MaxView

‹ 06_grad_accumCase: 07_eval08_checkpoint_async_true_resume ›

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

MetricLinen  574ad3fb9NNX  03aaf29baDiff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s92.03792.499+0.462
Tok/s13872.813942.4+69.605
Avg s/step2.7972.783-0.014
Memory %1.381.380
JAX0.9.20.9.2

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

Linen  ·  574ad3fb9  ·  feat_nnx_post_train_fixes_20260420_153543  ·  full log
XPK Start: Mon Apr 20 16:01:50 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-20 16:02:14.947017: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0420 16:02:15.124485 133292680877888 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 16:02:24.166855 133292680877888 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 16:02:24.169314 133292680877888 distributed.py:172] Connecting to JAX distributed service on mt-07-eval-sd6r6-slice-job-0-0.mt-07-eval-sd6r6:8482
I0420 16:02:25.079352 133292680877888 max_utils.py:284] Jax distributed system initialized!
I0420 16:02:31.296610 133292680877888 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 16:02:31.296710 133292680877888 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 16:02:31.296748 133292680877888 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
I0420 16:02:31.296783 133292680877888 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 16:02:31.981173 133292680877888 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 16:02:31.981767 133292680877888 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 16:02:31.981944 133292680877888 checkpointing.py:688] Setting up checkpoint logger...
I0420 16:02:31.981993 133292680877888 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 16:02:31.982035 133292680877888 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 16:02:31.982374 133292680877888 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 0x793a0630a600>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 16:02:36.408969 133292680877888 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 16:02:36.409220 133292680877888 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 0x7938db6e84d0>}, handler_registry=None
I0420 16:02:36.409455 133292680877888 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7938db6e84d0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0420 16:02:36.409503 133292680877888 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7925c80c8500>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0420 16:02:36.409539 133292680877888 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 0x7938db6e84d0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7938db6e84d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7925c80c8500>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7925c80c8500>}).
I0420 16:02:36.409861 133292680877888 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 16:02:36.409933 133292680877888 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x792598113740> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 16:02:37.776868 133292680877888 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints
I0420 16:02:38.179798 133292680877888 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_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7924c8161610>
I0420 16:02:38.179965 133292680877888 checkpointing.py:302] Checkpoint manager created!
I0420 16:02:39.108124 133292680877888 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 16:02:39.108220 133292680877888 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 16:02:39.488513 133292680877888 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 16:02:39.488604 133292680877888 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 16:02:39.504935 133292680877888 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 16:02:39.504995 133292680877888 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 16:02:39.528384 133292680877888 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 16:02:39.528453 133292680877888 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 16:02:39.544624 133292680877888 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 16:02:39.544684 133292680877888 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 16:02:39.560753 133292680877888 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 16:02:39.560810 133292680877888 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 16:02:39.585397 133292680877888 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 16:02:39.585464 133292680877888 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 16:02:39.606174 133292680877888 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 16:02:39.606232 133292680877888 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 16:02:39.822102 133292680877888 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 16:02:39.822202 133292680877888 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 16:02:41.248255 133292680877888 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 16:02:41.248373 133292680877888 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)
I0420 16:02:41.248426 133292680877888 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)
I0420 16:02:41.248492 133292680877888 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')
I0420 16:02:41.248547 133292680877888 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)
I0420 16:02:41.248585 133292680877888 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)
I0420 16:02:41.248636 133292680877888 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)
I0420 16:02:41.248686 133292680877888 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')
I0420 16:02:41.248726 133292680877888 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)
I0420 16:02:41.248760 133292680877888 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)
I0420 16:02:41.248806 133292680877888 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0420 16:02:41.248853 133292680877888 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0420 16:02:41.745221 133292680877888 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 16:02:41.745315 133292680877888 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 16:02:41.760577 133292680877888 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 16:02:41.760640 133292680877888 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 16:02:52.898319 133292680877888 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0420 16:02:52.906212 133292680877888 metric_logger.py:301] number parameters: 1.104 billion
I0420 16:03:04.475815 133292680877888 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 16:03:04.629204 133292680877888 checkpointing.py:798] Waited 0.15337157249450684 seconds for step 0 to finish before starting checkpointing.
