MaxView

Timing Test Reference Updates
‹ 07_evalCase: 08_checkpoint_async_true_resume08_checkpoint_async_true_save ›

Metrics: main (b117f50cf) vs main (59e0f1759)

Metricmain  b117f50cfmain  59e0f1759Diff (main − main)
Parameters1.104 billion1.104 billion—
Final loss5.39105.39100
TFLOP/s92.14592.001-0.144
Tok/s13889.213867.4-21.818
Avg s/step8.4018.365-0.036
Memory %1.381.380
JAX0.9.20.9.2—

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

main  ·  b117f50cf  ·  main_20260424_070227  ·  full log 
XPK Start: Fri Apr 24 07:30:36 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. 
2026-04-24 07:31:01.384448: 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)
I0424 07:31:01.595814 136980369393472 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 07:31:10.635934 136980369393472 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 07:31:10.638240 136980369393472 distributed.py:172] Connecting to JAX distributed service on mt-08-checkpoint-async--cejvn-slice-job-0-0.mt-08-checkpoint-async--cejvn:8482
I0424 07:31:11.684225 136980369393472 max_utils.py:284] Jax distributed system initialized!
I0424 07:31:17.602452 136980369393472 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 07:31:17.602559 136980369393472 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 07:31:17.602598 136980369393472 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
I0424 07:31:17.602634 136980369393472 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 07:31:18.293917 136980369393472 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 07:31:18.294197 136980369393472 checkpointing.py:677] Setting up checkpoint logger...
I0424 07:31:18.294250 136980369393472 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 07:31:18.294295 136980369393472 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 07:31:18.294629 136980369393472 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 0x7c9497510200>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 07:31:19.557190 136980369393472 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0424 07:31:19.557428 136980369393472 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c810438eb40>}, handler_registry=None
I0424 07:31:19.557684 136980369393472 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c810438eb40>` 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`.
I0424 07:31:19.557734 136980369393472 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c8104231250>` 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`.
I0424 07:31:19.557771 136980369393472 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 0x7c810438eb40>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c810438eb40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c8104231250>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c8104231250>}).
I0424 07:31:19.558089 136980369393472 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 07:31:19.558157 136980369393472 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c803457d580> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 07:31:21.184610 136862907733760 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777015716831445276, 'commit_timestamp_nsecs': 1777015761727898639, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/9/_CHECKPOINT_METADATA
I0424 07:31:21.431775 136862907733760 checkpoint_manager.py:1871] Missing metrics for step 9
I0424 07:31:21.467833 136980369393472 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints
I0424 07:31:21.876595 136980369393472 checkpoint_manager.py:929] [process=5][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_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c810422eb10>
I0424 07:31:21.876802 136980369393472 checkpointing.py:301] Checkpoint manager created!
I0424 07:31:22.810427 136980369393472 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 07:31:22.810545 136980369393472 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:31:23.191558 136980369393472 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 07:31:23.191664 136980369393472 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:31:23.207943 136980369393472 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 07:31:23.208003 136980369393472 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:31:23.232154 136980369393472 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 07:31:23.232220 136980369393472 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:31:23.248808 136980369393472 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 07:31:23.248868 136980369393472 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:31:23.265215 136980369393472 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 07:31:23.265279 136980369393472 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:31:23.289998 136980369393472 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0424 07:31:23.290070 136980369393472 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:31:23.311427 136980369393472 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 07:31:23.311496 136980369393472 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 07:31:23.525327 136980369393472 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 07:31:23.525442 136980369393472 checkpointing.py:581] restoring from this run's directory step 9
I0424 07:31:23.783023 136980369393472 event_tracking.py:70] [process=5] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/9.
I0424 07:31:24.031473 136980369393472 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/9.
I0424 07:31:24.031938 136980369393472 event_tracking.py:125] [process=5] [sync] Finished blocking load in 0.25 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/9.
