main (51c7f2b8e) vs feat/migrate-nnx-utils (dc4d58891)| Metric | main 51c7f2b8e | feat/migrate-nnx-utils dc4d58891 | Diff (feat/migrate-nnx-utils − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 1.4490 | 1.4490 | 0 |
| TFLOP/s | 114.508 | 114.977 | +0.469 |
| Tok/s | 17259.8 | 17330.5 | +70.691 |
| Avg s/step | 4.558 | 4.938 | +0.38 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-15 22:15:45.371873: 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) I0415 22:15:45.489251 123448993188992 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 22:16:42.873478 123448993188992 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 22:16:42.873588 123448993188992 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 22:16:42.873626 123448993188992 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0415 22:16:42.873660 123448993188992 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 22:16:42.873684 123448993188992 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0415 22:16:42.873794 123448993188992 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 22:16:42.967671 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 22:16:43.382148 123448993188992 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0415 22:16:43.382514 123448993188992 checkpointing.py:677] Setting up checkpoint logger... I0415 22:16:43.382562 123448993188992 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 22:16:43.382637 123448993188992 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 22:16:43.383375 123448993188992 base_pytree_checkpoint_handler.py:411] 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 0x70461f1add90>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 22:16:45.016290 123448993188992 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 22:16:45.016502 123448993188992 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 0x703fa293cbc0>}, handler_registry=None I0415 22:16:45.016847 123448993188992 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x703fa293cbc0>` 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`. I0415 22:16:45.016899 123448993188992 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x703fa293d880>` 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`. I0415 22:16:45.016928 123448993188992 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 0x703fa293cbc0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x703fa293cbc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x703fa293d880>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x703fa293d880>}). I0415 22:16:45.017254 123448993188992 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 22:16:45.017310 123448993188992 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x703fa29353a0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0415 22:16:45.408756 123342883391040 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776291226823400748, 'commit_timestamp_nsecs': 1776291269094302607, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/5/_CHECKPOINT_METADATA I0415 22:16:45.489178 123342883391040 checkpoint_manager.py:1847] Missing metrics for step 5 I0415 22:16:45.519407 123448993188992 checkpoint_manager.py:1788] Found 3 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints I0415 22:16:45.519865 123448993188992 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_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x703fa293c260> I0415 22:16:45.519962 123448993188992 checkpointing.py:301] Checkpoint manager created! W0415 22:16:45.536926 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 22:16:45.718611 123448993188992 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 22:16:45.718708 123448993188992 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 22:16:45.811550 123448993188992 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 22:16:45.811626 123448993188992 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 22:16:45.825885 123448993188992 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 22:16:45.825932 123448993188992 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 22:16:45.846523 123448993188992 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 22:16:45.846578 123448993188992 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 22:16:45.860917 123448993188992 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 22:16:45.860966 123448993188992 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 22:16:45.875312 123448993188992 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 22:16:45.875360 123448993188992 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 22:16:45.897171 123448993188992 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 22:16:45.897227 123448993188992 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 22:16:45.915627 123448993188992 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 22:16:45.915681 123448993188992 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0415 22:16:46.102486 123448993188992 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 22:16:46.102582 123448993188992 checkpointing.py:581] restoring from this run's directory step 9 I0415 22:16:46.281381 123448993188992 checkpointer.py:304] Restoring checkpoint from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/9. I0415 22:16:46.806285 3294311 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 22:16:46.806351 3294311 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 22:16:52.573064 123448993188992 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/read/gbytes_per_sec: 2.073 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 5 seconds) (per-host) I0415 22:16:52.573457 123448993188992 checkpointer.py:318] Finished restoring checkpoint in 6.38 seconds from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/9. I0415 22:16:52.573587 123448993188992 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints', 'checkpointer_start_time': 1776291406.1975315, 'checkpointer_duration_secs': 6.3759925365448, 'checkpoint_manager_start_time': 1776291406.1028256, 'checkpoint_manager_duration_secs': 6.470703601837158} [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 I0415 22:16:52.956852 123448993188992 