main (b341212ce) vs main (172d0f10a)| Metric | main b341212ce | main 172d0f10a | Diff (main − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 1.4490 | 1.4490 | 0 |
| TFLOP/s | 112.558 | 101.135 | -11.423 |
| Tok/s | 16966.0 | 15244.1 | -1721.895 |
| Avg s/step | 2.412 | 1.791 | -0.621 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.8.1 | 0.8.1 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-21 15:53:37.764888: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered WARNING: All log messages before absl::InitializeLog() is called are written to STDERR E0000 00:00:1776786817.775502 1541159 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776786817.778749 1541159 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776786817.789724 1541159 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776786817.789735 1541159 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776786817.789737 1541159 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776786817.789738 1541159 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. 2026-04-21 15:53:41.172700: 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) I0421 15:53:41.289894 126285427031168 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0421 15:54:38.304672 126285427031168 max_utils.py:800] System Information: Jax Version: 0.8.1 I0421 15:54:38.304860 126285427031168 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0421 15:54:38.304908 126285427031168 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709 I0421 15:54:38.304938 126285427031168 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system I0421 15:54:38.304962 126285427031168 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 15:54:38.305082 126285427031168 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0421 15:54:38.669029 126285427031168 maxtext_utils.py:1551] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 15:54:38.669404 126285427031168 checkpointing.py:677] Setting up checkpoint logger... I0421 15:54:38.669451 126285427031168 checkpointing.py:233] Creating checkpoint manager with ocdbt=False and zarr3=True I0421 15:54:38.669543 126285427031168 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0421 15:54:38.670346 126285427031168 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x72dad1b9b110>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 15:54:40.316493 126285427031168 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0421 15:54:40.316834 126285427031168 checkpoint_manager.py:702] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72d929b80bf0>}, handler_registry=None I0421 15:54:40.317274 126285427031168 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72d929b80bf0>` 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`. I0421 15:54:40.317321 126285427031168 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72d929b829c0>` 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`. I0421 15:54:40.317356 126285427031168 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 0x72d929b80bf0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72d929b80bf0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72d929b829c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72d929b829c0>}). I0421 15:54:40.317725 126285427031168 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0421 15:54:40.317794 126285427031168 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x72d929ba9260> timeout: 600 secs and primary_host=0 for async checkpoint writes I0421 15:54:40.776699 126201790203456 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776786770412013273, 'commit_timestamp_nsecs': 1776786790964829132, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/9/_CHECKPOINT_METADATA I0421 15:54:40.864947 126201790203456 checkpoint_manager.py:1847] Missing metrics for step 9 I0421 15:54:40.918763 126285427031168 checkpoint_manager.py:1788] Found 3 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints I0421 15:54:40.919296 126285427031168 checkpoint_manager.py:921] [process=0][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_hns=False, 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), root_directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x72d929b81340> I0421 15:54:40.919394 126285427031168 checkpointing.py:301] Checkpoint manager created! I0421 15:54:41.577264 126285427031168 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 15:54:41.577356 126285427031168 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:54:41.854923 126285427031168 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 15:54:41.855001 126285427031168 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:54:41.867026 126285427031168 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 15:54:41.867082 126285427031168 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:54:41.885896 126285427031168 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:54:41.885950 126285427031168 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:54:41.898025 126285427031168 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:54:41.898085 126285427031168 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:54:41.910124 126285427031168 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:54:41.910170 126285427031168 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:54:41.929168 126285427031168 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 15:54:41.929224 126285427031168 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:54:41.944261 126285427031168 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 15:54:41.944309 126285427031168 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0421 15:54:42.091295 126285427031168 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0421 15:54:42.091389 126285427031168 checkpointing.py:581] restoring from this run's directory step 9 I0421 15:54:42.250895 126285427031168 checkpointer.py:304] Restoring checkpoint from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/9. I0421 15:54:43.133699 1542688 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0421 15:54:43.133757 1542688 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0421 15:54:49.310798 126285427031168 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/read/gbytes_per_sec: 1.847 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 6 seconds) (per-host) I0421 15:54:49.311066 126285427031168 checkpointer.py:318] Finished restoring checkpoint in 7.14 seconds from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/9. I0421 15:54:49.311133 126285427031168 