I0420 16:03:04.631602 133292680877888 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 16:03:04.633289 133292680877888 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0420 16:03:04.634747 133292680877888 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0.
I0420 16:03:05.038127 133292680877888 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 16:03:05.052310 133292680877888 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 16:03:05.052377 133292680877888 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0420 16:03:05.322314 133163552179968 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0
I0420 16:03:05.324501 133292680877888 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.272981s
I0420 16:03:05.325404 133292680877888 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.521 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27938365936279297 s) (per-host)
I0420 16:03:05.325479 133292680877888 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279479s (batch_requests_ready=0.002209s, total_serialization_initiated=0.276450s, others=0.000820s)
I0420 16:03:05.325582 133292680877888 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283667s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.283650)
I0420 16:03:05.326413 133292680877888 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.69 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0.
I0420 16:03:05.326711 133163560572672 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 16:23:05.326679
I0420 16:03:05.335343 133292680877888 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 16:03:05.335625 133162457048832 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 16:03:05.335781 133292680877888 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776700984.6315827, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776700984.6333277, 'checkpointer_blocking_duration_secs': 0.6935257911682129, 'get_old_steps_start_time': 1776700985.3268793, 'get_old_steps_duration_secs': 2.7894973754882812e-05, 'checkpoint_manager_blocking_start_time': 1776700984.629646, 'checkpoint_manager_blocking_duration_secs': 0.7060964107513428}
I0420 16:03:05.335887 133292680877888 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 16:03:05.335939 133292680877888 max_utils.py:750] 
Memstats: After params initialized:
I0420 16:03:05.335990 133292680877888 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0420 16:03:05.336023 133292680877888 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0420 16:03:05.336051 133292680877888 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0420 16:03:05.336076 133292680877888 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0420 16:03:05.678694 133292680877888 metric_logger.py:196] completed step: 0, seconds: 11.238, TFLOP/s/device: 1.209, Tokens/s/device: 182.233, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 16:03:05.681522 133292680877888 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/tensorboard/'
I0420 16:03:06.107513 133292680877888 metric_logger.py:196] completed step: 1, seconds: 1.201, TFLOP/s/device: 11.314, Tokens/s/device: 1705.368, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 16:03:06.253535 133292680877888 metric_logger.py:196] completed step: 2, seconds: 0.430, TFLOP/s/device: 31.628, Tokens/s/device: 4767.325, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0420 16:03:06.401103 133292680877888 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1453.480, Tokens/s/device: 219084.296, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0420 16:03:07.966447 133163535394560 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/items
I0420 16:03:08.083942 133160897607424 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776700987611753994, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776700987611753994, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0420 16:03:09.324313    2829 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 16:03:12.421861 133162508220160 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_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/items/array_metadatas/process_0
I0420 16:03:29.343910 133292680877888 metric_logger.py:196] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.266, Tokens/s/device: 13907.280, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0420 16:03:29.352586 133292680877888 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.216, Tokens/s/device: 13899.729, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0420 16:03:29.498704 133292680877888 metric_logger.py:196] completed step: 6, seconds: 22.941, TFLOP/s/device: 0.592, Tokens/s/device: 89.271, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0420 16:03:29.646219 133292680877888 metric_logger.py:196] completed step: 7, seconds: 0.006, TFLOP/s/device: 2116.705, Tokens/s/device: 319052.812, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0420 16:03:29.793570 133292680877888 metric_logger.py:196] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.326, Tokens/s/device: 13916.352, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0420 16:03:29.940184 133292680877888 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 16:03:29.940778 133292680877888 checkpointing.py:798] Waited 0.0006163120269775391 seconds for step 9 to finish before starting checkpointing.
I0420 16:03:29.942922 133292680877888 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 16:03:47.025510 133162465441536 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.700106s (commit=36.422368s, array_metadata_write=1.277737s)
I0420 16:03:47.026665 133163560572672 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 37.627 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.980602741241455 s) (per-host)
I0420 16:03:47.026798 133163560572672 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.699976s.