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0415 22:16:52.956960 123448993188992 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0415 22:16:52.957001 123448993188992 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0415 22:16:52.957058 123448993188992 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0415 22:16:52.957102 123448993188992 maxtext_utils.py:1620] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0415 22:16:52.957129 123448993188992 maxtext_utils.py:1620] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0415 22:16:52.957166 123448993188992 maxtext_utils.py:1620] 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) I0415 22:16:52.957204 123448993188992 maxtext_utils.py:1620] 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') I0415 22:16:52.957233 123448993188992 maxtext_utils.py:1620] 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) I0415 22:16:52.957259 123448993188992 maxtext_utils.py:1620] 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) I0415 22:16:52.957294 123448993188992 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0415 22:16:52.957328 123448993188992 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0415 22:16:53.380434 123448993188992 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 22:16:53.380516 123448993188992 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0415 22:16:53.394158 123448993188992 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 22:16:53.394209 123448993188992 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0415 22:16:54.059412 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 22:16:54.124540 123448993188992 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. I0415 22:16:54.125034 123448993188992 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 22:16:54.125191 123448993188992 metric_logger.py:289] number parameters: 1.104 billion W0415 22:16:54.941079 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 22:16:55.011954 123448993188992 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0415 22:16:55.128859 123448993188992 checkpointing.py:776] Waited 0.11688852310180664 seconds for step 10 to finish before starting checkpointing. I0415 22:16:55.129301 123448993188992 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 22:16:55.129459 123448993188992 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0415 22:16:55.129750 123448993188992 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10. I0415 22:16:55.213442 123448993188992 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0415 22:16:55.293747 123448993188992 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 22:16:55.293835 123448993188992 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 I0415 22:16:55.305340 123341262292544 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10 W0415 22:16:55.517842 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:16:55.540091 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:16:55.544611 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:16:55.548721 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:16:55.552595 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:16:55.556797 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:16:55.560663 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:16:55.564432 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 22:16:55.974024 123341251806784 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10/items I0415 22:16:56.128537 123341220349504 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776291415893097045, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776291415893097045, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10/_CHECKPOINT_METADATA I0415 22:17:04.146846 123448993188992 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 8.853453s I0415 22:17:04.151200 123448993188992 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.381 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 8 seconds) (per-host) I0415 22:17:04.151261 123448993188992 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 8.936941s (batch_requests_ready=0.076427s, total_serialization_initiated=8.856272s, others=0.004243s) I0415 22:17:04.151333 123448993188992 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 8.937474s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=8.937457) I0415 22:17:04.152574 123341272778304 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 22:17:04.152655 123448993188992 async_checkpointer.py:561] Finished blocking save. Time taken: 9.023154s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10. I0415 22:17:04.152838 123448993188992 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0415 22:17:04.153002 123341262292544 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 22:17:04.153103 123448993188992 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776291415.1292872, 'wait_for_prev_duration_secs': 4.1961669921875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776291415.129482, 'checkpointer_blocking_duration_secs': 9.023280382156372, 'get_old_steps_start_time': 1776291424.1527789, 'get_old_steps_duration_secs': 2.384185791015625e-05, 'checkpoint_manager_blocking_start_time': 1776291415.1292038, 'checkpoint_manager_blocking_duration_secs': 9.023872137069702} I0415 22:17:04.153213 123448993188992 checkpointing.py:408] Started an asynchronous checkpoint save for step 10 I0415 22:17:04.153252 123448993188992 max_utils.py:750] Memstats: After params initialized: I0415 22:17:04.153296 123448993188992 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0415 22:17:04.153321 123448993188992 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0415 22:17:04.153340 123448993188992 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0415 22:17:04.153359 123448993188992 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0415 22:17:04.153377 123448993188992 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0415 22:17:04.153397 123448993188992 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0415 22:17:04.153415 123448993188992 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0415 22:17:04.153433 123448993188992 