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints', 'checkpointer_start_time': 1776786882.1665788, 'checkpointer_duration_secs': 7.144531011581421, 'checkpoint_manager_start_time': 1776786882.0915906, 'checkpoint_manager_duration_secs': 7.219520807266235} [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0421 15:54:49.638550 126285427031168 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0421 15:54:49.638662 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0421 15:54:49.638705 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0421 15:54:49.638749 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0421 15:54:49.638790 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0421 15:54:49.638818 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0421 15:54:49.638856 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0421 15:54:49.638895 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: PartitionSpec('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0421 15:54:49.638931 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0421 15:54:49.638958 126285427031168 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0421 15:54:49.638993 126285427031168 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 15:54:49.639028 126285427031168 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 15:54:50.034511 126285427031168 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 15:54:50.034592 126285427031168 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0421 15:54:50.046084 126285427031168 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 15:54:50.046133 126285427031168 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0421 15:54:50.706502 126285427031168 max_utils.py:791] Total memory size: 3.6 GB, Output size: 1.5 GB, Temp size: 2.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0421 15:54:50.710012 126285427031168 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0421 15:54:50.710204 126285427031168 metric_logger.py:301] number parameters: 1.104 billion I0421 15:54:52.376949 126285427031168 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0421 15:54:52.493122 126285427031168 checkpointing.py:776] Waited 0.11621475219726562 seconds for step 10 to finish before starting checkpointing. I0421 15:54:52.493847 126285427031168 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 15:54:52.494094 126285427031168 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0421 15:54:52.494702 126285427031168 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10. I0421 15:54:52.563208 126285427031168 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0421 15:54:52.653955 126285427031168 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:54:52.654065 126285427031168 replica_slices.py:410] 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 I0421 15:54:52.661482 126201052005952 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10 I0421 15:54:53.379783 126201041520192 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10/items I0421 15:54:53.532575 126201010062912 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776786893298126495, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776786893298126495, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10/_CHECKPOINT_METADATA I0421 15:54:56.622678 126285427031168 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 3.969145s I0421 15:54:56.627000 126285427031168 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 3.038 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 4 seconds) (per-host) I0421 15:54:56.627082 126285427031168 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 4.062829s (batch_requests_ready=0.086479s, total_serialization_initiated=3.972104s, others=0.004246s) I0421 15:54:56.627154 126285427031168 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 4.063451s (all_items=0.000043s, per_item={'items': '0.00004268'}, temp_paths=4.063409) I0421 15:54:56.628496 126201062491712 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:54:56.628592 126285427031168 async_checkpointer.py:561] Finished blocking save. Time taken: 4.134449s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10. I0421 15:54:56.628808 126285427031168 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:54:56.629012 126201052005952 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:54:56.629146 126285427031168 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776786892.4938262, 'wait_for_prev_duration_secs': 4.76837158203125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776786892.4941158, 'checkpointer_blocking_duration_secs': 4.13459587097168, 'get_old_steps_start_time': 1776786896.6287274, 'get_old_steps_duration_secs': 3.9577484130859375e-05, 'checkpoint_manager_blocking_start_time': 1776786892.4936206, 'checkpoint_manager_blocking_duration_secs': 4.1355016231536865} I0421 15:54:56.629227 126285427031168 checkpointing.py:408] Started an asynchronous checkpoint save for step 10 I0421 15:54:56.629278 126285427031168 max_utils.py:750] Memstats: After params initialized: I0421 15:54:56.629327 126285427031168 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0421 15:54:56.629351 126285427031168 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0421 15:54:56.629371 126285427031168 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0421 15:54:56.629390 126285427031168 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0421 15:54:56.629408 126285427031168 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0421 15:54:56.629425 126285427031168 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0421 15:54:56.629441 126285427031168 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0421 15:54:56.629457 126285427031168 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0421 15:54:56.967480 126285427031168 metric_logger.py:196] completed step: 10, seconds: 1.666, TFLOP/s/device: 8.155, Tokens/s/device: 1229.211, total_weights: 16384, loss: 3.602, lm_loss: 3.602, perplexity: 36.686 I0421 15:54:57.083525 126285427031168 metric_logger.py:196] completed step: 11, seconds: 4.588, TFLOP/s/device: 2.962, Tokens/s/device: 446.399, total_weights: 16384, loss: 3.101, lm_loss: 3.101, perplexity: 22.211 I0421 15:54:57.202365 126285427031168 metric_logger.py:196] completed step: 12, seconds: 0.007, TFLOP/s/device: 1842.822, Tokens/s/device: 277770.243, total_weights: 16384, loss: 2.684, lm_loss: 2.684, perplexity: 14.644 I0421 15:54:57.322154 126285427031168 metric_logger.py:196] completed step: 13, seconds: 0.122, TFLOP/s/device: 