MaxView

‹ —Case: 01_base02_synthetic ›

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

MetricLinen  7f06c99acNNX  7f06c99acDiff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s92.27391.498-0.775
Tok/s13908.413791.5-116.89
Avg s/step2.7642.768+0.004
Memory %1.381.380
JAX0.9.20.9.2

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

Linen  ·  7f06c99ac  ·  feat_nnx_post_train_fixes_20260422_123906  ·  full log
XPK Start: Wed Apr 22 12:39:42 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-22 12:40:06.974617: 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)
I0422 12:40:07.509227 138132989814592 max_utils.py:273] Attempting to initialize the jax distributed system...
I0422 12:40:16.552016 138132989814592 distributed.py:149] Starting JAX distributed service on [::]:8482
I0422 12:40:16.554450 138132989814592 distributed.py:172] Connecting to JAX distributed service on mt-01-base-ild4x-slice-job-0-0.mt-01-base-ild4x:8482
I0422 12:41:37.002501 138132989814592 max_utils.py:284] Jax distributed system initialized!
I0422 12:41:43.265319 138132989814592 max_utils.py:800] System Information: Jax Version: 0.9.2
I0422 12:41:43.265419 138132989814592 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0422 12:41:43.265459 138132989814592 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
I0422 12:41:43.265492 138132989814592 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0422 12:41:43.951322 138132989814592 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0422 12:41:43.951917 138132989814592 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0422 12:41:43.952093 138132989814592 checkpointing.py:688] Setting up checkpoint logger...
I0422 12:41:43.952143 138132989814592 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0422 12:41:43.952185 138132989814592 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0422 12:41:43.952518 138132989814592 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 0x7da101514440>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0422 12:41:47.653812 138132989814592 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0422 12:41:47.654046 138132989814592 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d8cbc156b70>}, handler_registry=None
I0422 12:41:47.654279 138132989814592 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d8cbc156b70>` 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`.
I0422 12:41:47.654328 138132989814592 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d8ca4282f60>` 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`.
I0422 12:41:47.654368 138132989814592 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 0x7d8cbc156b70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d8cbc156b70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d8ca4282f60>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d8ca4282f60>}).
I0422 12:41:47.654701 138132989814592 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0422 12:41:47.654772 138132989814592 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d8bcc184180> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0422 12:41:48.814407 138132989814592 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints
I0422 12:41:49.042887 138132989814592 checkpoint_manager.py:929] [process=0][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_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d8ca4561820>
I0422 12:41:49.043057 138132989814592 checkpointing.py:302] Checkpoint manager created!
I0422 12:41:49.953849 138132989814592 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0422 12:41:49.953949 138132989814592 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 12:41:50.331826 138132989814592 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 12:41:50.331914 138132989814592 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 12:41:50.348369 138132989814592 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 12:41:50.348433 138132989814592 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 12:41:50.372325 138132989814592 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 12:41:50.372392 138132989814592 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 12:41:50.389039 138132989814592 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 12:41:50.389105 138132989814592 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 12:41:50.405571 138132989814592 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 12:41:50.405640 138132989814592 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 12:41:50.431112 138132989814592 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0422 12:41:50.431185 138132989814592 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 12:41:50.452124 138132989814592 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0422 12:41:50.452189 138132989814592 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0422 12:41:50.665438 138132989814592 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0422 12:41:50.665539 138132989814592 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0422 12:41:52.081599 138132989814592 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0422 12:41:52.081721 138132989814592 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0422 12:41:52.081774 138132989814592 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0422 12:41:52.081830 138132989814592 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0422 12:41:52.081881 138132989814592 maxtext_utils.py:1827]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0422 12:41:52.081918 138132989814592 maxtext_utils.py:1827]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0422 12:41:52.081970 138132989814592 maxtext_utils.py:1827]  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)
I0422 12:41:52.082022 138132989814592 maxtext_utils.py:1827]  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')
I0422 12:41:52.082062 138132989814592 maxtext_utils.py:1827]  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)
I0422 12:41:52.082103 138132989814592 maxtext_utils.py:1827]  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)
I0422 12:41:52.082151 138132989814592 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0422 12:41:52.082198 138132989814592 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0422 12:41:52.583673 138132989814592 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0422 12:41:52.583763 138132989814592 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0422 12:41:52.599161 138132989814592 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0422 12:41:52.599223 138132989814592 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0422 12:42:03.489425 138132989814592 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.
