 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.752 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.98039650917053 s) (per-host)
I0417 02:44:38.511971 136820989871872 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.930725s.
I0417 02:44:47.577087 136820989871872 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 51.995827s.
I0417 02:44:47.577367 136818867566336 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 02:44:47.577491 136818867566336 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 02:44:47.577537 136818867566336 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0417 02:44:47.579468 136818867566336 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0417 02:44:47.579652 136950310844224 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0417 02:44:47.579837 136950310844224 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 02:44:47.580493 136950310844224 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.167, Tokens/s/device: 13892.469, total_weights: 65536, loss: 8.179
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 17 02:44:59 UTC 2026
EXIT_CODE=0
NNX  ·  2fbc8c2f1  ·  feat_nnx_post_train_fixes_20260417_023710  ·  full log
XPK Start: Fri Apr 17 04:42:58 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-17 04:43:22.426544: 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)
I0417 04:43:22.603658 137978123536192 max_utils.py:273] Attempting to initialize the jax distributed system...
I0417 04:43:31.643570 137978123536192 distributed.py:149] Starting JAX distributed service on [::]:8482
I0417 04:43:31.645928 137978123536192 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-o9iuy-slice-job-0-0.mt-02-synthetic-o9iuy:8482
I0417 04:43:33.103887 137978123536192 max_utils.py:284] Jax distributed system initialized!
I0417 04:43:39.344256 137978123536192 max_utils.py:800] System Information: Jax Version: 0.9.2
I0417 04:43:39.344355 137978123536192 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0417 04:43:39.344394 137978123536192 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0417 04:43:39.344429 137978123536192 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0417 04:43:40.027284 137978123536192 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 04:43:40.139035 137978123536192 checkpointing.py:688] Setting up checkpoint logger...
I0417 04:43:40.139153 137978123536192 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0417 04:43:40.139197 137978123536192 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0417 04:43:40.139393 137978123536192 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7d7cd144c050>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0417 04:43:43.401887 137978123536192 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0417 04:43:43.402121 137978123536192 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d67a4695c70>}, handler_registry=None
I0417 04:43:43.402357 137978123536192 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d67a4695c70>` 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`.
I0417 04:43:43.402403 137978123536192 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d688c31e8a0>` 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`.
I0417 04:43:43.402438 137978123536192 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 0x7d67a4695c70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d67a4695c70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d688c31e8a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d688c31e8a0>}).
I0417 04:43:43.402773 137978123536192 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0417 04:43:43.402854 137978123536192 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d67a4116020> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0417 04:43:44.525395 137978123536192 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints
I0417 04:43:44.970064 137978123536192 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d67a468ce00>
I0417 04:43:44.970238 137978123536192 checkpointing.py:302] Checkpoint manager created!
I0417 04:43:45.964460 137978123536192 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0417 04:43:45.964579 137978123536192 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0417 04:43:47.356381 137978123536192 maxtext_utils.py:1805]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0417 04:43:47.356496 137978123536192 maxtext_utils.py:1805]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0417 04:43:47.356547 137978123536192 maxtext_utils.py:1805]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0417 04:43:47.356589 137978123536192 maxtext_utils.py:1805]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0417 04:43:47.356626 137978123536192 maxtext_utils.py:1805]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0417 04:43:47.356661 137978123536192 maxtext_utils.py:1805]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0417 04:43:47.356697 137978123536192 maxtext_utils.py:1805]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0417 04:43:47.356734 137978123536192 maxtext_utils.py:1805]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0417 04:43:47.356769 137978123536192 maxtext_utils.py:1805]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0417 04:43:47.356803 137978123536192 maxtext_utils.py:1805]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0417 04:43:47.356848 137978123536192 maxtext_utils.py:1805]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0417 04:43:47.356887 137978123536192 maxtext_utils.py:1805]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')

I0417 04:43:47.757178 137978123536192 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0417 04:43:47.757272 137978123536192 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 04:43:47.762472 137978123536192 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0417 04:43:47.762529 137978123536192 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 04:43:47.778594 137978123536192 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 04:43:47.778658 137978123536192 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 04:43:47.794322 137978123536192 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 04:43:47.794381 137978123536192 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 04:43:47.817950 137978123536192 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 04:43:47.818023 137978123536192 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 04:43:47.833648 137978123536192 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 04:43:47.833710 137978123536192 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 04:43:47.849170 137978123536192 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 04:43:47.849236 137978123536192 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 04:43:47.878446 137978123536192 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0417 04:43:47.878519 137978123536192 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 04:43:47.898110 137978123536192 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0417 04:43:47.898180 137978123536192 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0417 04:43:58.566242 137978123536192 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.