I0420 16:03:47.758700 133163560572672 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776700987611753994, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0420 16:03:48.494396 133160897607424 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776700987611753994, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0420 16:03:48.499727 133163560572672 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000253s.
I0420 16:03:49.282471 133163560572672 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 16:03:50.219861 133163560572672 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.337330s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/items
I0420 16:03:50.221668 133163560572672 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/items
I0420 16:03:50.785820 133163560572672 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0
I0420 16:03:52.213617 133163560572672 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_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0`.
I0420 16:03:52.214493 133163560572672 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 47.58 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0
I0420 16:03:52.215956 133163560572672 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 46.889132s.
I0420 16:03:52.216186 133162457048832 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 16:03:52.216309 133162457048832 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 16:03:52.216378 133162457048832 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 16:03:52.218167 133162457048832 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 16:03:52.218365 133292680877888 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.
W0420 16:03:52.218506 133292680877888 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.275584 seconds. If this number is high, consider checkpointing less frequently.
I0420 16:03:52.220405 133292680877888 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0420 16:03:52.222420 133292680877888 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9.
I0420 16:03:52.621422 133292680877888 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 16:03:52.621514 133292680877888 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0420 16:03:52.647955 133292680877888 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027301s
I0420 16:03:52.648765 133292680877888 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 48.349 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03190469741821289 s) (per-host)
I0420 16:03:52.648838 133292680877888 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031996s (batch_requests_ready=0.001918s, total_serialization_initiated=0.029363s, others=0.000715s)
I0420 16:03:52.648923 133292680877888 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.036132s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.036117)
I0420 16:03:52.649667 133292680877888 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.43 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9.
I0420 16:03:52.649940 133162457048832 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 16:23:52.649913
I0420 16:03:52.660902 133292680877888 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 16:03:52.661215 133161929385728 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 16:03:52.661384 133292680877888 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776701009.942889, 'wait_for_prev_duration_secs': 22.275583505630493, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776701032.220447, 'checkpointer_blocking_duration_secs': 0.4296140670776367, 'get_old_steps_start_time': 1776701032.6500978, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1776701009.9410017, 'checkpoint_manager_blocking_duration_secs': 22.720349073410034}
I0420 16:03:52.661497 133292680877888 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 16:03:52.661540 133292680877888 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 16:03:52.856370 133163560572672 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9
I0420 16:03:54.658339 133163019912960 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9/items
I0420 16:03:58.025533 133161969104640 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_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9/items/array_metadatas/process_0
I0420 16:04:37.342019 133161937778432 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.536816s (commit=40.201624s, array_metadata_write=1.335192s)
I0420 16:04:37.343230 133162457048832 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 35.317 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.72634172439575 s) (per-host)
I0420 16:04:37.343356 133162457048832 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.693319s.
I0420 16:04:39.549981 133162457048832 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000236s.
I0420 16:04:39.688797 133162457048832 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 16:04:40.628120 133162457048832 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.060738s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9/items
I0420 16:04:40.629848 133162457048832 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9/items
I0420 16:04:41.586507 133162457048832 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9
I0420 16:04:43.912837 133162457048832 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_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9`.
I0420 16:04:43.913752 133162457048832 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 51.69 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9
I0420 16:04:43.915269 133162457048832 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 51.265229s.
I0420 16:04:43.915460 133161929385728 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 16:04:43.915667 133161929385728 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 16:04:43.915772 133161929385728 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 16:04:43.917518 133161929385728 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 16:04:43.917683 133292680877888 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.