I0422 12:42:03.497046 138132989814592 metric_logger.py:301] number parameters: 1.104 billion
I0422 12:42:14.952623 138132989814592 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0422 12:42:15.105007 138132989814592 checkpointing.py:798] Waited 0.1523609161376953 seconds for step 0 to finish before starting checkpointing.
I0422 12:42:15.107412 138132989814592 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 12:42:15.109440 138132989814592 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0422 12:42:15.111033 138132989814592 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0.
I0422 12:42:15.427024 138132989814592 signaling_client.py:364] Using JaxDistributedSignalingClient
I0422 12:42:15.442699 138132989814592 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0422 12:42:15.442767 138132989814592 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
I0422 12:42:15.683201 138003490682624 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0
I0422 12:42:15.708964 138132989814592 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.268718s
I0422 12:42:15.709800 138132989814592 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.610 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2749631404876709 s) (per-host)
I0422 12:42:15.709881 138132989814592 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.275067s (batch_requests_ready=0.002148s, total_serialization_initiated=0.272150s, others=0.000768s)
I0422 12:42:15.709966 138132989814592 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.278883s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.278865)
I0422 12:42:15.710829 138132989814592 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.60 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0.
I0422 12:42:15.711171 138001875187456 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 13:02:15.711127
I0422 12:42:16.148641 138132989814592 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 12:42:16.149016 138003499075328 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0422 12:42:16.149158 138132989814592 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776861735.1073942, 'wait_for_prev_duration_secs': 5.8650970458984375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776861735.1094797, 'checkpointer_blocking_duration_secs': 0.6018466949462891, 'get_old_steps_start_time': 1776861735.7113516, 'get_old_steps_duration_secs': 2.6941299438476562e-05, 'checkpoint_manager_blocking_start_time': 1776861735.1054769, 'checkpoint_manager_blocking_duration_secs': 1.0436396598815918}
I0422 12:42:16.149318 138132989814592 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0422 12:42:16.149374 138132989814592 max_utils.py:750] 
Memstats: After params initialized:
I0422 12:42:16.149430 138132989814592 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0422 12:42:16.149466 138132989814592 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0422 12:42:16.149493 138132989814592 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0422 12:42:16.149517 138132989814592 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0422 12:42:16.486083 138132989814592 metric_logger.py:196] completed step: 0, seconds: 11.152, TFLOP/s/device: 1.218, Tokens/s/device: 183.649, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0422 12:42:16.489251 138132989814592 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/tensorboard/'
I0422 12:42:16.902481 138132989814592 metric_logger.py:196] completed step: 1, seconds: 1.532, TFLOP/s/device: 8.871, Tokens/s/device: 1337.147, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0422 12:42:17.048641 138132989814592 metric_logger.py:196] completed step: 2, seconds: 0.417, TFLOP/s/device: 32.566, Tokens/s/device: 4908.646, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0422 12:42:17.195709 138132989814592 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1513.886, Tokens/s/device: 228189.415, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0422 12:42:18.392124 138003003471616 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0/items
I0422 12:42:18.561130 138000808290048 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776861738138504234, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776861738138504234, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0/_CHECKPOINT_METADATA
I0422 12:42:19.525957    2897 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0422 12:42:21.732331 138001891972864 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0/items/array_metadatas/process_0
I0422 12:42:39.521968 138132989814592 metric_logger.py:196] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.355, Tokens/s/device: 13920.704, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0422 12:42:39.531165 138132989814592 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.328, Tokens/s/device: 13916.731, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0422 12:42:39.677225 138132989814592 metric_logger.py:196] completed step: 6, seconds: 22.325, TFLOP/s/device: 0.609, Tokens/s/device: 91.734, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0422 12:42:39.824591 138132989814592 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 2032.784, Tokens/s/device: 306403.351, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0422 12:42:39.971904 138132989814592 metric_logger.py:196] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.477, Tokens/s/device: 13939.179, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0422 12:42:40.118497 138132989814592 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0422 12:42:40.119255 138132989814592 checkpointing.py:798] Waited 0.0007793903350830078 seconds for step 9 to finish before starting checkpointing.