I0424 07:31:25.425582    1411 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 07:31:31.584236 136980369393472 jax_array_handlers.py:843] [process=5] /jax/orbax/read/worker/io/requested throughput: 256.359 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.164309501647949 s) (per-host)
I0424 07:31:31.589983 136980369393472 base_pytree_checkpoint_handler.py:130] [process=5] /jax/checkpoint/read/gbytes_per_sec: 240.998 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.566833734512329 s) (per-host)
I0424 07:31:32.052338 136980369393472 event_tracking.py:138] [process=5] [sync] Finished load in 8.27 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/9
I0424 07:31:32.052493 136980369393472 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints', 'checkpointer_start_time': 1777015883.7829218, 'checkpointer_duration_secs': 8.269538640975952, 'checkpoint_manager_start_time': 1777015883.525702, 'checkpoint_manager_duration_secs': 8.526761054992676}
fsdp: 32
I0424 07:31:32.470428 136980369393472 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 07:31:32.470553 136980369393472 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)
I0424 07:31:32.470607 136980369393472 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)
I0424 07:31:32.470679 136980369393472 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')
I0424 07:31:32.470738 136980369393472 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)
I0424 07:31:32.470776 136980369393472 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)
I0424 07:31:32.470828 136980369393472 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)
I0424 07:31:32.470880 136980369393472 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')
I0424 07:31:32.470919 136980369393472 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)
I0424 07:31:32.470955 136980369393472 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)
I0424 07:31:32.471000 136980369393472 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0424 07:31:32.471046 136980369393472 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0424 07:31:32.957794 136980369393472 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 07:31:32.957885 136980369393472 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 07:31:32.973410 136980369393472 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 07:31:32.973469 136980369393472 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 07:31:44.044034 136980369393472 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.
I0424 07:31:44.044835 136980369393472 metric_logger.py:301] number parameters: 1.104 billion
I0424 07:31:55.226280 136980369393472 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0424 07:31:55.646880 136980369393472 checkpointing.py:776] Waited 0.420579195022583 seconds for step 10 to finish before starting checkpointing.
I0424 07:31:55.649422 136980369393472 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:31:55.651225 136980369393472 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 10
I0424 07:31:55.652818 136980369393472 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/10.
I0424 07:31:55.988033 136980369393472 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 07:31:55.988947 136980369393472 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 07:31:55.989005 136980369393472 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
I0424 07:31:56.263694 136980369393472 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.275735s
I0424 07:31:56.263877 136980369393472 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.516 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2796518802642822 s) (per-host)
I0424 07:31:56.263931 136980369393472 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279718s (batch_requests_ready=0.002200s, total_serialization_initiated=0.277442s, others=0.000077s)
I0424 07:31:56.264024 136980369393472 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283745s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.283727)
I0424 07:31:56.264598 136980369393472 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/10.
I0424 07:31:56.264848 136860206626560 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 07:51:56.264806
I0424 07:31:56.272843 136980369393472 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 07:31:56.273137 136862320539392 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 07:31:56.273288 136980369393472 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777015915.6494033, 'wait_for_prev_duration_secs': 6.127357482910156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777015915.651266, 'checkpointer_blocking_duration_secs': 0.613750696182251, 'get_old_steps_start_time': 1777015916.265039, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1777015915.6474411, 'checkpoint_manager_blocking_duration_secs': 0.6258132457733154}
I0424 07:31:56.273391 136980369393472 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0424 07:31:56.273452 136980369393472 max_utils.py:750] 
Memstats: After params initialized:
I0424 07:31:56.273505 136980369393472 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0))
I0424 07:31:56.273538 136980369393472 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0))
I0424 07:31:56.273565 136980369393472 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0))
I0424 07:31:56.273591 136980369393472 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0))
I0424 07:31:56.584820 136980369393472 metric_logger.py:196] completed step: 10, seconds: 11.181, TFLOP/s/device: 1.215, Tokens/s/device: 183.162, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0424 07:31:57.172387 136980369393472 metric_logger.py:196] completed step: 11, seconds: 1.357, TFLOP/s/device: 10.012, Tokens/s/device: 1509.173, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0424 07:31:57.319709 136980369393472 metric_logger.py:196] completed step: 12, seconds: 0.441, TFLOP/s/device: 30.817, Tokens/s/device: 4645.055, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0424 07:31:57.467309 136980369393472 metric_logger.py:196] completed step: 13, seconds: 0.153, TFLOP/s/device: 89.062, Tokens/s/device: 13424.403, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0424 07:31:57.471920 136980369393472 checkpointing.py:772] Waiting for step 15 to finish before checkpoint...
I0424 07:31:57.761626 136980369393472 checkpointing.py:776] Waited 0.2896716594696045 seconds for step 15 to finish before starting checkpointing.