I0420 16:04:43.917831 133292680877888 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 16:04:43.918745 133292680877888 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.037, Tokens/s/device: 13872.801, 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: Mon Apr 20 16:04:56 UTC 2026
EXIT_CODE=0
NNX  ·  03aaf29ba  ·  feat_nnx_post_train_fixes_20260420_153543  ·  full log
XPK Start: Mon Apr 20 17:46:56 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-20 17:47:20.505649: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0420 17:47:20.684544 138815951419200 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 17:47:29.724189 138815951419200 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 17:47:29.726455 138815951419200 distributed.py:172] Connecting to JAX distributed service on mt-07-eval-wrv83-slice-job-0-0.mt-07-eval-wrv83:8482
I0420 17:47:32.632542 138815951419200 max_utils.py:284] Jax distributed system initialized!
I0420 17:47:38.819301 138815951419200 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 17:47:38.819401 138815951419200 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 17:47:38.819441 138815951419200 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
I0420 17:47:38.819475 138815951419200 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 17:47:39.859188 138815951419200 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 17:47:39.973129 138815951419200 checkpointing.py:688] Setting up checkpoint logger...
I0420 17:47:39.973240 138815951419200 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 17:47:39.973284 138815951419200 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 17:47:39.973472 138815951419200 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 0x7e3ffd71b1d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 17:47:42.500048 138815951419200 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 17:47:42.500297 138815951419200 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 0x7e3ed78eda30>}, handler_registry=None
I0420 17:47:42.500529 138815951419200 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e3ed78eda30>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0420 17:47:42.500577 138815951419200 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e2bd44cb170>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0420 17:47:42.500612 138815951419200 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 0x7e3ed78eda30>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e3ed78eda30>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e2bd44cb170>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e2bd44cb170>}).
I0420 17:47:42.500940 138815951419200 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 17:47:42.501009 138815951419200 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e2afc62c4a0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 17:47:43.833566 138815951419200 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints
I0420 17:47:43.985531 138815951419200 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/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e3ed7d52ab0>
I0420 17:47:43.985698 138815951419200 checkpointing.py:302] Checkpoint manager created!
I0420 17:47:44.933767 138815951419200 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 17:47:44.933885 138815951419200 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 17:47:46.294951 138815951419200 maxtext_utils.py:1836]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0420 17:47:46.295065 138815951419200 maxtext_utils.py:1836]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0420 17:47:46.295129 138815951419200 maxtext_utils.py:1836]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0420 17:47:46.295173 138815951419200 maxtext_utils.py:1836]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0420 17:47:46.295209 138815951419200 maxtext_utils.py:1836]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0420 17:47:46.295242 138815951419200 maxtext_utils.py:1836]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0420 17:47:46.295278 138815951419200 maxtext_utils.py:1836]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:47:46.295313 138815951419200 maxtext_utils.py:1836]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0420 17:47:46.295347 138815951419200 maxtext_utils.py:1836]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:47:46.295379 138815951419200 maxtext_utils.py:1836]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:47:46.295413 138815951419200 maxtext_utils.py:1836]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0420 17:47:46.295445 138815951419200 maxtext_utils.py:1836]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')

I0420 17:47:46.704479 138815951419200 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 17:47:46.704571 138815951419200 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:47:46.711278 138815951419200 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 17:47:46.711386 138815951419200 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:47:46.732039 138815951419200 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 17:47:46.732117 138815951419200 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:47:46.747739 138815951419200 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 17:47:46.747799 138815951419200 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:47:46.777245 138815951419200 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:47:46.777344 138815951419200 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:47:46.793245 138815951419200 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:47:46.793306 138815951419200 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:47:46.808883 138815951419200 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:47:46.808942 138815951419200 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:47:46.839016 138815951419200 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 17:47:46.839098 138815951419200 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:47:46.858796 138815951419200 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 17:47:46.858859 138815951419200 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 17:47:58.258748 138815951419200 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0420 17:47:58.268199 138815951419200 metric_logger.py:301] number parameters: 1.104 billion
I0420 17:48:09.801244 138815951419200 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 17:48:09.952838 138815951419200 checkpointing.py:798] Waited 0.15157485008239746 seconds for step 0 to finish before starting checkpointing.
I0420 17:48:09.955102 138815951419200 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 17:48:09.956938 138815951419200 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0420 17:48:09.958472 138815951419200 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0.