I0422 12:42:40.121405 138132989814592 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 12:42:56.981374 138001883580160 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.455424s (commit=35.590429s, array_metadata_write=1.864994s)
I0422 12:42:56.982503 138001875187456 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 38.019 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.54763889312744 s) (per-host)
I0422 12:42:56.982643 138001875187456 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.271352s.
I0422 12:42:57.641692 138001875187456 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776861738138504234, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0/_CHECKPOINT_METADATA
I0422 12:42:58.426706 138000808290048 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776861738138504234, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0/_CHECKPOINT_METADATA
I0422 12:42:59.020568 138001875187456 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0422 12:42:59.233109 138001875187456 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000227s.
I0422 12:42:59.944262 138001875187456 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.172240s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0/items
I0422 12:42:59.945947 138001875187456 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0/items
I0422 12:43:00.462480 138001875187456 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0
I0422 12:43:02.767766 138001875187456 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0`.
I0422 12:43:02.768604 138001875187456 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 47.66 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0
I0422 12:43:02.770059 138001875187456 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 47.058763s.
I0422 12:43:02.770278 138003499075328 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 12:43:02.770404 138003499075328 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0422 12:43:02.770466 138003499075328 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0422 12:43:02.773365 138003499075328 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0422 12:43:02.773487 138132989814592 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0422 12:43:02.773592 138132989814592 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.652179 seconds. If this number is high, consider checkpointing less frequently.
I0422 12:43:02.775454 138132989814592 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0422 12:43:02.777595 138132989814592 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9.
I0422 12:43:03.072513 138132989814592 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0422 12:43:03.072617 138132989814592 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
I0422 12:43:03.098658 138132989814592 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026853s
I0422 12:43:03.099362 138132989814592 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 48.599 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03174090385437012 s) (per-host)
I0422 12:43:03.099428 138132989814592 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031825s (batch_requests_ready=0.001901s, total_serialization_initiated=0.029268s, others=0.000657s)
I0422 12:43:03.099502 138132989814592 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035687s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.035670)
I0422 12:43:03.100202 138132989814592 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9.
I0422 12:43:03.100515 138003499075328 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 13:03:03.100478
I0422 12:43:03.107889 138132989814592 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 12:43:03.108135 138001904551680 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0422 12:43:03.108269 138132989814592 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776861760.1213744, 'wait_for_prev_duration_secs': 22.65217876434326, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776861782.775494, 'checkpointer_blocking_duration_secs': 0.32512903213500977, 'get_old_steps_start_time': 1776861783.100656, 'get_old_steps_duration_secs': 3.552436828613281e-05, 'checkpoint_manager_blocking_start_time': 1776861760.119502, 'checkpoint_manager_blocking_duration_secs': 22.98872661590576}
I0422 12:43:03.108397 138132989814592 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0422 12:43:03.108453 138132989814592 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 12:43:03.309337 138001875187456 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9
I0422 12:43:05.501394 137997864462080 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9/items
I0422 12:43:08.801852 138002437232384 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9/items/array_metadatas/process_0
I0422 12:43:47.703794 138001921337088 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.076366s (commit=39.858499s, array_metadata_write=1.217867s)
I0422 12:43:47.704924 138003499075328 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 35.387 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.6372754573822 s) (per-host)
I0422 12:43:47.705050 138003499075328 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.603804s.
I0422 12:43:49.204932 138003499075328 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000233s.