111.719, Tokens/s/device: 16839.474, total_weights: 16384, loss: 2.348, lm_loss: 2.348, perplexity: 10.461 I0421 15:54:57.339439 126285427031168 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0421 15:54:57.384822 126201020548672 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10/items/array_metadatas/process_0 I0421 15:54:57.564296 126285427031168 checkpointing.py:776] Waited 0.22484183311462402 seconds for step 15 to finish before starting checkpointing. I0421 15:54:57.564723 126285427031168 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:55:02.901400 126201041520192 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 6.273668s (commit=5.881005s, array_metadata_write=0.392663s) I0421 15:55:02.902388 126201062491712 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 1.194 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 10 seconds) (per-host) I0421 15:55:02.902500 126201062491712 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 6.273862s. W0421 15:55:03.209344 126201062491712 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10/items, I0421 15:55:03.302793 126201062491712 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0421 15:55:03.302899 126201062491712 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.154769s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10/items I0421 15:55:03.303541 126201062491712 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10/items I0421 15:55:03.462567 126201010062912 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776786893298126495, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10/_CHECKPOINT_METADATA I0421 15:55:03.546238 126201062491712 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10 I0421 15:55:04.236640 126201062491712 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10`. I0421 15:55:04.237303 126201062491712 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 11.743167s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/10 I0421 15:55:04.237388 126201062491712 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 7.608750s. I0421 15:55:04.237575 126201052005952 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:55:04.237687 126201052005952 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:55:04.237758 126201052005952 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:55:04.237808 126201052005952 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0421 15:55:04.238301 126285427031168 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0421 15:55:04.238538 126285427031168 checkpoint_manager.py:1441] Waiting for previous save to complete took 6.673833 seconds. If this number is high, consider checkpointing less frequently. I0421 15:55:04.239318 126285427031168 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 15 I0421 15:55:04.239605 126285427031168 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15. I0421 15:55:04.400516 126201052005952 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15 I0421 15:55:04.429059 126285427031168 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:55:04.429154 126285427031168 replica_slices.py:410] 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 I0421 15:55:05.057069 126201041520192 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15/items I0421 15:55:14.571394 126285427031168 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.142741s I0421 15:55:14.576600 126285427031168 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.204 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 10 seconds) (per-host) I0421 15:55:14.576661 126285427031168 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 10.247512s (batch_requests_ready=0.096617s, total_serialization_initiated=10.145798s, others=0.005098s) I0421 15:55:14.576737 126285427031168 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 10.248067s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=10.248050) I0421 15:55:14.578140 126201031034432 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:55:14.578223 126285427031168 async_checkpointer.py:561] Finished blocking save. Time taken: 10.338861s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15. I0421 15:55:14.578416 126285427031168 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:55:14.578619 126199787423296 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:55:14.578728 126285427031168 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776786897.5646894, 'wait_for_prev_duration_secs': 6.67383337020874, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776786904.239344, 'checkpointer_blocking_duration_secs': 10.338986158370972, 'get_old_steps_start_time': 1776786914.5783484, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1776786897.5646129, 'checkpoint_manager_blocking_duration_secs': 17.014091968536377} I0421 15:55:14.578864 126285427031168 checkpointing.py:408] Started an asynchronous checkpoint save for step 15 I0421 15:55:14.579561 126285427031168 metric_logger.py:196] completed step: 14, seconds: 0.122, TFLOP/s/device: 111.751, Tokens/s/device: 16844.322, total_weights: 16384, loss: 2.083, lm_loss: 2.083, perplexity: 8.030 I0421 15:55:14.588415 126285427031168 metric_logger.py:196] completed step: 15, seconds: 0.124, TFLOP/s/device: 109.720, Tokens/s/device: 16538.135, total_weights: 16384, loss: 1.880, lm_loss: 1.880, perplexity: 6.551 I0421 15:55:14.704562 126285427031168 metric_logger.py:196] completed step: 16, seconds: 17.247, TFLOP/s/device: 0.788, Tokens/s/device: 118.746, total_weights: 16384, loss: 1.726, lm_loss: 1.726, perplexity: 5.616 I0421 15:55:14.824033 126285427031168 metric_logger.py:196] completed step: 17, seconds: 0.006, TFLOP/s/device: 2114.070, Tokens/s/device: 318655.671, total_weights: 16384, loss: 1.610, lm_loss: 1.610, perplexity: 5.002 I0421 15:55:14.943088 126285427031168 metric_logger.py:196] completed step: 18, seconds: 0.117, TFLOP/s/device: 115.893, Tokens/s/device: 17468.590, total_weights: 16384, loss: 1.521, lm_loss: 1.521, perplexity: 4.577 I0421 15:55:15.061568 126285427031168 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0421 15:55:15.062554 126285427031168 checkpointing.py:776] Waited 0.0009884834289550781 seconds for step 19 to finish before starting checkpointing. I0421 15:55:15.062870 126285427031168 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:55:15.347863 126201020548672 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15/items/array_metadatas/process_0 I0421 15:55:20.768117 126201041520192 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 