I0420 17:48:10.724127 138815951419200 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 17:48:10.743722 138815951419200 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0420 17:48:10.743794 138815951419200 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0420 17:48:10.989564 138688805705472 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0
I0420 17:48:11.015735 138815951419200 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273447s
I0420 17:48:11.016541 138815951419200 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.426 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2842984199523926 s) (per-host)
I0420 17:48:11.016606 138815951419200 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.284379s (batch_requests_ready=0.003029s, total_serialization_initiated=0.280621s, others=0.000729s)
I0420 17:48:11.016676 138815951419200 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288433s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.288415)
I0420 17:48:11.017422 138815951419200 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.06 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0.
I0420 17:48:11.017749 138688814098176 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 18:08:11.017711
I0420 17:48:11.026936 138815951419200 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:48:11.027232 138686662428416 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:48:11.027387 138815951419200 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776707289.9550686, 'wait_for_prev_duration_secs': 7.43865966796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776707289.9569776, 'checkpointer_blocking_duration_secs': 1.0608723163604736, 'get_old_steps_start_time': 1776707291.0178735, 'get_old_steps_duration_secs': 2.6702880859375e-05, 'checkpoint_manager_blocking_start_time': 1776707289.9532552, 'checkpoint_manager_blocking_duration_secs': 1.074094533920288}
I0420 17:48:11.027498 138815951419200 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 17:48:11.027570 138815951419200 max_utils.py:750] 
Memstats: After params initialized:
I0420 17:48:11.027621 138815951419200 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0420 17:48:11.027652 138815951419200 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0420 17:48:11.027680 138815951419200 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0420 17:48:11.027706 138815951419200 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0420 17:48:11.382699 138815951419200 metric_logger.py:196] completed step: 0, seconds: 11.199, TFLOP/s/device: 1.213, Tokens/s/device: 182.882, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0420 17:48:11.385826 138815951419200 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/tensorboard/'
I0420 17:48:11.813339 138815951419200 metric_logger.py:196] completed step: 1, seconds: 1.577, TFLOP/s/device: 8.614, Tokens/s/device: 1298.325, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0420 17:48:11.955096 138815951419200 metric_logger.py:196] completed step: 2, seconds: 0.432, TFLOP/s/device: 31.455, Tokens/s/device: 4741.300, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28632.133
I0420 17:48:12.102574 138815951419200 metric_logger.py:196] completed step: 3, seconds: 0.013, TFLOP/s/device: 1062.491, Tokens/s/device: 160150.141, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16855.789
I0420 17:48:12.965862 138688269321984 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/items
I0420 17:48:13.147346 138685046195968 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776707292707210489, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776707292707210489, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0420 17:48:14.145651    2838 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 17:48:15.912185 138687190447872 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/items/array_metadatas/process_0
I0420 17:48:34.541498 138815951419200 metric_logger.py:196] completed step: 4, seconds: 0.143, TFLOP/s/device: 94.916, Tokens/s/device: 14306.771, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10661.594
I0420 17:48:34.552063 138815951419200 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.574, Tokens/s/device: 13953.805, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7245.978
I0420 17:48:34.696721 138815951419200 metric_logger.py:196] completed step: 6, seconds: 22.437, TFLOP/s/device: 0.606, Tokens/s/device: 91.276, total_weights: 65536, loss: 8.589, lm_loss: 8.589, perplexity: 5371.156
I0420 17:48:34.844160 138815951419200 metric_logger.py:196] completed step: 7, seconds: 0.008, TFLOP/s/device: 1706.283, Tokens/s/device: 257189.501, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4360.465
I0420 17:48:34.991519 138815951419200 metric_logger.py:196] completed step: 8, seconds: 0.146, TFLOP/s/device: 92.862, Tokens/s/device: 13997.198, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.256
I0420 17:48:35.139769 138815951419200 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 17:48:35.140480 138815951419200 checkpointing.py:798] Waited 0.0007233619689941406 seconds for step 9 to finish before starting checkpointing.