I0422 12:43:49.584741 138003499075328 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0422 12:43:50.209072 138003499075328 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.592225s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9/items
I0422 12:43:50.210744 138003499075328 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9/items
I0422 12:43:50.784270 138003499075328 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9
I0422 12:43:52.205983 138003499075328 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9`.
I0422 12:43:52.206806 138003499075328 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 49.43 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9
I0422 12:43:52.208186 138003499075328 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 49.106939s.
I0422 12:43:52.208401 138001904551680 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 12:43:52.208514 138001904551680 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0422 12:43:52.208589 138001904551680 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0422 12:43:52.210767 138001904551680 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0422 12:43:52.210947 138132989814592 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0422 12:43:52.211089 138132989814592 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 12:43:52.212081 138132989814592 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.273, Tokens/s/device: 13908.414, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Wed Apr 22 12:44:03 UTC 2026
EXIT_CODE=0
NNX  ·  7f06c99ac  ·  feat_nnx_post_train_fixes_20260422_123906  ·  full log
XPK Start: Wed Apr 22 14:17:29 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-22 14:17:53.817628: 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)
I0422 14:17:54.001724 139740245141312 max_utils.py:273] Attempting to initialize the jax distributed system...
I0422 14:18:03.044355 139740245141312 distributed.py:149] Starting JAX distributed service on [::]:8482
I0422 14:18:03.046688 139740245141312 distributed.py:172] Connecting to JAX distributed service on mt-01-base-lzegb-slice-job-0-0.mt-01-base-lzegb:8482
I0422 14:18:04.401181 139740245141312 max_utils.py:284] Jax distributed system initialized!
I0422 14:18:10.490783 139740245141312 max_utils.py:800] System Information: Jax Version: 0.9.2
I0422 14:18:10.490882 139740245141312 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0422 14:18:10.490922 139740245141312 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
I0422 14:18:10.490956 139740245141312 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0422 14:18:11.195937 139740245141312 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0422 14:18:11.309167 139740245141312 checkpointing.py:688] Setting up checkpoint logger...
I0422 14:18:11.309284 139740245141312 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0422 14:18:11.309330 139740245141312 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0422 14:18:11.309524 139740245141312 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 0x7f1764fc26c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0422 14:18:15.130670 139740245141312 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0422 14:18:15.130904 139740245141312 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 0x7f01f843f7d0>}, handler_registry=None
I0422 14:18:15.131155 139740245141312 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f01f843f7d0>` 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`.
I0422 14:18:15.131204 139740245141312 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f01d8669460>` 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`.
I0422 14:18:15.131239 139740245141312 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 0x7f01f843f7d0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f01f843f7d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f01d8669460>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f01d8669460>}).
I0422 14:18:15.131572 139740245141312 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0422 14:18:15.131645 139740245141312 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f02d81c5da0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0422 14:18:15.849968 139740245141312 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints
I0422 14:18:16.036856 139740245141312 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_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f02d837fad0>
I0422 14:18:16.037015 139740245141312 checkpointing.py:302] Checkpoint manager created!
I0422 14:18:16.995045 139740245141312 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0422 14:18:16.995173 139740245141312 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0422 14:18:18.360872 139740245141312 maxtext_utils.py:1836]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0422 14:18:18.360984 139740245141312 maxtext_utils.py:1836]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0422 14:18:18.361036 139740245141312 maxtext_utils.py:1836]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0422 14:18:18.361078 139740245141312 maxtext_utils.py:1836]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0422 14:18:18.361129 139740245141312 maxtext_utils.py:1836]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0422 14:18:18.361166 139740245141312 maxtext_utils.py:1836]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0422 14:18:18.361203 139740245141312 maxtext_utils.py:1836]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0422 14:18:18.361244 139740245141312 maxtext_utils.py:1836]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0422 14:18:18.361279 139740245141312 maxtext_utils.py:1836]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0422 14:18:18.361316 139740245141312 maxtext_utils.py:1836]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0422 14:18:18.361349 139740245141312 maxtext_utils.py:1836]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)

I0422 14:18:18.361381 139740245141312 maxtext_utils.py:1836]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')
I0422 14:18:18.760135 139740245141312 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0422 14:18:18.760227 139740245141312 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 14:18:18.765470 139740245141312 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0422 14:18:18.765524 139740245141312 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 14:18:18.781788 139740245141312 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 14:18:18.781847 139740245141312 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 14:18:18.797403 139740245141312 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 14:18:18.797458 139740245141312 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 14:18:18.822157 139740245141312 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 14:18:18.822218 139740245141312 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 14:18:18.838088 139740245141312 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 14:18:18.838160 139740245141312 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 14:18:18.853867 139740245141312 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 14:18:18.853927 139740245141312 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 14:18:18.884062 139740245141312 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0422 14:18:18.884142 139740245141312 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 14:18:18.903902 139740245141312 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0422 14:18:18.903958 139740245141312 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0422 14:18:30.138141 139740245141312 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.