I0424 07:31:57.764874 136980369393472 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 07:32:01.382457 136861246797568 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/10/items/array_metadatas/process_5
I0424 07:32:38.736590 136860206626560 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 36.948 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.752328872680664 s) (per-host)
I0424 07:32:38.736738 136860206626560 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.471769s.
I0424 07:32:46.522500 136860206626560 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.257514s.
I0424 07:32:46.522825 136862320539392 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 07:32:46.522949 136862320539392 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 07:32:46.522997 136862320539392 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0424 07:32:46.525684 136862320539392 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0424 07:32:46.525858 136980369393472 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0424 07:32:46.525989 136980369393472 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.761114 seconds. If this number is high, consider checkpointing less frequently.
I0424 07:32:46.528066 136980369393472 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 15
I0424 07:32:46.530568 136980369393472 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/15.
I0424 07:32:47.074904 136980369393472 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 07:32:47.075011 136980369393472 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
I0424 07:32:47.116086 136980369393472 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042291s
I0424 07:32:47.116233 136980369393472 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 33.745 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04571247100830078 s) (per-host)
I0424 07:32:47.116282 136980369393472 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045771s (batch_requests_ready=0.001765s, total_serialization_initiated=0.043941s, others=0.000065s)
I0424 07:32:47.116370 136980369393472 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050755s (all_items=0.000013s, per_item={'items': '0.00001335'}, temp_paths=0.050742)
I0424 07:32:47.117002 136980369393472 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.59 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/15.
I0424 07:32:47.117284 136862320539392 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 07:52:47.117261
I0424 07:32:47.119183 136980369393472 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 07:32:47.119396 136858059142912 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 07:32:47.119515 136980369393472 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777015917.764845, 'wait_for_prev_duration_secs': 48.76111435890198, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777015966.5281038, 'checkpointer_blocking_duration_secs': 0.5892789363861084, 'get_old_steps_start_time': 1777015967.1174073, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1777015917.762289, 'checkpoint_manager_blocking_duration_secs': 49.357192516326904}
I0424 07:32:47.119621 136980369393472 checkpointing.py:408] Started an asynchronous checkpoint save for step 15
I0424 07:32:47.120437 136980369393472 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 92.017, Tokens/s/device: 13869.795, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0424 07:32:47.126490 136980369393472 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.192, Tokens/s/device: 13896.240, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0424 07:32:52.240301 136861246797568 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/15/items/array_metadatas/process_5
I0424 07:33:07.902184 136980369393472 metric_logger.py:196] completed step: 16, seconds: 49.653, TFLOP/s/device: 0.274, Tokens/s/device: 41.246, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0424 07:33:08.049815 136980369393472 metric_logger.py:196] completed step: 17, seconds: 20.630, TFLOP/s/device: 0.659, Tokens/s/device: 99.275, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0424 07:33:08.197209 136980369393472 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.257, Tokens/s/device: 13453.857, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0424 07:33:08.343857 136980369393472 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0424 07:33:08.344555 136980369393472 checkpointing.py:776] Waited 0.000720977783203125 seconds for step 19 to finish before starting checkpointing.
I0424 07:33:08.346668 136980369393472 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 07:33:29.751693 136862320539392 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.009 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.681114196777344 s) (per-host)
I0424 07:33:29.751805 136862320539392 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.634444s.
I0424 07:33:36.992799 136862320539392 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 49.875421s.
I0424 07:33:36.993042 136858059142912 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 07:33:36.993103 136858059142912 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 07:33:36.993151 136858059142912 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0424 07:33:36.994741 136858059142912 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts.
I0424 07:33:36.994904 136980369393472 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15.
W0424 07:33:36.995039 136980369393472 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.648392 seconds. If this number is high, consider checkpointing less frequently.
I0424 07:33:36.996526 136980369393472 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 19
I0424 07:33:36.998426 136980369393472 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/19.
I0424 07:33:37.299405 136980369393472 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 07:33:37.299499 136980369393472 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
I0424 07:33:37.342690 136980369393472 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.044067s
I0424 07:33:37.342856 136980369393472 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 32.451 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04753565788269043 s) (per-host)
I0424 07:33:37.342904 136980369393472 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.047595s (batch_requests_ready=0.001654s, total_serialization_initiated=0.045877s, others=0.000064s)
I0424 07:33:37.342993 136980369393472 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.051660s (all_items=0.000010s, per_item={'items': '0.00001049'}, temp_paths=0.051649)
I0424 07:33:37.343713 136980369393472 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/19.