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0415 22:17:04.158851 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:17:04.164761 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 22:17:04.169025 3292719 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 22:17:04.356245 123448993188992 metric_logger.py:185] completed step: 10, seconds: 0.886, TFLOP/s/device: 15.331, Tokens/s/device: 2310.850, total_weights: 16384, loss: 3.602 I0415 22:17:04.472384 123448993188992 metric_logger.py:185] completed step: 11, seconds: 9.343, TFLOP/s/device: 1.454, Tokens/s/device: 219.208, total_weights: 16384, loss: 3.101 I0415 22:17:04.590579 123448993188992 metric_logger.py:185] completed step: 12, seconds: 0.005, TFLOP/s/device: 2687.328, Tokens/s/device: 405063.291, total_weights: 16384, loss: 2.684 I0415 22:17:04.709017 123448993188992 metric_logger.py:185] completed step: 13, seconds: 0.117, TFLOP/s/device: 116.469, Tokens/s/device: 17555.439, total_weights: 16384, loss: 2.348 I0415 22:17:04.713377 123448993188992 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0415 22:17:04.875970 123341230835264 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10/items/array_metadatas/process_0 I0415 22:17:04.945168 123448993188992 checkpointing.py:776] Waited 0.23176813125610352 seconds for step 15 to finish before starting checkpointing. I0415 22:17:04.945541 123448993188992 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 22:17:33.911066 123341251806784 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 29.759162s (commit=29.315727s, array_metadata_write=0.443435s) I0415 22:17:33.912119 123341272778304 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 326.551 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 38 seconds) (per-host) I0415 22:17:33.912177 123341272778304 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 29.759473s. I0415 22:17:34.312394 123341272778304 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 22:17:34.518285 123341220349504 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776291415893097045, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10/_CHECKPOINT_METADATA I0415 22:17:34.765127 123341272778304 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 22:17:34.765699 123341272778304 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.591568s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10/items I0415 22:17:34.766375 123341272778304 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10/items I0415 22:17:35.001126 123341272778304 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10 I0415 22:17:35.680818 123341272778304 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10`. I0415 22:17:35.681423 123341272778304 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 40.551929s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/10 I0415 22:17:35.681498 123341272778304 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 31.528795s. I0415 22:17:35.681680 123341262292544 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 22:17:35.681785 123341262292544 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 22:17:35.681849 123341262292544 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0415 22:17:35.681897 123341262292544 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0415 22:17:35.682282 123448993188992 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. W0415 22:17:35.683244 123448993188992 checkpoint_manager.py:1441] Waiting for previous save to complete took 30.737703 seconds. If this number is high, consider checkpointing less frequently. I0415 22:17:35.684070 123448993188992 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 15 I0415 22:17:35.684354 123448993188992 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15. I0415 22:17:35.878474 123448993188992 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 22:17:35.878561 123448993188992 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 I0415 22:17:35.880227 123341262292544 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15 I0415 22:17:36.583682 123341209863744 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15/items I0415 22:17:39.451666 123448993188992 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 3.573586s I0415 22:17:39.455722 123448993188992 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 3.359 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 3 seconds) (per-host) I0415 22:17:39.455783 123448993188992 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 3.674327s (batch_requests_ready=0.093479s, total_serialization_initiated=3.576895s, others=0.003953s) I0415 22:17:39.455856 123448993188992 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 3.675063s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=3.675048) I0415 22:17:39.457401 123342830962240 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 22:17:39.457484 123448993188992 async_checkpointer.py:561] Finished blocking save. Time taken: 3.773369s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15. I0415 22:17:39.457685 123448993188992 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0415 22:17:39.457840 123341272778304 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 22:17:39.457942 123448993188992 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776291424.9455185, 'wait_for_prev_duration_secs': 30.73770260810852, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776291455.6840954, 'checkpointer_blocking_duration_secs': 3.773502826690674, 'get_old_steps_start_time': 1776291459.4576142, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1776291424.9454522, 'checkpoint_manager_blocking_duration_secs': 34.51246523857117} I0415 22:17:39.458085 123448993188992 checkpointing.py:408] Started an asynchronous checkpoint save for step 15 I0415 22:17:39.458532 123448993188992 metric_logger.py:185] completed step: 14, seconds: 0.118, TFLOP/s/device: 114.793, Tokens/s/device: 17302.851, total_weights: 16384, loss: 2.083 I0415 22:17:39.465489 123448993188992 metric_logger.py:185] completed step: 15, seconds: 0.118, TFLOP/s/device: 114.688, Tokens/s/device: 17287.077, total_weights: 16384, loss: 1.880 I0415 22:17:39.581810 123448993188992 