I0422 14:18:30.140772 139740245141312 metric_logger.py:301] number parameters: 1.104 billion
I0422 14:18:41.285842 139740245141312 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0422 14:18:41.436886 139740245141312 checkpointing.py:798] Waited 0.1510298252105713 seconds for step 0 to finish before starting checkpointing.
I0422 14:18:41.439318 139740245141312 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 14:18:41.440999 139740245141312 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0422 14:18:41.442447 139740245141312 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0.
I0422 14:18:41.804821 139740245141312 signaling_client.py:364] Using JaxDistributedSignalingClient
I0422 14:18:41.805991 139740245141312 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0422 14:18:41.806053 139740245141312 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
I0422 14:18:42.080789 139740245141312 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.276766s
I0422 14:18:42.080956 139740245141312 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.397 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2858095169067383 s) (per-host)
I0422 14:18:42.081007 139740245141312 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.285869s (batch_requests_ready=0.002941s, total_serialization_initiated=0.282862s, others=0.000067s)
I0422 14:18:42.081134 139740245141312 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.289904s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.289887)
I0422 14:18:42.081923 139740245141312 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.64 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0.
I0422 14:18:42.082263 139610657306368 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 14:38:42.082223
I0422 14:18:42.099401 139740245141312 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 14:18:42.099737 139609934272256 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0422 14:18:42.099896 139740245141312 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776867521.4392989, 'wait_for_prev_duration_secs': 6.031990051269531e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776867521.4410388, 'checkpointer_blocking_duration_secs': 0.6413822174072266, 'get_old_steps_start_time': 1776867522.0824487, 'get_old_steps_duration_secs': 3.647804260253906e-05, 'checkpoint_manager_blocking_start_time': 1776867521.4373286, 'checkpoint_manager_blocking_duration_secs': 0.6625256538391113}
I0422 14:18:42.100006 139740245141312 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0422 14:18:42.100073 139740245141312 max_utils.py:750] 
Memstats: After params initialized:
I0422 14:18:42.100136 139740245141312 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0422 14:18:42.100170 139740245141312 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0422 14:18:42.100197 139740245141312 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0422 14:18:42.100221 139740245141312 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0422 14:18:42.418867 139740245141312 metric_logger.py:196] completed step: 0, seconds: 11.144, TFLOP/s/device: 1.219, Tokens/s/device: 183.777, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0422 14:18:42.595121 139740245141312 metric_logger.py:196] completed step: 1, seconds: 1.131, TFLOP/s/device: 12.013, Tokens/s/device: 1810.680, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0422 14:18:43.038325 139740245141312 metric_logger.py:196] completed step: 2, seconds: 0.030, TFLOP/s/device: 453.161, Tokens/s/device: 68305.373, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28632.133
I0422 14:18:43.186128 139740245141312 metric_logger.py:196] completed step: 3, seconds: 0.433, TFLOP/s/device: 31.369, Tokens/s/device: 4728.231, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16855.789
I0422 14:18:43.481759 139740245141312 metric_logger.py:196] completed step: 4, seconds: 0.165, TFLOP/s/device: 82.572, Tokens/s/device: 12446.216, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10661.594
I0422 14:18:43.489531 139740245141312 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.177, Tokens/s/device: 13893.883, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7245.978
I0422 14:18:51.022820    2827 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0422 14:18:53.115927 139609975543552 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_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/0/items/array_metadatas/process_6
I0422 14:19:06.046526 139740245141312 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.841, Tokens/s/device: 6909.628, total_weights: 65536, loss: 8.589, lm_loss: 8.589, perplexity: 5371.156
I0422 14:19:06.193994 139740245141312 metric_logger.py:196] completed step: 7, seconds: 22.411, TFLOP/s/device: 0.606, Tokens/s/device: 91.386, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4360.465
I0422 14:19:06.341856 139740245141312 metric_logger.py:196] completed step: 8, seconds: 0.154, TFLOP/s/device: 88.435, Tokens/s/device: 13329.949, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.256
I0422 14:19:06.490061 139740245141312 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0422 14:19:06.490796 139740245141312 checkpointing.py:798] Waited 0.0007431507110595703 seconds for step 9 to finish before starting checkpointing.