6.190783s (commit=5.777143s, array_metadata_write=0.413640s) I0421 15:55:20.768984 126201031034432 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 768.670 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 16 seconds) (per-host) I0421 15:55:20.769035 126201031034432 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 6.190769s. W0421 15:55:21.097814 126201031034432 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15/items, I0421 15:55:21.185788 126201031034432 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0421 15:55:21.185895 126201031034432 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.151519s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15/items I0421 15:55:21.186545 126201031034432 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15/items I0421 15:55:21.416101 126201031034432 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15 I0421 15:55:22.099579 126201031034432 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15`. I0421 15:55:22.100214 126201031034432 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 17.860857s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/15 I0421 15:55:22.100282 126201031034432 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 7.522017s. I0421 15:55:22.100450 126199787423296 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:55:22.100556 126199787423296 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:55:22.100608 126199787423296 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:55:22.100651 126199787423296 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0421 15:55:22.100795 126285427031168 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15. W0421 15:55:22.100902 126285427031168 checkpoint_manager.py:1441] Waiting for previous save to complete took 7.038052 seconds. If this number is high, consider checkpointing less frequently. I0421 15:55:22.101115 126285427031168 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0421 15:55:22.101397 126285427031168 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19. I0421 15:55:22.272467 126285427031168 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:55:22.272557 126285427031168 replica_slices.py:410] 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 I0421 15:55:22.274985 126201031034432 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19 I0421 15:55:22.928575 126201041520192 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19/items I0421 15:55:29.920936 126285427031168 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 7.648792s I0421 15:55:29.925370 126285427031168 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.595 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 7 seconds) (per-host) I0421 15:55:29.925439 126285427031168 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 7.736841s (batch_requests_ready=0.080378s, total_serialization_initiated=7.652150s, others=0.004313s) I0421 15:55:29.925531 126285427031168 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 7.737358s (all_items=0.000012s, per_item={'items': '0.00001240'}, temp_paths=7.737346) I0421 15:55:29.926830 126201031034432 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:55:29.926930 126285427031168 async_checkpointer.py:561] Finished blocking save. Time taken: 7.825773s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19. I0421 15:55:29.927178 126285427031168 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:55:29.927374 126199535765056 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:55:29.927471 126285427031168 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776786915.0628378, 'wait_for_prev_duration_secs': 7.038051605224609, 'time_between_consecutive_saves_sec': 10.825014114379883, 'checkpointer_blocking_start_time': 1776786922.1011384, 'checkpointer_blocking_duration_secs': 7.825978994369507, 'get_old_steps_start_time': 1776786929.9271278, 'get_old_steps_duration_secs': 1.9073486328125e-05, 'checkpoint_manager_blocking_start_time': 1776786915.0628018, 'checkpoint_manager_blocking_duration_secs': 14.864646196365356} I0421 15:55:29.927566 126285427031168 checkpointing.py:408] Started an asynchronous checkpoint save for step 19 I0421 15:55:29.927597 126285427031168 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:55:30.662657 126200999577152 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19/items/array_metadatas/process_0 I0421 15:55:36.321797 126201041520192 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 6.395672s (commit=6.024359s, array_metadata_write=0.371313s) I0421 15:55:36.322760 126201031034432 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 894.065 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 14 seconds) (per-host) I0421 15:55:36.322859 126201031034432 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 6.395885s. W0421 15:55:36.625607 126201031034432 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19/items, I0421 15:55:36.742720 126201031034432 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0421 15:55:36.742825 126201031034432 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.173566s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19/items I0421 15:55:36.743471 126201031034432 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19/items I0421 15:55:36.988537 126201031034432 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19 I0421 15:55:37.648160 126201031034432 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19`. I0421 15:55:37.648780 126201031034432 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 15.547630s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_15_ocdbt_false/checkpoints/19 I0421 15:55:37.648842 126201031034432 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 7.721869s. I0421 15:55:37.648952 126199535765056 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:55:37.649001 126199535765056 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:55:37.649038 126199535765056 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:55:37.649085 126199535765056 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0421 15:55:37.649209 126285427031168 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0421 15:55:37.649337 126285427031168 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 15:55:37.650112 126285427031168 metric_logger.py:196] completed step: 19, seconds: 0.121, TFLOP/s/device: 112.558, Tokens/s/device: 16966.002, total_weights: 16384, loss: 1.449, lm_loss: 1.449, perplexity: 4.260 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