metric_logger.py:185] completed step: 16, seconds: 34.751, TFLOP/s/device: 0.391, Tokens/s/device: 58.934, total_weights: 16384, loss: 1.726 I0415 22:17:39.700253 123448993188992 metric_logger.py:185] completed step: 17, seconds: 0.005, TFLOP/s/device: 2553.012, Tokens/s/device: 384817.738, total_weights: 16384, loss: 1.610 I0415 22:17:39.818475 123448993188992 metric_logger.py:185] completed step: 18, seconds: 0.117, TFLOP/s/device: 115.885, Tokens/s/device: 17467.398, total_weights: 16384, loss: 1.521 I0415 22:17:39.936327 123448993188992 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0415 22:17:39.937187 123448993188992 checkpointing.py:776] Waited 0.0008909702301025391 seconds for step 19 to finish before starting checkpointing. I0415 22:17:39.937424 123448993188992 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 22:17:40.202001 123342841448000 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15/items/array_metadatas/process_0 I0415 22:18:04.662126 123341209863744 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 25.205681s (commit=24.748200s, array_metadata_write=0.457481s) I0415 22:18:04.663063 123342830962240 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 437.538 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 28 seconds) (per-host) I0415 22:18:04.663113 123342830962240 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 25.205586s. I0415 22:18:05.099053 123342830962240 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 22:18:05.542353 123342830962240 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 22:18:05.543015 123342830962240 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.582975s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15/items I0415 22:18:05.543659 123342830962240 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15/items I0415 22:18:05.798960 123342830962240 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15 I0415 22:18:06.485565 123342830962240 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15`. I0415 22:18:06.486201 123342830962240 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 30.802091s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/15 I0415 22:18:06.486269 123342830962240 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 27.028743s. I0415 22:18:06.486440 123341272778304 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 22:18:06.486536 123341272778304 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 22:18:06.486597 123341272778304 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0415 22:18:06.486650 123341272778304 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0415 22:18:06.486899 123448993188992 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. W0415 22:18:06.487486 123448993188992 checkpoint_manager.py:1441] Waiting for previous save to complete took 26.550066 seconds. If this number is high, consider checkpointing less frequently. I0415 22:18:06.487962 123448993188992 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0415 22:18:06.488391 123448993188992 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19. I0415 22:18:06.664440 123341272778304 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19 I0415 22:18:06.670215 123448993188992 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 22:18:06.670305 123448993188992 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 I0415 22:18:07.367842 123342799504960 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19/items I0415 22:18:16.562609 123448993188992 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.892791s I0415 22:18:16.566644 123448993188992 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.236 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0415 22:18:16.566709 123448993188992 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.986058s (batch_requests_ready=0.086097s, total_serialization_initiated=9.896034s, others=0.003928s) I0415 22:18:16.566783 123448993188992 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.986555s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=9.986538) I0415 22:18:16.568367 123341272778304 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 22:18:16.568472 123448993188992 async_checkpointer.py:561] Finished blocking save. Time taken: 10.080466s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19. I0415 22:18:16.568639 123448993188992 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0415 22:18:16.568807 123342830962240 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 22:18:16.568916 123448993188992 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776291459.9373991, 'wait_for_prev_duration_secs': 26.550065994262695, 'time_between_consecutive_saves_sec': 4.255486488342285, 'checkpointer_blocking_start_time': 1776291486.487988, 'checkpointer_blocking_duration_secs': 10.080584526062012, 'get_old_steps_start_time': 1776291496.5685835, 'get_old_steps_duration_secs': 2.1457672119140625e-05, 'checkpoint_manager_blocking_start_time': 1776291459.9373674, 'checkpoint_manager_blocking_duration_secs': 36.63152289390564} I0415 22:18:16.569040 123448993188992 checkpointing.py:408] Started an asynchronous checkpoint save for step 19 I0415 22:18:16.569085 123448993188992 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 22:18:17.315836 123342841448000 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19/items/array_metadatas/process_0 I0415 22:18:52.185615 123342799504960 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 35.618229s (commit=35.163767s, array_metadata_write=0.454462s) I0415 22:18:52.186681 123341272778304 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 277.086 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 45 seconds) (per-host) I0415 22:18:52.186742 123341272778304 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.618228s. I0415 22:18:52.639396 123341272778304 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 22:18:53.056246 123341272778304 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 22:18:53.056912 123341272778304 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.571064s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19/items I0415 22:18:53.057581 