I0420 17:48:35.142611 138815951419200 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:48:51.470255 138686708578048 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.337914s (commit=36.103671s, array_metadata_write=1.234243s)
I0420 17:48:51.471314 138688814098176 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 38.774 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.739041328430176 s) (per-host)
I0420 17:48:51.471436 138688814098176 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.453016s.
I0420 17:48:52.254723 138688814098176 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776707292707210489, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0420 17:48:52.987885 138688814098176 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000425s.
I0420 17:48:53.045236 138685046195968 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776707292707210489, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0420 17:48:53.658584 138688814098176 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 17:48:54.590931 138688814098176 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.207905s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/items
I0420 17:48:54.592778 138688814098176 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0/items
I0420 17:48:55.131526 138688814098176 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0
I0420 17:48:56.971383 138688814098176 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0`.
I0420 17:48:56.972066 138688814098176 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 47.02 seconds @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/0
I0420 17:48:56.973702 138688814098176 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 45.955279s.
I0420 17:48:56.973891 138686662428416 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:48:56.974007 138686662428416 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:48:56.974069 138686662428416 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:48:56.976737 138686662428416 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 17:48:56.976866 138815951419200 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.
W0420 17:48:56.977007 138815951419200 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.834395 seconds. If this number is high, consider checkpointing less frequently.
I0420 17:48:56.978940 138815951419200 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0420 17:48:56.981027 138815951419200 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9.
I0420 17:48:57.277532 138815951419200 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0420 17:48:57.277630 138815951419200 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0420 17:48:57.305509 138815951419200 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.029342s
I0420 17:48:57.306376 138815951419200 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 40.505 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.038083791732788086 s) (per-host)
I0420 17:48:57.306452 138815951419200 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038178s (batch_requests_ready=0.002647s, total_serialization_initiated=0.034747s, others=0.000784s)
I0420 17:48:57.306535 138815951419200 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042134s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.042119)
I0420 17:48:57.307267 138815951419200 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9.
I0420 17:48:57.307527 138686662428416 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 18:08:57.307506
I0420 17:48:57.314512 138815951419200 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:48:57.314747 138686691792640 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:48:57.314872 138815951419200 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776707315.1425803, 'wait_for_prev_duration_secs': 21.834394693374634, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776707336.9789772, 'checkpointer_blocking_duration_secs': 0.3286399841308594, 'get_old_steps_start_time': 1776707337.3076391, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776707315.1407037, 'checkpoint_manager_blocking_duration_secs': 22.174136638641357}
I0420 17:48:57.314981 138815951419200 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 17:48:57.315023 138815951419200 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:48:57.518344 138688814098176 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9
I0420 17:48:59.367192 138687773927168 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9/items
I0420 17:49:02.946491 138687190447872 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9/items/array_metadatas/process_0
I0420 17:49:42.144118 138686708578048 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.632353s (commit=40.426464s, array_metadata_write=1.205889s)
I0420 17:49:42.145417 138686662428416 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 35.198 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.8770911693573 s) (per-host)
I0420 17:49:42.145543 138686662428416 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.837945s.
I0420 17:49:43.656327 138686662428416 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000429s.
I0420 17:49:43.863554 138686662428416 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 17:49:45.002333 138686662428416 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.977430s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9/items
I0420 17:49:45.005331 138686662428416 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9/items
I0420 17:49:45.521784 138686662428416 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9
I0420 17:49:46.922250 138686662428416 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9`.
I0420 17:49:46.922937 138686662428416 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 49.94 seconds @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_07_eval/checkpoints/9
I0420 17:49:46.924541 138686662428416 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 49.616942s.
I0420 17:49:46.924688 138686691792640 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:49:46.924752 138686691792640 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:49:46.924808 138686691792640 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:49:46.926918 138686691792640 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 17:49:46.927104 138815951419200 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.
I0420 17:49:46.927275 138815951419200 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 17:49:46.928254 138815951419200 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.499, Tokens/s/device: 13942.406, 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: Mon Apr 20 17:49:56 UTC 2026
EXIT_CODE=0