2026-04-23 02:24:59.329226: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered WARNING: All log messages before absl::InitializeLog() is called are written to STDERR E0000 00:00:1776911099.342448 2320751 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776911099.346261 2320751 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776911099.357770 2320751 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776911099.357786 2320751 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776911099.357788 2320751 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776911099.357789 2320751 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} `rope_scaling`'s factor field must be a float >= 1, got 40 `rope_scaling`'s beta_fast field must be a float, got 32 `rope_scaling`'s beta_slow field must be a float, got 1 Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} 2026-04-23 02:25:03.329867: 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 02:25:03.465980 137884523878208 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0423 02:26:01.772751 137884523878208 max_utils.py:800] System Information: Jax Version: 0.8.1 I0423 02:26:01.772862 137884523878208 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0423 02:26:01.772899 137884523878208 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709 I0423 02:26:01.772925 137884523878208 train_utils.py:348] WARNING: 'dataset_path' might be pointing your local file system I0423 02:26:01.772948 137884523878208 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 02:26:01.773067 137884523878208 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0423 02:26:02.438713 137884523878208 maxtext_utils.py:1604] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 02:26:02.438982 137884523878208 checkpointing.py:677] Setting up checkpoint logger... I0423 02:26:02.439029 137884523878208 checkpointing.py:233] Creating checkpoint manager with ocdbt=False and zarr3=True I0423 02:26:02.439084 137884523878208 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0423 02:26:02.439531 137884523878208 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7d67791bad50>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 02:26:04.112258 137884523878208 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0423 02:26:04.112627 137884523878208 checkpoint_manager.py:702] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d659f3a9d90>}, handler_registry=None I0423 02:26:04.112946 137884523878208 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d659f3a9d90>` 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 02:26:04.112992 137884523878208 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d659f3aa3f0>` 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 02:26:04.113021 137884523878208 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 0x7d659f3a9d90>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d659f3a9d90>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d659f3aa3f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d659f3aa3f0>}). I0423 02:26:04.113317 137884523878208 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0423 02:26:04.113369 137884523878208 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7d659f3c8900> timeout: 600 secs and primary_host=0 for async checkpoint writes I0423 02:26:04.649085 137854850172480 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776911049717085118, 'commit_timestamp_nsecs': 1776911058144697246, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/5/_CHECKPOINT_METADATA I0423 02:26:04.736185 137854850172480 checkpoint_manager.py:1847] Missing metrics for step 5 I0423 02:26:04.751745 137884523878208 checkpoint_manager.py:1788] Found 3 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints I0423 02:26:04.752208 137884523878208 checkpoint_manager.py:921] [process=0][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_hns=False, 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), root_directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d659f3a8e00> I0423 02:26:04.752328 137884523878208 checkpointing.py:301] Checkpoint manager created! I0423 02:26:05.156143 137884523878208 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 02:26:05.156236 137884523878208 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0423 02:26:05.436206 137884523878208 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 02:26:05.436292 137884523878208 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0423 02:26:05.448700 137884523878208 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 02:26:05.448747 137884523878208 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0423 02:26:05.467801 137884523878208 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 02:26:05.467862 137884523878208 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0423 02:26:05.480218 137884523878208 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 02:26:05.480267 137884523878208 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0423 02:26:05.492459 137884523878208 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 02:26:05.492511 137884523878208 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0423 02:26:05.511945 137884523878208 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0423 02:26:05.512004 137884523878208 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0423 02:26:05.527431 137884523878208 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 02:26:05.527484 137884523878208 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0423 02:26:05.678835 137884523878208 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0423 02:26:05.678935 137884523878208 checkpointing.py:581] restoring from this run's directory step 9 I0423 02:26:05.847275 137884523878208 checkpointer.py:304] Restoring checkpoint from gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/9. I0423 02:26:06.701156 2322337 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0423 02:26:06.701208 2322337 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0423 02:26:12.672733 137884523878208 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/read/gbytes_per_sec: 1.911 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 6 seconds) (per-host) I0423 02:26:12.672992 137884523878208 checkpointer.py:318] Finished restoring checkpoint in 6.91 seconds from gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/9. I0423 02:26:12.673071 137884523878208 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints', 'checkpointer_start_time': 1776911165.7665188, 