123341272778304 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19/items I0415 22:18:53.304157 123341272778304 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19 I0415 22:18:53.978709 123341272778304 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19`. I0415 22:18:53.979351 123341272778304 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 47.491352s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_16_zarr3_true/checkpoints/19 I0415 22:18:53.979417 123341272778304 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 37.410903s. I0415 22:18:53.979597 123342830962240 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 22:18:53.979705 123342830962240 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 22:18:53.979766 123342830962240 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0415 22:18:53.979811 123342830962240 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0415 22:18:53.979957 123448993188992 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. I0415 22:18:53.980112 123448993188992 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 22:18:53.981008 123448993188992 metric_logger.py:185] completed step: 19, seconds: 0.119, TFLOP/s/device: 114.508, Tokens/s/device: 17259.833, total_weights: 16384, loss: 1.449 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
2026-04-16 00:18:25.170001: 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) I0416 00:18:25.285876 132748576648320 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. ~/maxtext_venv/lib/python3.12/site-packages/jax/_src/xla_bridge.py:219: UserWarning: TPU backend initialization is taking more than 60.0 seconds. Did you run your code on all TPU hosts? See https://docs.jax.dev/en/latest/multi_process.html for more information. warnings.warn( I0416 00:19:25.885281 132748576648320 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 00:19:25.885391 132748576648320 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 00:19:25.885425 132748576648320 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0416 00:19:25.885452 132748576648320 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0416 00:19:25.885475 132748576648320 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 00:19:25.885546 132748576648320 train.py:715] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 00:19:25.974701 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 00:19:26.373174 132748576648320 maxtext_utils.py:1520] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 00:19:26.373428 132748576648320 checkpointing.py:677] Setting up checkpoint logger... I0416 00:19:26.373470 132748576648320 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 00:19:26.373505 132748576648320 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 00:19:26.373938 132748576648320 base_pytree_checkpoint_handler.py:411] 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 0x78bb58df8ad0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 00:19:27.974922 132748576648320 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0416 00:19:27.975133 132748576648320 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 0x78b4dcb42b40>}, handler_registry=None I0416 00:19:27.975370 132748576648320 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78b4dcb42b40>` 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`. I0416 00:19:27.975411 132748576648320 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b6540a86b0>` 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`. I0416 00:19:27.975439 132748576648320 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 0x78b4dcb42b40>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78b4dcb42b40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b6540a86b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b6540a86b0>}). I0416 00:19:27.975728 132748576648320 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 00:19:27.975777 132748576648320 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x78b4dcb49580> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 00:19:28.369177 132642435499584 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776298538562580504, 'commit_timestamp_nsecs': 1776298575746587943, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/0/_CHECKPOINT_METADATA I0416 00:19:28.460925 132642435499584 checkpoint_manager.py:1847] Missing metrics for step 0 I0416 00:19:28.497839 132748576648320 checkpoint_manager.py:1788] Found 3 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints I0416 00:19:28.498309 132748576648320 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_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x78b4dcb42f90> I0416 00:19:28.498396 132748576648320 checkpointing.py:301] Checkpoint manager created! W0416 00:19:28.514915 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 00:19:28.693499 132748576648320 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 00:19:28.693603 132748576648320 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 00:19:28.777261 132748576648320 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 00:19:28.777341 132748576648320 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 00:19:28.791589 132748576648320 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 00:19:28.791639 132748576648320 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 00:19:28.812163 132748576648320 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 00:19:28.812220 132748576648320 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 00:19:28.826590 132748576648320 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 00:19:28.826639 132748576648320 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 00:19:28.843071 132748576648320 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 00:19:28.843122 132748576648320 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 00:19:28.864751 132748576648320 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 00:19:28.864809 132748576648320 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 00:19:28.883168 132748576648320 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 00:19:28.883228 132748576648320 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 