I0424 07:33:37.344065 136858059142912 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 07:53:37.344028
I0424 07:33:37.346034 136980369393472 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 07:33:37.346321 136857497093888 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 07:33:37.346453 136980369393472 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777015988.3466225, 'wait_for_prev_duration_secs': 28.648391723632812, 'time_between_consecutive_saves_sec': 21.820897579193115, 'checkpointer_blocking_start_time': 1777016016.9965656, 'checkpointer_blocking_duration_secs': 0.34764695167541504, 'get_old_steps_start_time': 1777016017.3442311, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1777015988.344802, 'checkpoint_manager_blocking_duration_secs': 29.001619577407837}
I0424 07:33:37.346610 136980369393472 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0424 07:33:37.346663 136980369393472 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 07:33:42.591087 136861246797568 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_08_checkpoint_async_true/checkpoints/19/items/array_metadatas/process_5
I0424 07:34:19.399309 136858059142912 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.517 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.103941917419434 s) (per-host)
I0424 07:34:19.399431 136858059142912 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.055255s.
I0424 07:34:25.973201 136858059142912 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 48.629003s.
I0424 07:34:25.973493 136857497093888 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 07:34:25.973611 136857497093888 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 07:34:25.973667 136857497093888 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0424 07:34:25.975240 136857497093888 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0424 07:34:25.975421 136980369393472 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0424 07:34:25.975574 136980369393472 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:34:25.976531 136980369393472 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.145, Tokens/s/device: 13889.171, 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: Fri Apr 24 07:34:37 UTC 2026
EXIT_CODE=0
main  ·  59e0f1759  ·  main_20260425_071456  ·  full log 
XPK Start: Sat Apr 25 07:38:43 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. 
2026-04-25 07:39:08.214704: 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)
I0425 07:39:08.424495 138952591587136 max_utils.py:273] Attempting to initialize the jax distributed system...
I0425 07:39:17.464858 138952591587136 distributed.py:149] Starting JAX distributed service on [::]:8482
I0425 07:39:17.467160 138952591587136 distributed.py:172] Connecting to JAX distributed service on mt-08-checkpoint-async--i4ahv-slice-job-0-0.mt-08-checkpoint-async--i4ahv:8482
I0425 07:39:18.927439 138952591587136 max_utils.py:284] Jax distributed system initialized!
I0425 07:39:24.933264 138952591587136 max_utils.py:800] System Information: Jax Version: 0.9.2
I0425 07:39:24.933371 138952591587136 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0425 07:39:24.933412 138952591587136 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
I0425 07:39:24.933452 138952591587136 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0425 07:39:25.629763 138952591587136 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0425 07:39:25.630045 138952591587136 checkpointing.py:677] Setting up checkpoint logger...
I0425 07:39:25.630100 138952591587136 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0425 07:39:25.630144 138952591587136 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0425 07:39:25.630481 138952591587136 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 0x7e5fcad18950>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0425 07:39:26.884775 138952591587136 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0425 07:39:26.885016 138952591587136 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4b843199a0>}, handler_registry=None
I0425 07:39:26.885252 138952591587136 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4b843199a0>` 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`.
I0425 07:39:26.885299 138952591587136 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4b8431d430>` 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`.
I0425 07:39:26.885335 138952591587136 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 0x7e4b843199a0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4b843199a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4b8431d430>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4b8431d430>}).
I0425 07:39:26.885661 138952591587136 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0425 07:39:26.885732 138952591587136 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e4b6c7c5800> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0425 07:39:28.391002 138830757738240 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777102511644862103, 'commit_timestamp_nsecs': 1777102553908884070, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/0/_CHECKPOINT_METADATA
I0425 07:39:28.635838 138830757738240 checkpoint_manager.py:1871] Missing metrics for step 0
I0425 07:39:29.000308 138952591587136 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints
I0425 07:39:29.111919 138952591587136 checkpoint_manager.py:929] [process=5][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_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e4b843193a0>
I0425 07:39:29.112096 138952591587136 checkpointing.py:301] Checkpoint manager created!