'checkpointer_duration_secs': 6.906516790390015, 'checkpoint_manager_start_time': 1776911165.67915, 'checkpoint_manager_duration_secs': 6.993886947631836} [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0423 02:26:13.000226 137884523878208 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0423 02:26:13.000337 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 02:26:13.000379 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 02:26:13.000424 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0423 02:26:13.000464 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0423 02:26:13.000493 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0423 02:26:13.000533 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0423 02:26:13.000573 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: PartitionSpec('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0423 02:26:13.000601 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0423 02:26:13.000627 137884523878208 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0423 02:26:13.000671 137884523878208 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 02:26:13.000706 137884523878208 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 02:26:13.400249 137884523878208 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0423 02:26:13.400331 137884523878208 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0423 02:26:13.411876 137884523878208 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0423 02:26:13.411929 137884523878208 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0423 02:26:14.212199 137884523878208 max_utils.py:791] Total memory size: 3.6 GB, Output size: 1.5 GB, Temp size: 2.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0423 02:26:14.215726 137884523878208 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0423 02:26:14.215882 137884523878208 metric_logger.py:301] number parameters: 1.104 billion I0423 02:26:16.172334 137884523878208 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0423 02:26:16.290051 137884523878208 checkpointing.py:776] Waited 0.11769819259643555 seconds for step 10 to finish before starting checkpointing. I0423 02:26:16.290671 137884523878208 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 02:26:16.290822 137884523878208 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0423 02:26:16.291135 137884523878208 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10. I0423 02:26:16.371030 137884523878208 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0423 02:26:16.458606 137884523878208 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0423 02:26:16.458759 137852444739136 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10 I0423 02:26:16.458844 137884523878208 replica_slices.py:410] 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 02:26:17.200258 137852434253376 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10/items I0423 02:26:17.342670 137852224538176 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776911177116346547, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776911177116346547, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10/_CHECKPOINT_METADATA I0423 02:26:19.474843 137884523878208 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 3.016790s I0423 02:26:19.478975 137884523878208 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 3.972 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 3 seconds) (per-host) I0423 02:26:19.479038 137884523878208 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 3.106665s (batch_requests_ready=0.082972s, total_serialization_initiated=3.019679s, others=0.004014s) I0423 02:26:19.479138 137884523878208 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 3.107642s (all_items=0.000021s, per_item={'items': '0.00002122'}, temp_paths=3.107621) I0423 02:26:19.480824 137852455224896 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0423 02:26:19.480976 137884523878208 async_checkpointer.py:561] Finished blocking save. Time taken: 3.190113s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10. I0423 02:26:19.481145 137884523878208 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0423 02:26:19.481360 137852423767616 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0423 02:26:19.481459 137884523878208 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776911176.2906575, 'wait_for_prev_duration_secs': 4.100799560546875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776911176.2908442, 'checkpointer_blocking_duration_secs': 3.1901910305023193, 'get_old_steps_start_time': 1776911179.481074, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776911176.2905712, 'checkpoint_manager_blocking_duration_secs': 3.190861701965332} I0423 02:26:19.481572 137884523878208 checkpointing.py:408] Started an asynchronous checkpoint save for step 10 I0423 02:26:19.481612 137884523878208 max_utils.py:750] Memstats: After params initialized: I0423 02:26:19.481660 137884523878208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0423 02:26:19.481685 137884523878208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0423 02:26:19.481706 137884523878208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0423 02:26:19.481725 137884523878208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0423 02:26:19.481745 137884523878208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0423 02:26:19.481764 137884523878208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0423 02:26:19.481784 137884523878208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0423 02:26:19.481800 137884523878208 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0423 02:26:19.750367 137884523878208 metric_logger.py:196] completed step: 10, seconds: 1.956, TFLOP/s/device: 6.947, Tokens/s/device: 1047.080, total_weights: 16384, loss: 3.602, lm_loss: 3.602, perplexity: 36.686 I0423 02:26:19.867352 137884523878208 metric_logger.py:196] completed step: 11, seconds: 3.576, TFLOP/s/device: 3.800, Tokens/s/device: 572.771, total_weights: 16384, loss: 3.101, lm_loss: 3.101, perplexity: 22.211 I0423 02:26:19.999534 137884523878208 metric_logger.py:196] completed step: 12, seconds: 0.007, TFLOP/s/device: 1966.015, Tokens/s/device: 296339.169, total_weights: 16384, loss: 2.684, lm_loss: 2.684, perplexity: 14.644 I0423 02:26:20.109465 137884523878208 metric_logger.py:196] completed step: 13, seconds: 0.119, TFLOP/s/device: 113.796, Tokens/s/device: 17152.572, total_weights: 16384, loss: 2.348, lm_loss: 2.348, perplexity: 10.461 I0423 02:26:20.127020 137884523878208 