I0417 04:43:58.568628 137978123536192 metric_logger.py:289] number parameters: 1.104 billion
I0417 04:44:09.854140 137978123536192 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0417 04:44:10.304040 137978123536192 checkpointing.py:798] Waited 0.44988512992858887 seconds for step 0 to finish before starting checkpointing.
I0417 04:44:10.306383 137978123536192 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 04:44:10.308160 137978123536192 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0417 04:44:10.309560 137978123536192 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/0.
I0417 04:44:11.052208 137978123536192 signaling_client.py:364] Using JaxDistributedSignalingClient
I0417 04:44:11.053119 137978123536192 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0417 04:44:11.053177 137978123536192 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
I0417 04:44:11.332123 137978123536192 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.280674s
I0417 04:44:11.332289 137978123536192 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.322 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28987574577331543 s) (per-host)
I0417 04:44:11.332341 137978123536192 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.289938s (batch_requests_ready=0.003235s, total_serialization_initiated=0.286633s, others=0.000070s)
I0417 04:44:11.332425 137978123536192 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.293971s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.293953)
I0417 04:44:11.333236 137978123536192 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.02 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/0.
I0417 04:44:11.333571 137849928152832 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 05:04:11.333534
I0417 04:44:11.340899 137978123536192 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 04:44:11.341204 137849873610496 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 04:44:11.341362 137978123536192 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776401050.306364, 'wait_for_prev_duration_secs': 6.222724914550781e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776401050.308199, 'checkpointer_blocking_duration_secs': 1.0255322456359863, 'get_old_steps_start_time': 1776401051.333754, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776401050.3044252, 'checkpoint_manager_blocking_duration_secs': 1.036895513534546}
I0417 04:44:11.341474 137978123536192 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0417 04:44:11.341543 137978123536192 max_utils.py:750] 
Memstats: After params initialized:
I0417 04:44:11.341592 137978123536192 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0417 04:44:11.341630 137978123536192 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0417 04:44:11.341656 137978123536192 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0417 04:44:11.341681 137978123536192 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0417 04:44:11.659514 137978123536192 metric_logger.py:185] completed step: 0, seconds: 11.284, TFLOP/s/device: 1.204, Tokens/s/device: 181.488, total_weights: 65536, loss: 10.874
I0417 04:44:11.839210 137978123536192 metric_logger.py:185] completed step: 1, seconds: 1.804, TFLOP/s/device: 7.533, Tokens/s/device: 1135.417, total_weights: 65536, loss: 10.874
I0417 04:44:12.274017 137978123536192 metric_logger.py:185] completed step: 2, seconds: 0.034, TFLOP/s/device: 401.653, Tokens/s/device: 60541.563, total_weights: 65536, loss: 10.262
I0417 04:44:12.421495 137978123536192 metric_logger.py:185] completed step: 3, seconds: 0.424, TFLOP/s/device: 32.059, Tokens/s/device: 4832.297, total_weights: 65536, loss: 9.732
I0417 04:44:12.717365 137978123536192 metric_logger.py:185] completed step: 4, seconds: 0.165, TFLOP/s/device: 82.548, Tokens/s/device: 12442.511, total_weights: 65536, loss: 9.274
I0417 04:44:12.724716 137978123536192 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.364, Tokens/s/device: 13922.123, total_weights: 65536, loss: 8.888
I0417 04:44:15.661230    2785 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0417 04:44:17.704971 137849882003200 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/0/items/array_metadatas/process_6
I0417 04:44:34.915537 137978123536192 metric_logger.py:185] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.838, Tokens/s/device: 6909.185, total_weights: 65536, loss: 8.589
I0417 04:44:35.062883 137978123536192 metric_logger.py:185] completed step: 7, seconds: 22.045, TFLOP/s/device: 0.616, Tokens/s/device: 92.902, total_weights: 65536, loss: 8.380
I0417 04:44:35.210349 137978123536192 metric_logger.py:185] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.871, Tokens/s/device: 13395.602, total_weights: 65536, loss: 8.251
I0417 04:44:35.358914 137978123536192 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0417 04:44:35.359591 137978123536192 checkpointing.py:798] Waited 0.0006899833679199219 seconds for step 9 to finish before starting checkpointing.
I0417 04:44:35.361495 137978123536192 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 04:44:47.194017 137849928152832 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 43.694 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.15154957771301 s) (per-host)
I0417 04:44:47.194141 137849928152832 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.860452s.
I0417 04:44:57.217473 137849928152832 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 45.883768s.
I0417 04:44:57.217755 137849873610496 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 04:44:57.217900 137849873610496 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 04:44:57.217952 137849873610496 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0417 04:44:57.219383 137849873610496 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0417 04:44:57.219542 137978123536192 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0417 04:44:57.219700 137978123536192 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.858197 seconds. If this number is high, consider checkpointing less frequently.
I0417 04:44:57.221681 137978123536192 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0417 04:44:57.223770 137978123536192 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/9.
I0417 04:44:57.917626 137978123536192 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0417 04:44:57.917717 137978123536192 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
I0417 04:44:57.955496 137978123536192 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039324s
I0417 04:44:57.955644 137978123536192 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 33.173 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.046500444412231445 s) (per-host)
I0417 04:44:57.955695 137978123536192 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046563s (batch_requests_ready=0.002900s, total_serialization_initiated=0.043595s, others=0.000067s)
I0417 04:44:57.955780 137978123536192 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050579s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.050564)
I0417 04:44:57.956464 137978123536192 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.73 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/9.
I0417 04:44:57.956805 137849873610496 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 05:04:57.956764
I0417 04:44:57.958776 137978123536192 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 04:44:57.959057 137849911445248 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0417 04:44:57.959186 137978123536192 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776401075.3614657, 'wait_for_prev_duration_secs': 21.85819721221924, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776401097.2217214, 'checkpointer_blocking_duration_secs': 0.7352385520935059, 'get_old_steps_start_time': 1776401097.9569843, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776401075.3598392, 'checkpoint_manager_blocking_duration_secs': 22.59930443763733}
I0417 04:44:57.959324 137978123536192 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0417 04:44:57.959381 137978123536192 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 04:45:03.509736 137849882003200 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_02_synthetic/checkpoints/9/items/array_metadatas/process_6
I0417 04:45:39.448125 137849873610496 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.027 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.53894233703613 s) (per-host)
I0417 04:45:39.448245 137849873610496 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.491314s.
I0417 04:45:45.963738 137849873610496 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.006791s.
I0417 04:45:45.963985 137849911445248 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 04:45:45.964045 137849911445248 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0417 04:45:45.964088 137849911445248 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0417 04:45:45.965771 137849911445248 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0417 04:45:45.965951 137978123536192 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0417 04:45:45.966131 137978123536192 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 04:45:45.966816 137978123536192 metric_logger.py:185] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.963, Tokens/s/device: 13861.627, total_weights: 65536, loss: 8.175
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 17 04:45:53 UTC 2026
EXIT_CODE=0