00:19:29.066780 132748576648320 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0416 00:19:29.066877 132748576648320 checkpointing.py:581] restoring from this run's directory step 9 I0416 00:19:29.237802 132748576648320 checkpointer.py:304] Restoring checkpoint from gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/9. I0416 00:19:29.779413 3424856 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 00:19:29.779482 3424856 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 00:19:35.742447 132748576648320 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/read/gbytes_per_sec: 2.003 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 6 seconds) (per-host) I0416 00:19:35.742934 132748576648320 checkpointer.py:318] Finished restoring checkpoint in 6.58 seconds from gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/9. I0416 00:19:35.743112 132748576648320 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints', 'checkpointer_start_time': 1776298769.1608663, 'checkpointer_duration_secs': 6.582141637802124, 'checkpoint_manager_start_time': 1776298769.0671022, 'checkpoint_manager_duration_secs': 6.6759092807769775} [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 I0416 00:19:36.365566 132748576648320 maxtext_utils.py:1623] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 00:19:36.365680 132748576648320 maxtext_utils.py:1623] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 00:19:36.365721 132748576648320 maxtext_utils.py:1623] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 00:19:36.365764 132748576648320 maxtext_utils.py:1623] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 00:19:36.365803 132748576648320 maxtext_utils.py:1623] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 00:19:36.365831 132748576648320 maxtext_utils.py:1623] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 00:19:36.365880 132748576648320 maxtext_utils.py:1623] 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) I0416 00:19:36.365918 132748576648320 maxtext_utils.py:1623] 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') I0416 00:19:36.365949 132748576648320 maxtext_utils.py:1623] 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) I0416 00:19:36.365976 132748576648320 maxtext_utils.py:1623] 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) I0416 00:19:36.366017 132748576648320 maxtext_utils.py:1623] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed', 'vocab') Physical: ('fsdp', None) I0416 00:19:36.366109 132748576648320 maxtext_utils.py:1623] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed') Physical: (None, 'fsdp') I0416 00:19:36.898998 132748576648320 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 00:19:36.899091 132748576648320 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 00:19:36.912820 132748576648320 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 00:19:36.912867 132748576648320 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 00:19:37.646493 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 00:19:37.711062 132748576648320 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. I0416 00:19:37.711515 132748576648320 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 00:19:37.711671 132748576648320 metric_logger.py:289] number parameters: 1.104 billion W0416 00:19:38.682472 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 00:19:38.756138 132748576648320 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0416 00:19:38.872873 132748576648320 checkpointing.py:776] Waited 0.11672258377075195 seconds for step 10 to finish before starting checkpointing. I0416 00:19:38.873307 132748576648320 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 00:19:38.873473 132748576648320 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0416 00:19:38.873770 132748576648320 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10. I0416 00:19:38.956170 132748576648320 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 00:19:39.051599 132748576648320 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 00:19:39.051740 132640747292224 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10 I0416 00:19:39.051817 132748576648320 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 W0416 00:19:39.272139 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:39.289097 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:39.301646 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:39.306120 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:39.310110 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:39.313898 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:39.317744 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:39.321574 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 00:19:39.770523 132640736806464 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10/items I0416 00:19:39.927295 132640705349184 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776298779685173287, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776298779685173287, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10/_CHECKPOINT_METADATA I0416 00:19:48.160719 132748576648320 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.109851s I0416 00:19:48.165038 132748576648320 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.340 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0416 00:19:48.165110 132748576648320 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.208023s (batch_requests_ready=0.090758s, total_serialization_initiated=9.113043s, others=0.004222s) I0416 00:19:48.165179 132748576648320 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.208559s (all_items=0.000019s, per_item={'items': '0.00001931'}, temp_paths=9.208540) I0416 00:19:48.166572 132640747292224 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 00:19:48.166686 132748576648320 async_checkpointer.py:561] Finished blocking save. Time taken: 9.293173s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10. I0416 00:19:48.166891 132748576648320 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0416 00:19:48.167073 132640757777984 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 00:19:48.167167 