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0423 02:26:20.310909 137852274869824 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10/items/array_metadatas/process_0 I0423 02:26:20.347779 137884523878208 checkpointing.py:776] Waited 0.22071433067321777 seconds for step 15 to finish before starting checkpointing. I0423 02:26:20.348939 137884523878208 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 02:26:26.328513 137852434253376 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 6.848480s (commit=6.436277s, array_metadata_write=0.412203s) I0423 02:26:26.329393 137852455224896 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 1.239 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0423 02:26:26.329443 137852455224896 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 6.848535s. W0423 02:26:26.631322 137852455224896 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10/items, I0423 02:26:26.745294 137852455224896 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0423 02:26:26.745411 137852455224896 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.161029s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10/items I0423 02:26:26.746064 137852455224896 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10/items I0423 02:26:26.928281 137852224538176 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776911177116346547, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10/_CHECKPOINT_METADATA I0423 02:26:26.965571 137852455224896 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10 I0423 02:26:27.628887 137852455224896 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10`. I0423 02:26:27.629621 137852455224896 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 11.338763s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/10 I0423 02:26:27.629694 137852455224896 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 8.148787s. I0423 02:26:27.629856 137852423767616 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 02:26:27.629967 137852423767616 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0423 02:26:27.630018 137852423767616 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0423 02:26:27.630074 137852423767616 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0423 02:26:27.630248 137884523878208 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0423 02:26:27.630405 137884523878208 checkpoint_manager.py:1441] Waiting for previous save to complete took 7.281487 seconds. If this number is high, consider checkpointing less frequently. I0423 02:26:27.631263 137884523878208 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 15 I0423 02:26:27.631492 137884523878208 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15. I0423 02:26:27.798054 137884523878208 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0423 02:26:27.798144 137884523878208 replica_slices.py:410] 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 02:26:27.803494 137852455224896 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15 I0423 02:26:28.465925 137852274869824 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15/items I0423 02:26:31.850752 137884523878208 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 4.053117s I0423 02:26:31.854620 137884523878208 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.982 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 4 seconds) (per-host) I0423 02:26:31.854683 137884523878208 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 4.138876s (batch_requests_ready=0.079079s, total_serialization_initiated=4.056046s, others=0.003751s) I0423 02:26:31.854761 137884523878208 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 4.139399s (all_items=0.000013s, per_item={'items': '0.00001264'}, temp_paths=4.139387) I0423 02:26:31.856204 137852423767616 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0423 02:26:31.856289 137884523878208 async_checkpointer.py:561] Finished blocking save. Time taken: 4.224979s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15. I0423 02:26:31.856471 137884523878208 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0423 02:26:31.856631 137852455224896 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0423 02:26:31.856729 137884523878208 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776911180.3488984, 'wait_for_prev_duration_secs': 7.281487464904785, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776911187.6312902, 'checkpointer_blocking_duration_secs': 4.225105285644531, 'get_old_steps_start_time': 1776911191.8564112, 'get_old_steps_duration_secs': 2.288818359375e-05, 'checkpoint_manager_blocking_start_time': 1776911180.3487992, 'checkpoint_manager_blocking_duration_secs': 11.507904529571533} I0423 02:26:31.856836 137884523878208 checkpointing.py:408] Started an asynchronous checkpoint save for step 15 I0423 02:26:31.857550 137884523878208 metric_logger.py:196] completed step: 14, seconds: 0.139, TFLOP/s/device: 97.881, Tokens/s/device: 14753.661, total_weights: 16384, loss: 2.083, lm_loss: 2.083, perplexity: 8.030 I0423 02:26:31.865848 137884523878208 metric_logger.py:196] completed step: 15, seconds: 0.114, TFLOP/s/device: 119.542, Tokens/s/device: 18018.652, total_weights: 16384, loss: 1.880, lm_loss: 1.880, perplexity: 6.551 I0423 02:26:31.982333 137884523878208 metric_logger.py:196] completed step: 16, seconds: 11.737, TFLOP/s/device: 1.158, Tokens/s/device: 174.493, total_weights: 16384, loss: 1.726, lm_loss: 1.726, perplexity: 5.616 I0423 02:26:32.111105 137884523878208 metric_logger.py:196] completed step: 17, seconds: 0.007, TFLOP/s/device: 2057.098, Tokens/s/device: 310068.130, total_weights: 16384, loss: 1.610, lm_loss: 1.610, perplexity: 5.002 I0423 02:26:32.222334 137884523878208 metric_logger.py:196] completed step: 18, seconds: 0.117, TFLOP/s/device: 115.639, Tokens/s/device: 17430.381, total_weights: 16384, loss: 1.521, lm_loss: 1.521, perplexity: 4.577 I0423 02:26:32.341266 137884523878208 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0423 02:26:32.342618 137884523878208 checkpointing.py:776] Waited 0.0013873577117919922 seconds for step 19 to finish before starting checkpointing. I0423 02:26:32.342997 137884523878208 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 02:26:32.608834 137852444739136 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15/items/array_metadatas/process_0 I0423 02:26:40.086664 137852274869824 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 8.231426s (commit=7.796989s, array_metadata_write=0.434437s) I0423 02:26:40.087553 