I0422 14:19:06.493047 139740245141312 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 14:19:26.386564 139610657306368 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 35.424 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.591346979141235 s) (per-host)
I0422 14:19:26.386690 139610657306368 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.304317s.
I0422 14:19:31.831054 139610657306368 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.748666s.
I0422 14:19:31.831355 139609934272256 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 14:19:31.831473 139609934272256 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0422 14:19:31.831542 139609934272256 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0422 14:19:31.833453 139609934272256 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0422 14:19:31.833651 139740245141312 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.
W0422 14:19:31.833785 139740245141312 checkpoint_manager.py:1452] Waiting for previous save to complete took 25.340738 seconds. If this number is high, consider checkpointing less frequently.
I0422 14:19:31.835722 139740245141312 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0422 14:19:31.837839 139740245141312 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9.
I0422 14:19:32.555218 139740245141312 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0422 14:19:32.555310 139740245141312 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
I0422 14:19:32.592524 139740245141312 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038947s
I0422 14:19:32.592685 139740245141312 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 33.534 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04600071907043457 s) (per-host)
I0422 14:19:32.592736 139740245141312 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046061s (batch_requests_ready=0.002645s, total_serialization_initiated=0.043349s, others=0.000067s)
I0422 14:19:32.592832 139740245141312 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050864s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.050848)
I0422 14:19:32.593569 139740245141312 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9.
I0422 14:19:32.593865 139609934272256 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 14:39:32.593832
I0422 14:19:32.598085 139740245141312 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 14:19:32.598383 139610640520960 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0422 14:19:32.598551 139740245141312 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776867546.4930177, 'wait_for_prev_duration_secs': 25.34073758125305, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776867571.8357625, 'checkpointer_blocking_duration_secs': 0.7582550048828125, 'get_old_steps_start_time': 1776867572.5940402, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776867546.4910326, 'checkpoint_manager_blocking_duration_secs': 26.107484579086304}
I0422 14:19:32.598664 139740245141312 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0422 14:19:32.598707 139740245141312 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 14:19:38.220992 139609975543552 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_20260422_123906/nnx_xpk_feat_nnx_post_train_fixes_20260422_123906_01_base/checkpoints/9/items/array_metadatas/process_6
I0422 14:20:15.800077 139609934272256 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.520 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.253352642059326 s) (per-host)
I0422 14:20:15.800219 139609934272256 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.206234s.
I0422 14:20:21.405513 139609934272256 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.811514s.
I0422 14:20:21.405833 139610640520960 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 14:20:21.405972 139610640520960 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0422 14:20:21.406026 139610640520960 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0422 14:20:21.408512 139610640520960 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0422 14:20:21.408736 139740245141312 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.
I0422 14:20:21.408922 139740245141312 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 14:20:21.409969 139740245141312 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.498, Tokens/s/device: 13791.524, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3551.438
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Wed Apr 22 14:20:29 UTC 2026
EXIT_CODE=0