132748576648320 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776298778.873294, 'wait_for_prev_duration_secs': 4.076957702636719e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776298778.8734946, 'checkpointer_blocking_duration_secs': 9.29332184791565, 'get_old_steps_start_time': 1776298788.1668305, 'get_old_steps_duration_secs': 2.3365020751953125e-05, 'checkpoint_manager_blocking_start_time': 1776298778.8732097, 'checkpoint_manager_blocking_duration_secs': 9.293932676315308} I0416 00:19:48.167281 132748576648320 checkpointing.py:408] Started an asynchronous checkpoint save for step 10 I0416 00:19:48.167325 132748576648320 max_utils.py:750] Memstats: After params initialized: I0416 00:19:48.167372 132748576648320 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0416 00:19:48.167397 132748576648320 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 00:19:48.167418 132748576648320 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 00:19:48.167436 132748576648320 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 00:19:48.167454 132748576648320 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 00:19:48.167472 132748576648320 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 00:19:48.167490 132748576648320 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 00:19:48.167506 132748576648320 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 00:19:48.173116 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:48.178883 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 00:19:48.183220 3423223 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 00:19:48.368904 132748576648320 metric_logger.py:185] completed step: 10, seconds: 1.044, TFLOP/s/device: 13.015, Tokens/s/device: 1961.729, total_weights: 16384, loss: 3.602 I0416 00:19:48.485387 132748576648320 metric_logger.py:185] completed step: 11, seconds: 9.611, TFLOP/s/device: 1.414, Tokens/s/device: 213.085, total_weights: 16384, loss: 3.101 I0416 00:19:48.603787 132748576648320 metric_logger.py:185] completed step: 12, seconds: 0.005, TFLOP/s/device: 2590.492, Tokens/s/device: 390467.112, total_weights: 16384, loss: 2.684 I0416 00:19:48.722510 132748576648320 metric_logger.py:185] completed step: 13, seconds: 0.117, TFLOP/s/device: 115.992, Tokens/s/device: 17483.652, total_weights: 16384, loss: 2.348 I0416 00:19:48.727276 132748576648320 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0416 00:19:48.903590 132640715834944 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10/items/array_metadatas/process_0 I0416 00:19:48.959584 132748576648320 checkpointing.py:776] Waited 0.23227739334106445 seconds for step 15 to finish before starting checkpointing. I0416 00:19:48.959974 132748576648320 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 00:20:13.948669 132640736806464 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 25.782928s (commit=25.347514s, array_metadata_write=0.435414s) I0416 00:20:13.949858 132640747292224 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 361.126 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 34 seconds) (per-host) I0416 00:20:13.949967 132640747292224 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 25.783221s. I0416 00:20:14.373386 132640747292224 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 00:20:14.569108 132640705349184 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776298779685173287, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10/_CHECKPOINT_METADATA I0416 00:20:14.810157 132640747292224 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 00:20:14.810754 132640747292224 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.585026s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10/items I0416 00:20:14.811403 132640747292224 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10/items I0416 00:20:15.065009 132640747292224 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10 I0416 00:20:15.738166 132640747292224 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10`. I0416 00:20:15.738781 132640747292224 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 36.865274s. directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/10 I0416 00:20:15.738861 132640747292224 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 27.572116s. I0416 00:20:15.739073 132640757777984 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 00:20:15.739179 132640757777984 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 00:20:15.739247 132640757777984 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0416 00:20:15.739298 132640757777984 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0416 00:20:15.739434 132748576648320 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. W0416 00:20:15.739554 132748576648320 checkpoint_manager.py:1441] Waiting for previous save to complete took 26.779577 seconds. If this number is high, consider checkpointing less frequently. I0416 00:20:15.740388 132748576648320 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 15 I0416 00:20:15.740663 132748576648320 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15. I0416 00:20:15.911786 132640757777984 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15 I0416 00:20:15.923692 132748576648320 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 00:20:15.923784 132748576648320 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 I0416 00:20:16.644987 132640694863424 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15/items I0416 00:20:26.837224 132748576648320 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.913939s I0416 00:20:26.841510 132748576648320 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.120 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 11 seconds) (per-host) I0416 00:20:26.841576 132748576648320 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 11.014129s (batch_requests_ready=0.092064s, total_serialization_initiated=10.917875s, others=0.004190s) I0416 00:20:26.841652 132748576648320 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 11.014877s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=11.014861) I0416 00:20:26.843073 132640757777984 