I0425 07:39:30.043500 138952591587136 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0425 07:39:30.043616 138952591587136 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 07:39:30.420730 138952591587136 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0425 07:39:30.420819 138952591587136 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 07:39:30.437313 138952591587136 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0425 07:39:30.437373 138952591587136 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 07:39:30.461026 138952591587136 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 07:39:30.461093 138952591587136 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 07:39:30.477812 138952591587136 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 07:39:30.477870 138952591587136 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 07:39:30.494496 138952591587136 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 07:39:30.494555 138952591587136 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 07:39:30.519360 138952591587136 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0425 07:39:30.519431 138952591587136 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 07:39:30.541546 138952591587136 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0425 07:39:30.541616 138952591587136 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0425 07:39:30.750854 138952591587136 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0425 07:39:30.750960 138952591587136 checkpointing.py:581] restoring from this run's directory step 9
I0425 07:39:30.996970 138952591587136 event_tracking.py:70] [process=5] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/9.
I0425 07:39:31.238502 138952591587136 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/9.
I0425 07:39:31.238950 138952591587136 event_tracking.py:125] [process=5] [sync] Finished blocking load in 0.24 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/9.
I0425 07:39:33.054344    1461 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0425 07:39:38.839129 138952591587136 jax_array_handlers.py:843] [process=5] /jax/orbax/read/worker/io/requested throughput: 272.893 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 5.790808439254761 s) (per-host)
I0425 07:39:38.843603 138952591587136 base_pytree_checkpoint_handler.py:130] [process=5] /jax/checkpoint/read/gbytes_per_sec: 256.279 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.175278902053833 s) (per-host)
I0425 07:39:40.096958 138952591587136 event_tracking.py:138] [process=5] [sync] Finished load in 9.10 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/9
I0425 07:39:40.097114 138952591587136 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints', 'checkpointer_start_time': 1777102770.996872, 'checkpointer_duration_secs': 9.100209474563599, 'checkpoint_manager_start_time': 1777102770.7512057, 'checkpoint_manager_duration_secs': 9.345877885818481}
fsdp: 32
I0425 07:39:40.514114 138952591587136 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0425 07:39:40.514244 138952591587136 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)
I0425 07:39:40.514298 138952591587136 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)
I0425 07:39:40.514357 138952591587136 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')
I0425 07:39:40.514408 138952591587136 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)
I0425 07:39:40.514446 138952591587136 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)
I0425 07:39:40.514497 138952591587136 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)
I0425 07:39:40.514548 138952591587136 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')
I0425 07:39:40.514587 138952591587136 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)
I0425 07:39:40.514624 138952591587136 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)
I0425 07:39:40.514684 138952591587136 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0425 07:39:40.514734 138952591587136 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0425 07:39:40.999165 138952591587136 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0425 07:39:40.999264 138952591587136 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0425 07:39:41.014898 138952591587136 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0425 07:39:41.014969 138952591587136 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0425 07:39:52.019056 138952591587136 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.
I0425 07:39:52.019806 138952591587136 metric_logger.py:301] number parameters: 1.104 billion
I0425 07:40:03.558957 138952591587136 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0425 07:40:03.711856 138952591587136 checkpointing.py:776] Waited 0.15288019180297852 seconds for step 10 to finish before starting checkpointing.
I0425 07:40:03.714487 138952591587136 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0425 07:40:03.716368 138952591587136 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 10
I0425 07:40:03.717838 138952591587136 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/10.
I0425 07:40:04.031196 138952591587136 signaling_client.py:364] Using JaxDistributedSignalingClient
I0425 07:40:04.032167 138952591587136 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0425 07:40:04.032224 138952591587136 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
I0425 07:40:04.310777 138952591587136 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279658s
I0425 07:40:04.310943 138952591587136 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.438 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28367090225219727 s) (per-host)
I0425 07:40:04.311000 138952591587136 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.283738s (batch_requests_ready=0.002302s, total_serialization_initiated=0.281361s, others=0.000074s)
I0425 07:40:04.311090 138952591587136 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287872s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.287855)
I0425 07:40:04.311838 138952591587136 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.60 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/10.