137852423767616 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 1021.426 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 12 seconds) (per-host) I0423 02:26:40.087605 137852423767616 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 8.231271s. W0423 02:26:40.419925 137852423767616 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15/items, I0423 02:26:40.535655 137852423767616 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0423 02:26:40.535768 137852423767616 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.166564s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15/items I0423 02:26:40.536457 137852423767616 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15/items I0423 02:26:40.758432 137852423767616 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15 I0423 02:26:41.413628 137852423767616 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15`. I0423 02:26:41.414290 137852423767616 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 13.782986s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/15 I0423 02:26:41.414365 137852423767616 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 9.558033s. I0423 02:26:41.414545 137852455224896 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 02:26:41.414650 137852455224896 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0423 02:26:41.414703 137852455224896 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0423 02:26:41.414742 137852455224896 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0423 02:26:41.414924 137884523878208 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15. W0423 02:26:41.415093 137884523878208 checkpoint_manager.py:1441] Waiting for previous save to complete took 9.072117 seconds. If this number is high, consider checkpointing less frequently. I0423 02:26:41.415592 137884523878208 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0423 02:26:41.415991 137884523878208 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19. I0423 02:26:41.579801 137884523878208 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0423 02:26:41.579896 137884523878208 replica_slices.py:410] 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 02:26:41.635342 137852423767616 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19 I0423 02:26:42.314390 137852444739136 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19/items I0423 02:26:46.275377 137884523878208 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 4.695953s I0423 02:26:46.279818 137884523878208 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.584 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 4 seconds) (per-host) I0423 02:26:46.279896 137884523878208 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 4.775573s (batch_requests_ready=0.072092s, total_serialization_initiated=4.699205s, others=0.004276s) I0423 02:26:46.279995 137884523878208 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 4.776097s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=4.776080) I0423 02:26:46.281719 137852037891648 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0423 02:26:46.281819 137884523878208 async_checkpointer.py:561] Finished blocking save. Time taken: 4.866107s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19. I0423 02:26:46.282033 137884523878208 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0423 02:26:46.282193 137852455224896 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0423 02:26:46.282264 137884523878208 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776911192.3429592, 'wait_for_prev_duration_secs': 9.07211685180664, 'time_between_consecutive_saves_sec': 4.712869644165039, 'checkpointer_blocking_start_time': 1776911201.4156923, 'checkpointer_blocking_duration_secs': 4.866242170333862, 'get_old_steps_start_time': 1776911206.2819562, 'get_old_steps_duration_secs': 3.3855438232421875e-05, 'checkpoint_manager_blocking_start_time': 1776911192.3429146, 'checkpoint_manager_blocking_duration_secs': 13.939323902130127} I0423 02:26:46.282394 137884523878208 checkpointing.py:408] Started an asynchronous checkpoint save for step 19 I0423 02:26:46.282427 137884523878208 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 02:26:47.041260 137852434253376 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19/items/array_metadatas/process_0 I0423 02:26:54.140673 137852444739136 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 7.860072s (commit=7.434976s, array_metadata_write=0.425096s) I0423 02:26:54.141704 137852037891648 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 999.959 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 12 seconds) (per-host) I0423 02:26:54.141822 137852037891648 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 7.859952s. W0423 02:26:54.460540 137852037891648 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19/items, I0423 02:26:54.568394 137852037891648 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0423 02:26:54.568512 137852037891648 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.158864s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19/items I0423 02:26:54.569175 137852037891648 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19/items I0423 02:26:54.798624 137852037891648 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19 I0423 02:26:55.512505 137852037891648 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19`. I0423 02:26:55.513257 137852037891648 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 14.097553s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260423_012248/linen_main_20260423_012248_15_ocdbt_false/checkpoints/19 I0423 02:26:55.513320 137852037891648 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 9.231453s. I0423 02:26:55.513422 137852455224896 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 02:26:55.513470 137852455224896 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0423 02:26:55.513529 137852455224896 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0423 02:26:55.513567 137852455224896 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0423 02:26:55.513682 137884523878208 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0423 02:26:55.513819 137884523878208 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 02:26:55.514740 137884523878208 metric_logger.py:196] completed step: 19, seconds: 0.134, TFLOP/s/device: 101.135, Tokens/s/device: 15244.107, total_weights: 16384, loss: 1.449, lm_loss: 1.449, perplexity: 4.260 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops