MaxView

Timing Test Reference Updates
‹ 15_ocdbt_false_saveCase: 15_ocdbt_true_resume15_ocdbt_true_save ›

Metrics: main (586e69205) vs feat/nnx-post-train-fixes (7f06c99ac)

Metricmain  586e69205feat/nnx-post-train-fixes  7f06c99acDiff (feat/nnx-post-train-fixes − main)
Parameters1.104 billion1.104 billion—
Final loss5.39105.39100
TFLOP/s92.23292.253+0.021
Tok/s13902.313905.4+3.115
Avg s/step8.4608.354-0.106
Memory %1.381.380
JAX0.9.20.9.2—

Diff = branch value − main value. Green = branch improved. Red = branch regressed.

main  ·  586e69205  ·  main_20260423_071538  ·  full log 
XPK Start: Thu Apr 23 08:24:43 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-23 08:25:08.493467: 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)
I0423 08:25:08.703954 140714551846720 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 08:25:17.746535 140714551846720 distributed.py:149] Starting JAX distributed service on [::]:8482
I0423 08:25:17.748891 140714551846720 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-true-rs-d9i41-slice-job-0-0.mt-15-ocdbt-true-rs-d9i41:8482
I0423 08:25:18.965893 140714551846720 max_utils.py:284] Jax distributed system initialized!
I0423 08:25:24.044172 140714551846720 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 08:25:24.044284 140714551846720 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 08:25:24.044337 140714551846720 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
I0423 08:25:24.044383 140714551846720 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 08:25:24.736042 140714551846720 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 08:25:24.736368 140714551846720 checkpointing.py:677] Setting up checkpoint logger...
I0423 08:25:24.736438 140714551846720 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 08:25:24.736503 140714551846720 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0423 08:25:24.736866 140714551846720 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 0x7ffa08bef380>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 08:25:26.399881 140714551846720 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0423 08:25:26.400261 140714551846720 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 0x7fe6707baea0>}, handler_registry=None
I0423 08:25:26.400684 140714551846720 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fe6707baea0>` 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`.
I0423 08:25:26.400747 140714551846720 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fe6707be870>` 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`.
I0423 08:25:26.400803 140714551846720 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 0x7fe6707baea0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fe6707baea0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fe6707be870>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fe6707be870>}).
I0423 08:25:26.401164 140714551846720 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0423 08:25:26.401259 140714551846720 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7fe670189800> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0423 08:25:28.423829 140596190623488 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776932565295311749, 'commit_timestamp_nsecs': 1776932611532394922, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/9/_CHECKPOINT_METADATA
I0423 08:25:29.018854 140596207408896 checkpoint_manager.py:1871] Missing metrics for step 0
I0423 08:25:29.114918 140714551846720 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints
I0423 08:25:30.261186 140714551846720 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=5), 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_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7fe6707bb380>
I0423 08:25:30.261372 140714551846720 checkpointing.py:301] Checkpoint manager created!
I0423 08:25:31.187525 140714551846720 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 08:25:31.187648 140714551846720 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 08:25:31.566146 140714551846720 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0423 08:25:31.566238 140714551846720 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 08:25:31.582827 140714551846720 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0423 08:25:31.582885 140714551846720 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 08:25:31.606784 140714551846720 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 08:25:31.606852 140714551846720 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 08:25:31.623582 140714551846720 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 08:25:31.623643 140714551846720 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 08:25:31.640255 140714551846720 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 08:25:31.640321 140714551846720 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 08:25:31.665119 140714551846720 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0423 08:25:31.665189 140714551846720 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 08:25:31.686465 140714551846720 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 08:25:31.686526 140714551846720 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 08:25:31.897239 140714551846720 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 08:25:31.897350 140714551846720 checkpointing.py:581] restoring from this run's directory step 9
I0423 08:25:32.152206 140714551846720 event_tracking.py:70] [process=6] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/9.
I0423 08:25:32.412147 140714551846720 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/9.
I0423 08:25:32.412596 140714551846720 event_tracking.py:125] [process=6] [sync] Finished blocking load in 0.26 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/9.
I0423 08:25:34.237133    1409 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 08:25:40.925737 140714551846720 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 236.043 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.694852590560913 s) (per-host)
I0423 08:25:40.931307 140714551846720 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 222.888 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 7.100405216217041 s) (per-host)
I0423 08:25:41.656169 140714551846720 event_tracking.py:138] [process=6] [sync] Finished load in 9.50 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/9
I0423 08:25:41.656327 140714551846720 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints', 'checkpointer_start_time': 1776932732.152088, 'checkpointer_duration_secs': 9.5042085647583, 'checkpoint_manager_start_time': 1776932731.8975968, 'checkpoint_manager_duration_secs': 9.75870156288147}
fsdp: 32
I0423 08:25:42.081601 140714551846720 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0423 08:25:42.081726 140714551846720 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 08:25:42.081778 140714551846720 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 08:25:42.081835 140714551846720 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 08:25:42.081897 140714551846720 maxtext_utils.py:1707]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 08:25:42.081935 140714551846720 maxtext_utils.py:1707]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 08:25:42.081986 140714551846720 maxtext_utils.py:1707]  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)
I0423 08:25:42.082036 140714551846720 maxtext_utils.py:1707]  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')
I0423 08:25:42.082075 140714551846720 maxtext_utils.py:1707]  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)
I0423 08:25:42.082137 140714551846720 maxtext_utils.py:1707]  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)
I0423 08:25:42.082189 140714551846720 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0423 08:25:42.082241 140714551846720 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0423 08:25:42.575443 140714551846720 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 08:25:42.575545 140714551846720 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 08:25:42.591354 140714551846720 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 08:25:42.591412 140714551846720 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 08:25:53.615693 140714551846720 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.
I0423 08:25:53.616461 140714551846720 metric_logger.py:301] number parameters: 1.104 billion
I0423 08:26:04.883043 140714551846720 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0423 08:26:05.090700 140714551846720 checkpointing.py:776] Waited 0.2076435089111328 seconds for step 10 to finish before starting checkpointing.
I0423 08:26:05.093304 140714551846720 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 08:26:05.095324 140714551846720 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0423 08:26:05.097019 140714551846720 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/10.
I0423 08:26:05.837588 140714551846720 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 08:26:05.838554 140714551846720 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 08:26:05.838609 140714551846720 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
I0423 08:26:06.111131 140714551846720 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273621s
I0423 08:26:06.111306 140714551846720 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.557 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27761411666870117 s) (per-host)
I0423 08:26:06.111360 140714551846720 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.277678s (batch_requests_ready=0.002290s, total_serialization_initiated=0.275316s, others=0.000072s)
I0423 08:26:06.111451 140714551846720 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.281850s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.281833)
I0423 08:26:06.112143 140714551846720 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.02 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/10.
I0423 08:26:06.112481 140595603429120 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:46:06.112442
I0423 08:26:06.121619 140714551846720 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 08:26:06.121869 140592373810944 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 08:26:06.121993 140714551846720 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776932765.093286, 'wait_for_prev_duration_secs': 6.079673767089844e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776932765.095365, 'checkpointer_blocking_duration_secs': 1.0172700881958008, 'get_old_steps_start_time': 1776932766.112658, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776932765.0911705, 'checkpoint_manager_blocking_duration_secs': 1.0307867527008057}
I0423 08:26:06.122108 140714551846720 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0423 08:26:06.122164 140714551846720 max_utils.py:750] 
Memstats: After params initialized:
I0423 08:26:06.122213 140714551846720 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0423 08:26:06.122246 140714551846720 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0423 08:26:06.122274 140714551846720 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0423 08:26:06.122299 140714551846720 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0423 08:26:06.431662 140714551846720 metric_logger.py:196] completed step: 10, seconds: 11.266, TFLOP/s/device: 1.206, Tokens/s/device: 181.779, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0423 08:26:07.021324 140714551846720 metric_logger.py:196] completed step: 11, seconds: 1.547, TFLOP/s/device: 8.783, Tokens/s/device: 1323.836, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0423 08:26:07.168718 140714551846720 metric_logger.py:196] completed step: 12, seconds: 0.443, TFLOP/s/device: 30.640, Tokens/s/device: 4618.417, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0423 08:26:07.316288 140714551846720 metric_logger.py:196] completed step: 13, seconds: 0.152, TFLOP/s/device: 89.375, Tokens/s/device: 13471.557, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0423 08:26:07.320601 140714551846720 checkpointing.py:772] Waiting for step 15 to finish before checkpoint...
I0423 08:26:07.610443 140714551846720 checkpointing.py:776] Waited 0.2898092269897461 seconds for step 15 to finish before starting checkpointing.
I0423 08:26:07.613471 140714551846720 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 08:26:11.442010 140592382203648 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/10/items/array_metadatas/process_6
I0423 08:26:47.782437 140595603429120 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.655 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.9487190246582 s) (per-host)
I0423 08:26:47.782541 140595603429120 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.669952s.
I0423 08:26:56.214391 140595603429120 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.101784s.
I0423 08:26:56.214704 140592373810944 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 08:26:56.214826 140592373810944 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 08:26:56.214874 140592373810944 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0423 08:26:56.218102 140592373810944 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0423 08:26:56.218280 140714551846720 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0423 08:26:56.218412 140714551846720 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.604940 seconds. If this number is high, consider checkpointing less frequently.
I0423 08:26:56.220532 140714551846720 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15
I0423 08:26:56.222846 140714551846720 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/15.
I0423 08:26:56.544971 140714551846720 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 08:26:56.545066 140714551846720 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
I0423 08:26:56.587674 140714551846720 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.043815s
I0423 08:26:56.587841 140714551846720 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 32.576 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.047353506088256836 s) (per-host)
I0423 08:26:56.587890 140714551846720 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.047412s (batch_requests_ready=0.001910s, total_serialization_initiated=0.045436s, others=0.000067s)
I0423 08:26:56.587979 140714551846720 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.052270s (all_items=0.000014s, per_item={'items': '0.00001383'}, temp_paths=0.052257)
I0423 08:26:56.588722 140714551846720 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.37 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/15.
I0423 08:26:56.589144 140592373810944 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:46:56.589074
I0423 08:26:56.591132 140714551846720 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 08:26:56.591413 140589706241792 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 08:26:56.591547 140714551846720 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776932767.6134405, 'wait_for_prev_duration_secs': 48.60494017601013, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776932816.2205718, 'checkpointer_blocking_duration_secs': 0.36872053146362305, 'get_old_steps_start_time': 1776932816.5893185, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776932767.6109831, 'checkpoint_manager_blocking_duration_secs': 48.980531454086304}
I0423 08:26:56.591716 140714551846720 checkpointing.py:408] Started an asynchronous checkpoint save for step 15
I0423 08:26:56.592484 140714551846720 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 92.014, Tokens/s/device: 13869.419, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0423 08:26:56.598428 140714551846720 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.227, Tokens/s/device: 13901.522, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0423 08:27:02.844535 140592382203648 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/15/items/array_metadatas/process_6
I0423 08:27:18.066896 140714551846720 metric_logger.py:196] completed step: 16, seconds: 49.276, TFLOP/s/device: 0.276, Tokens/s/device: 41.562, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0423 08:27:18.214176 140714551846720 metric_logger.py:196] completed step: 17, seconds: 21.322, TFLOP/s/device: 0.637, Tokens/s/device: 96.051, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0423 08:27:18.361487 140714551846720 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.205, Tokens/s/device: 13445.908, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0423 08:27:18.508058 140714551846720 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0423 08:27:18.508758 140714551846720 checkpointing.py:776] Waited 0.0007212162017822266 seconds for step 19 to finish before starting checkpointing.
I0423 08:27:18.510892 140714551846720 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 08:27:39.510528 140592373810944 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.760 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.97000026702881 s) (per-host)
I0423 08:27:39.510654 140592373810944 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.921390s.
I0423 08:27:48.978983 140592373810944 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 52.389706s.
I0423 08:27:48.979262 140589706241792 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 08:27:48.979396 140589706241792 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 08:27:48.979444 140589706241792 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0423 08:27:48.980898 140589706241792 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts.
I0423 08:27:48.981066 140714551846720 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15.
W0423 08:27:48.981208 140714551846720 checkpoint_manager.py:1452] Waiting for previous save to complete took 30.470323 seconds. If this number is high, consider checkpointing less frequently.
I0423 08:27:48.982739 140714551846720 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0423 08:27:48.984830 140714551846720 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/19.