I0425 07:40:04.312128 138830229260032 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:00:04.312094
I0425 07:40:04.314058 138952591587136 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 07:40:04.314295 138825867183872 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0425 07:40:04.314390 138952591587136 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777102803.7144675, 'wait_for_prev_duration_secs': 6.198883056640625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777102803.7164059, 'checkpointer_blocking_duration_secs': 0.595874547958374, 'get_old_steps_start_time': 1777102804.312304, 'get_old_steps_duration_secs': 3.337860107421875e-05, 'checkpoint_manager_blocking_start_time': 1777102803.7123668, 'checkpoint_manager_blocking_duration_secs': 0.6019895076751709}
I0425 07:40:04.314539 138952591587136 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0425 07:40:04.314589 138952591587136 max_utils.py:750] 
Memstats: After params initialized:
I0425 07:40:04.314639 138952591587136 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0))
I0425 07:40:04.314688 138952591587136 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0))
I0425 07:40:04.314718 138952591587136 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0))
I0425 07:40:04.314743 138952591587136 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0))
I0425 07:40:04.626347 138952591587136 metric_logger.py:196] completed step: 10, seconds: 11.539, TFLOP/s/device: 1.177, Tokens/s/device: 177.484, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0425 07:40:04.816727 138952591587136 metric_logger.py:196] completed step: 11, seconds: 1.066, TFLOP/s/device: 12.748, Tokens/s/device: 1921.464, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0425 07:40:04.964202 138952591587136 metric_logger.py:196] completed step: 12, seconds: 0.043, TFLOP/s/device: 315.254, Tokens/s/device: 47518.504, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0425 07:40:05.111894 138952591587136 metric_logger.py:196] completed step: 13, seconds: 0.153, TFLOP/s/device: 88.813, Tokens/s/device: 13386.846, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0425 07:40:05.116091 138952591587136 checkpointing.py:772] Waiting for step 15 to finish before checkpoint...
I0425 07:40:05.405877 138952591587136 checkpointing.py:776] Waited 0.2897510528564453 seconds for step 15 to finish before starting checkpointing.
I0425 07:40:05.408922 138952591587136 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 07:40:09.288563 138825875576576 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/10/items/array_metadatas/process_5
I0425 07:40:45.661553 138830229260032 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.940 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.634241580963135 s) (per-host)
I0425 07:40:45.661693 138830229260032 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.349454s.
I0425 07:40:53.881609 138830229260032 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 49.569354s.
I0425 07:40:53.881870 138825867183872 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 07:40:53.881939 138825867183872 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0425 07:40:53.881990 138825867183872 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0425 07:40:53.884303 138825867183872 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0425 07:40:53.884470 138952591587136 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0425 07:40:53.884605 138952591587136 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.475682 seconds. If this number is high, consider checkpointing less frequently.
I0425 07:40:53.887563 138952591587136 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 15
I0425 07:40:53.890148 138952591587136 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/15.
I0425 07:40:54.532386 138952591587136 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0425 07:40:54.532482 138952591587136 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
I0425 07:40:54.572747 138952591587136 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042121s
I0425 07:40:54.572916 138952591587136 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 33.748 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04570913314819336 s) (per-host)
I0425 07:40:54.572967 138952591587136 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045770s (batch_requests_ready=0.001930s, total_serialization_initiated=0.043770s, others=0.000071s)
I0425 07:40:54.573060 138952591587136 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050695s (all_items=0.000013s, per_item={'items': '0.00001335'}, temp_paths=0.050682)
I0425 07:40:54.573756 138952591587136 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.69 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/15.
I0425 07:40:54.574090 138825867183872 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:00:54.574053
I0425 07:40:54.576049 138952591587136 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 07:40:54.576349 138825875576576 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0425 07:40:54.576489 138952591587136 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777102805.408893, 'wait_for_prev_duration_secs': 48.47568154335022, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777102853.887604, 'checkpointer_blocking_duration_secs': 0.6866261959075928, 'get_old_steps_start_time': 1777102854.5742564, 'get_old_steps_duration_secs': 3.24249267578125e-05, 'checkpoint_manager_blocking_start_time': 1777102805.4064717, 'checkpoint_manager_blocking_duration_secs': 49.16998291015625}
I0425 07:40:54.576668 138952591587136 checkpointing.py:408] Started an asynchronous checkpoint save for step 15
I0425 07:40:54.577446 138952591587136 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 92.087, Tokens/s/device: 13880.417, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0425 07:40:54.583493 138952591587136 metric_logger.py:196] completed step: 15, seconds: 0.148, TFLOP/s/device: 92.076, Tokens/s/device: 13878.724, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0425 07:41:00.178115 138824239785728 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/15/items/array_metadatas/process_5
I0425 07:41:15.516174 138952591587136 metric_logger.py:196] completed step: 16, seconds: 49.466, TFLOP/s/device: 0.275, Tokens/s/device: 41.402, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0425 07:41:15.663851 138952591587136 metric_logger.py:196] completed step: 17, seconds: 20.787, TFLOP/s/device: 0.654, Tokens/s/device: 98.524, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0425 07:41:15.811354 138952591587136 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.375, Tokens/s/device: 13471.557, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0425 07:41:15.958069 138952591587136 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0425 07:41:15.958791 138952591587136 checkpointing.py:776] Waited 0.0007455348968505859 seconds for step 19 to finish before starting checkpointing.
I0425 07:41:15.961149 138952591587136 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 07:41:36.857453 138825867183872 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.316 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.330207109451294 s) (per-host)
I0425 07:41:36.857581 138825867183872 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.283380s.
I0425 07:41:43.020378 138825867183872 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 48.446161s.
I0425 07:41:43.020691 138825875576576 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 07:41:43.020808 138825875576576 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0425 07:41:43.020854 138825875576576 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0425 07:41:43.022253 138825875576576 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts.
I0425 07:41:43.022392 138952591587136 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15.
W0425 07:41:43.022470 138952591587136 checkpoint_manager.py:1452] Waiting for previous save to complete took 27.061331 seconds. If this number is high, consider checkpointing less frequently.
I0425 07:41:43.024360 138952591587136 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 19
I0425 07:41:43.026497 138952591587136 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/19.
I0425 07:41:43.314701 138952591587136 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0425 07:41:43.314795 138952591587136 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
I0425 07:41:43.355461 138952591587136 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041612s
I0425 07:41:43.355621 138952591587136 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 34.101 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.045235395431518555 s) (per-host)
I0425 07:41:43.355685 138952591587136 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045308s (batch_requests_ready=0.001787s, total_serialization_initiated=0.043444s, others=0.000077s)
I0425 07:41:43.355777 138952591587136 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049311s (all_items=0.000011s, per_item={'items': '0.00001073'}, temp_paths=0.049300)
I0425 07:41:43.356501 138952591587136 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/19.
I0425 07:41:43.356811 138825875576576 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:01:43.356778
I0425 07:41:43.358784 138952591587136 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 07:41:43.359080 138824239785728 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0425 07:41:43.359245 138952591587136 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777102875.9611194, 'wait_for_prev_duration_secs': 27.061330795288086, 'time_between_consecutive_saves_sec': 22.076775074005127, 'checkpointer_blocking_start_time': 1777102903.0243971, 'checkpointer_blocking_duration_secs': 0.33256101608276367, 'get_old_steps_start_time': 1777102903.3569756, 'get_old_steps_duration_secs': 2.6702880859375e-05, 'checkpoint_manager_blocking_start_time': 1777102875.9590228, 'checkpoint_manager_blocking_duration_secs': 27.400190114974976}
I0425 07:41:43.359352 138952591587136 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0425 07:41:43.359396 138952591587136 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 07:41:48.371646 138824281749248 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_08_checkpoint_async_true/checkpoints/19/items/array_metadatas/process_5
I0425 07:42:24.609368 138825875576576 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 38.248 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.29894161224365 s) (per-host)
I0425 07:42:24.609493 138825875576576 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.252569s.
I0425 07:42:31.719234 138825875576576 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 48.362295s.
I0425 07:42:31.719553 138824239785728 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 07:42:31.719704 138824239785728 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0425 07:42:31.719746 138824239785728 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0425 07:42:31.722257 138824239785728 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0425 07:42:31.722426 138952591587136 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0425 07:42:31.722570 138952591587136 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0425 07:42:31.723429 138952591587136 metric_logger.py:196] completed step: 19, seconds: 0.148, TFLOP/s/device: 92.001, Tokens/s/device: 13867.353, 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: Sat Apr 25 07:42:43 UTC 2026
EXIT_CODE=0