I0423 08:27:49.294409 140714551846720 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 08:27:49.294504 140714551846720 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
I0423 08:27:49.334792 140714551846720 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041247s
I0423 08:27:49.334954 140714551846720 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 34.430 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04480385780334473 s) (per-host)
I0423 08:27:49.335003 140714551846720 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044861s (batch_requests_ready=0.001761s, total_serialization_initiated=0.043037s, others=0.000063s)
I0423 08:27:49.335108 140714551846720 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048883s (all_items=0.000012s, per_item={'items': '0.00001216'}, temp_paths=0.048871)
I0423 08:27:49.335809 140714551846720 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/19.
I0423 08:27:49.336136 140589706241792 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:47:49.336088
I0423 08:27:49.338074 140714551846720 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 08:27:49.338320 140592382203648 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 08:27:49.338473 140714551846720 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776932838.5108616, 'wait_for_prev_duration_secs': 30.470323085784912, 'time_between_consecutive_saves_sec': 22.292718172073364, 'checkpointer_blocking_start_time': 1776932868.9827762, 'checkpointer_blocking_duration_secs': 0.35346245765686035, 'get_old_steps_start_time': 1776932869.336261, 'get_old_steps_duration_secs': 2.5033950805664062e-05, 'checkpoint_manager_blocking_start_time': 1776932838.5089865, 'checkpoint_manager_blocking_duration_secs': 30.829455137252808}
I0423 08:27:49.338585 140714551846720 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0423 08:27:49.338626 140714551846720 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 08:27:54.958194 140592365418240 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_true/checkpoints/19/items/array_metadatas/process_6
I0423 08:28:31.848119 140589706241792 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.116 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.55791616439819 s) (per-host)
I0423 08:28:31.848235 140589706241792 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.511495s.
I0423 08:28:39.539326 140589706241792 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.202570s.
I0423 08:28:39.539684 140592382203648 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 08:28:39.539805 140592382203648 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 08:28:39.539847 140592382203648 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0423 08:28:39.541255 140592382203648 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0423 08:28:39.541418 140714551846720 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0423 08:28:39.541566 140714551846720 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 08:28:39.542471 140714551846720 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.232, Tokens/s/device: 13902.277, total_weights: 65536, loss: 5.391, lm_loss: 5.391, perplexity: 219.349
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Thu Apr 23 08:28:48 UTC 2026
EXIT_CODE=0
feat/nnx-post-train-fixes  ·  7f06c99ac  ·  feat_nnx_post_train_fixes_20260423_124541  ·  full log 
XPK Start: Thu Apr 23 13:51:02 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-23 13:51:27.615013: 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)
I0423 13:51:27.824429 139565630281536 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 13:51:36.866982 139565630281536 distributed.py:149] Starting JAX distributed service on [::]:8482
I0423 13:51:36.869359 139565630281536 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-true-rs-3jv8p-slice-job-0-0.mt-15-ocdbt-true-rs-3jv8p:8482
I0423 13:51:37.969548 139565630281536 max_utils.py:284] Jax distributed system initialized!
I0423 13:51:44.231193 139565630281536 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 13:51:44.231300 139565630281536 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 13:51:44.231340 139565630281536 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
I0423 13:51:44.231376 139565630281536 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 13:51:44.927138 139565630281536 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 13:51:44.927728 139565630281536 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 13:51:44.927908 139565630281536 checkpointing.py:688] Setting up checkpoint logger...
I0423 13:51:44.927957 139565630281536 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 13:51:44.928002 139565630281536 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0423 13:51:44.928355 139565630281536 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 0x7eee70450260>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 13:51:46.649690 139565630281536 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0423 13:51:46.649928 139565630281536 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 0x7eda601a23f0>}, handler_registry=None
I0423 13:51:46.650178 139565630281536 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7eda601a23f0>` 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`.
I0423 13:51:46.650228 139565630281536 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eda601a49e0>` 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`.
I0423 13:51:46.650266 139565630281536 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 0x7eda601a23f0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7eda601a23f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eda601a49e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eda601a49e0>}).
I0423 13:51:46.650608 139565630281536 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0423 13:51:46.650679 139565630281536 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7eda38379d00> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0423 13:51:48.114146 139444057253632 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776952044397212372, 'commit_timestamp_nsecs': 1776952087198627892, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA
I0423 13:51:48.363777 139444057253632 checkpoint_manager.py:1871] Missing metrics for step 0
I0423 13:51:48.913280 139565630281536 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints
I0423 13:51:49.224922 139565630281536 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=5), 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_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7eda601a2c60>
I0423 13:51:49.225116 139565630281536 checkpointing.py:302] Checkpoint manager created!
I0423 13:51:50.172576 139565630281536 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 13:51:50.172687 139565630281536 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 13:51:50.556408 139565630281536 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 13:51:50.556501 139565630281536 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 13:51:50.572982 139565630281536 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 13:51:50.573040 139565630281536 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 13:51:50.597013 139565630281536 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 13:51:50.597085 139565630281536 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 13:51:50.613715 139565630281536 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 13:51:50.613779 139565630281536 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 13:51:50.630346 139565630281536 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 13:51:50.630409 139565630281536 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 13:51:50.661654 139565630281536 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 13:51:50.661762 139565630281536 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 13:51:50.683431 139565630281536 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 13:51:50.683500 139565630281536 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 13:51:50.913700 139565630281536 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 13:51:50.913814 139565630281536 checkpointing.py:582] restoring from this run's directory step 9
I0423 13:51:51.153398 139565630281536 event_tracking.py:70] [process=6] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/9.
I0423 13:51:51.397966 139565630281536 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/9.
I0423 13:51:51.398452 139565630281536 event_tracking.py:125] [process=6] [sync] Finished blocking load in 0.25 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/9.
I0423 13:51:52.804640    1469 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 13:52:00.971351 139565630281536 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 193.361 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 8.172668695449829 s) (per-host)
I0423 13:52:00.976980 139565630281536 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 184.491 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 8.578146696090698 s) (per-host)
I0423 13:52:00.979156 139565630281536 event_tracking.py:138] [process=6] [sync] Finished load in 9.83 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/9
I0423 13:52:00.979252 139565630281536 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints', 'checkpointer_start_time': 1776952311.1532986, 'checkpointer_duration_secs': 9.825926065444946, 'checkpoint_manager_start_time': 1776952310.91406, 'checkpoint_manager_duration_secs': 10.065165758132935}
fsdp: 32
I0423 13:52:01.398682 139565630281536 maxtext_utils.py:1880]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0423 13:52:01.398816 139565630281536 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 13:52:01.398871 139565630281536 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 13:52:01.398928 139565630281536 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 13:52:01.398978 139565630281536 maxtext_utils.py:1880]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 13:52:01.399014 139565630281536 maxtext_utils.py:1880]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 13:52:01.399065 139565630281536 maxtext_utils.py:1880]  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)
I0423 13:52:01.399143 139565630281536 maxtext_utils.py:1880]  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')
I0423 13:52:01.399191 139565630281536 maxtext_utils.py:1880]  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)
I0423 13:52:01.399228 139565630281536 maxtext_utils.py:1880]  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)
I0423 13:52:01.399277 139565630281536 maxtext_utils.py:1880]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0423 13:52:01.399331 139565630281536 maxtext_utils.py:1880]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0423 13:52:01.891791 139565630281536 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 13:52:01.891884 139565630281536 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 13:52:01.907495 139565630281536 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 13:52:01.907556 139565630281536 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 13:52:12.956663 139565630281536 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.
I0423 13:52:12.957500 139565630281536 metric_logger.py:301] number parameters: 1.104 billion
I0423 13:52:24.185305 139565630281536 checkpointing.py:794] Waiting for step 10 to finish before checkpoint...
I0423 13:52:24.341610 139565630281536 checkpointing.py:798] Waited 0.15628409385681152 seconds for step 10 to finish before starting checkpointing.
I0423 13:52:24.343961 139565630281536 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 13:52:24.345824 139565630281536 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0423 13:52:24.347216 139565630281536 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/10.
I0423 13:52:24.673475 139565630281536 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 13:52:24.674609 139565630281536 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 13:52:24.674668 139565630281536 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
I0423 13:52:24.945861 139565630281536 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.272457s
I0423 13:52:24.946038 139565630281536 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.583 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27628016471862793 s) (per-host)
I0423 13:52:24.946108 139565630281536 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.276346s (batch_requests_ready=0.002132s, total_serialization_initiated=0.274139s, others=0.000075s)
I0423 13:52:24.946212 139565630281536 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.280610s (all_items=0.000016s, per_item={'items': '0.00001645'}, temp_paths=0.280593)
I0423 13:52:24.946946 139565630281536 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.60 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/10.
I0423 13:52:24.947323 139443453273856 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:12:24.947284
I0423 13:52:24.961760 139565630281536 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 13:52:24.962066 139440223655680 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 13:52:24.962234 139565630281536 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776952344.3439429, 'wait_for_prev_duration_secs': 6.127357482910156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776952344.3458622, 'checkpointer_blocking_duration_secs': 0.6016128063201904, 'get_old_steps_start_time': 1776952344.947502, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776952344.3421438, 'checkpoint_manager_blocking_duration_secs': 0.6200547218322754}
I0423 13:52:24.962339 139565630281536 checkpointing.py:409] Started an asynchronous checkpoint save for step 10
I0423 13:52:24.962389 139565630281536 max_utils.py:750] 
Memstats: After params initialized:
I0423 13:52:24.962439 139565630281536 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0423 13:52:24.962471 139565630281536 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0423 13:52:24.962497 139565630281536 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0423 13:52:24.962522 139565630281536 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0423 13:52:25.278694 139565630281536 metric_logger.py:196] completed step: 10, seconds: 11.228, TFLOP/s/device: 1.210, Tokens/s/device: 182.406, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0423 13:52:25.884622 139565630281536 metric_logger.py:196] completed step: 11, seconds: 1.092, TFLOP/s/device: 12.444, Tokens/s/device: 1875.645, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0423 13:52:26.032136 139565630281536 metric_logger.py:196] completed step: 12, seconds: 0.459, TFLOP/s/device: 29.629, Tokens/s/device: 4465.941, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0423 13:52:26.179581 139565630281536 metric_logger.py:196] completed step: 13, seconds: 0.153, TFLOP/s/device: 88.572, Tokens/s/device: 13350.543, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0423 13:52:26.184043 139565630281536 checkpointing.py:794] Waiting for step 15 to finish before checkpoint...
I0423 13:52:26.473486 139565630281536 checkpointing.py:798] Waited 0.28940463066101074 seconds for step 15 to finish before starting checkpointing.
I0423 13:52:26.476659 139565630281536 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 13:52:30.315398 139442379532032 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/10/items/array_metadatas/process_6
I0423 13:53:06.821088 139443453273856 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.474 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.15129375457764 s) (per-host)
I0423 13:53:06.821261 139443453273856 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.873811s.
I0423 13:53:14.067178 139443453273856 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.119727s.
I0423 13:53:14.067488 139440223655680 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 13:53:14.067604 139440223655680 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 13:53:14.067652 139440223655680 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0423 13:53:14.068963 139440223655680 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0423 13:53:14.069172 139565630281536 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0423 13:53:14.069300 139565630281536 checkpoint_manager.py:1452] Waiting for previous save to complete took 47.592643 seconds. If this number is high, consider checkpointing less frequently.
I0423 13:53:14.071427 139565630281536 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15
I0423 13:53:14.074115 139565630281536 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/15.
I0423 13:53:14.813670 139565630281536 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 13:53:14.813763 139565630281536 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
I0423 13:53:14.841886 139565630281536 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.029358s
I0423 13:53:14.842051 139565630281536 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 47.088 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.032759666442871094 s) (per-host)
I0423 13:53:14.842116 139565630281536 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.032818s (batch_requests_ready=0.001748s, total_serialization_initiated=0.031004s, others=0.000066s)
I0423 13:53:14.842212 139565630281536 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.038044s (all_items=0.000013s, per_item={'items': '0.00001287'}, temp_paths=0.038031)
I0423 13:53:14.842862 139565630281536 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/15.
I0423 13:53:14.843209 139440223655680 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:13:14.843173
I0423 13:53:14.854008 139565630281536 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 13:53:14.854255 139438730508032 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 13:53:14.854384 139565630281536 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776952346.4766297, 'wait_for_prev_duration_secs': 47.59264302253723, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776952394.0714664, 'checkpointer_blocking_duration_secs': 0.7718896865844727, 'get_old_steps_start_time': 1776952394.8433814, 'get_old_steps_duration_secs': 3.24249267578125e-05, 'checkpoint_manager_blocking_start_time': 1776952346.4741516, 'checkpoint_manager_blocking_duration_secs': 48.38019943237305}
I0423 13:53:14.854499 139565630281536 checkpointing.py:409] Started an asynchronous checkpoint save for step 15
I0423 13:53:14.855283 139565630281536 metric_logger.py:196] completed step: 14, seconds: 0.147, TFLOP/s/device: 92.150, Tokens/s/device: 13889.925, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0423 13:53:14.861159 139565630281536 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.228, Tokens/s/device: 13901.616, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0423 13:53:20.875827 139439697417984 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/15/items/array_metadatas/process_6
I0423 13:53:36.348064 139565630281536 metric_logger.py:196] completed step: 16, seconds: 48.676, TFLOP/s/device: 0.279, Tokens/s/device: 42.074, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0423 13:53:36.495412 139565630281536 metric_logger.py:196] completed step: 17, seconds: 21.340, TFLOP/s/device: 0.637, Tokens/s/device: 95.968, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0423 13:53:36.642766 139565630281536 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.125, Tokens/s/device: 13433.825, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0423 13:53:36.789441 139565630281536 checkpointing.py:794] Waiting for step 19 to finish before checkpoint...
I0423 13:53:36.790176 139565630281536 checkpointing.py:798] Waited 0.00075531005859375 seconds for step 19 to finish before starting checkpointing.
I0423 13:53:36.792166 139565630281536 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 13:53:56.958234 139440223655680 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.477 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.14890193939209 s) (per-host)
I0423 13:53:56.958380 139440223655680 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.115051s.
I0423 13:54:05.586734 139440223655680 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.743389s.
I0423 13:54:05.587025 139438730508032 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 13:54:05.587160 139438730508032 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 13:54:05.587209 139438730508032 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0423 13:54:05.588701 139438730508032 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts.
I0423 13:54:05.588878 139565630281536 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15.
W0423 13:54:05.588999 139565630281536 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.796841 seconds. If this number is high, consider checkpointing less frequently.
I0423 13:54:05.590407 139565630281536 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0423 13:54:05.592804 139565630281536 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/19.
I0423 13:54:05.901781 139565630281536 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 13:54:05.901874 139565630281536 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
I0423 13:54:05.927408 139565630281536 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026469s
I0423 13:54:05.927534 139565630281536 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 51.605 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.0298917293548584 s) (per-host)
I0423 13:54:05.927579 139565630281536 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.029946s (batch_requests_ready=0.001672s, total_serialization_initiated=0.028215s, others=0.000059s)
I0423 13:54:05.927650 139565630281536 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.034044s (all_items=0.000011s, per_item={'items': '0.00001121'}, temp_paths=0.034033)
I0423 13:54:05.928385 139565630281536 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/19.
I0423 13:54:05.928730 139438730508032 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:14:05.928695
I0423 13:54:05.932259 139565630281536 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 13:54:05.932547 139439697417984 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 13:54:05.932711 139565630281536 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776952416.7921357, 'wait_for_prev_duration_secs': 28.79684066772461, 'time_between_consecutive_saves_sec': 22.723135709762573, 'checkpointer_blocking_start_time': 1776952445.5904477, 'checkpointer_blocking_duration_secs': 0.3384237289428711, 'get_old_steps_start_time': 1776952445.9288907, 'get_old_steps_duration_secs': 2.6464462280273438e-05, 'checkpoint_manager_blocking_start_time': 1776952416.7904172, 'checkpoint_manager_blocking_duration_secs': 29.14226245880127}
I0423 13:54:05.932815 139565630281536 checkpointing.py:409] Started an asynchronous checkpoint save for step 19
I0423 13:54:05.932857 139565630281536 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 13:54:11.082988 139437001615104 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_true/checkpoints/19/items/array_metadatas/process_6
I0423 13:54:47.382688 139438730508032 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.076 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.48500633239746 s) (per-host)
I0423 13:54:47.382810 139438730508032 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.453968s.
I0423 13:54:55.405759 139438730508032 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.476900s.
I0423 13:54:55.406039 139439697417984 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 13:54:55.406202 139439697417984 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 13:54:55.406249 139439697417984 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0423 13:54:55.408372 139439697417984 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0423 13:54:55.408611 139565630281536 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0423 13:54:55.408757 139565630281536 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 13:54:55.409835 139565630281536 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.253, Tokens/s/device: 13905.392, total_weights: 65536, loss: 5.391, lm_loss: 5.391, perplexity: 219.349
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Thu Apr 23 13:55:07 UTC 2026
EXIT_CODE=0