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 00:20:26.843157 132748576648320 async_checkpointer.py:561] Finished blocking save. Time taken: 11.102722s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15. I0416 00:20:26.843351 132748576648320 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0416 00:20:26.843526 132640747292224 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 00:20:26.843654 132748576648320 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776298788.959951, 'wait_for_prev_duration_secs': 26.779577255249023, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776298815.7404137, 'checkpointer_blocking_duration_secs': 11.1028470993042, 'get_old_steps_start_time': 1776298826.8432813, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1776298788.9598806, 'checkpoint_manager_blocking_duration_secs': 37.883747816085815} I0416 00:20:26.843748 132748576648320 checkpointing.py:408] Started an asynchronous checkpoint save for step 15 I0416 00:20:26.844188 132748576648320 metric_logger.py:185] completed step: 14, seconds: 0.119, TFLOP/s/device: 114.591, Tokens/s/device: 17272.352, total_weights: 16384, loss: 2.083 I0416 00:20:26.851521 132748576648320 metric_logger.py:185] completed step: 15, seconds: 0.119, TFLOP/s/device: 114.324, Tokens/s/device: 17232.095, total_weights: 16384, loss: 1.880 I0416 00:20:26.967973 132748576648320 metric_logger.py:185] completed step: 16, seconds: 38.123, TFLOP/s/device: 0.356, Tokens/s/device: 53.722, total_weights: 16384, loss: 1.726 I0416 00:20:27.086280 132748576648320 metric_logger.py:185] completed step: 17, seconds: 0.005, TFLOP/s/device: 2532.078, Tokens/s/device: 381662.318, total_weights: 16384, loss: 1.610 I0416 00:20:27.204942 132748576648320 metric_logger.py:185] completed step: 18, seconds: 0.117, TFLOP/s/device: 115.651, Tokens/s/device: 17432.161, total_weights: 16384, loss: 1.521 I0416 00:20:27.322648 132748576648320 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0416 00:20:27.323521 132748576648320 checkpointing.py:776] Waited 0.0008940696716308594 seconds for step 19 to finish before starting checkpointing. I0416 00:20:27.323760 132748576648320 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 00:20:27.629486 132642425013824 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15/items/array_metadatas/process_0 I0416 00:20:59.800201 132640694863424 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.957981s (commit=32.531183s, array_metadata_write=0.426798s) I0416 00:20:59.801285 132640757777984 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 287.371 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 43 seconds) (per-host) I0416 00:20:59.801340 132640757777984 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.958138s. I0416 00:21:00.217698 132640757777984 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 00:21:00.618077 132640757777984 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 00:21:00.618242 132640757777984 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.540148s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15/items I0416 00:21:00.618897 132640757777984 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15/items I0416 00:21:00.858939 132640757777984 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15 I0416 00:21:01.522944 132640757777984 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15`. I0416 00:21:01.523548 132640757777984 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 45.783120s. directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/15 I0416 00:21:01.523627 132640757777984 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.680426s. I0416 00:21:01.523817 132640747292224 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 00:21:01.523923 132640747292224 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 00:21:01.523998 132640747292224 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0416 00:21:01.524040 132640747292224 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0416 00:21:01.525070 132748576648320 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. W0416 00:21:01.525190 132748576648320 checkpoint_manager.py:1441] Waiting for previous save to complete took 34.201433 seconds. If this number is high, consider checkpointing less frequently. I0416 00:21:01.525403 132748576648320 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0416 00:21:01.525687 132748576648320 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19. I0416 00:21:01.717949 132748576648320 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 00:21:01.718034 132748576648320 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 I0416 00:21:01.725517 132640747292224 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19 I0416 00:21:02.443311 132640694863424 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19/items I0416 00:21:15.073663 132748576648320 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 13.356105s I0416 00:21:15.077924 132748576648320 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 938.988 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 13 seconds) (per-host) I0416 00:21:15.077988 132748576648320 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 13.457964s (batch_requests_ready=0.094235s, total_serialization_initiated=13.359577s, others=0.004153s) I0416 00:21:15.078072 132748576648320 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 13.458703s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=13.458689) I0416 00:21:15.079573 132642372585024 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 00:21:15.079654 132748576648320 async_checkpointer.py:561] Finished blocking save. Time taken: 13.554210s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19. I0416 00:21:15.079828 132748576648320 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0416 00:21:15.080053 132640757777984 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 00:21:15.080161 132748576648320 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776298827.3237355, 'wait_for_prev_duration_secs': 34.20143270492554, 'time_between_consecutive_saves_sec': 11.58442211151123, 'checkpointer_blocking_start_time': 1776298861.5254264, 'checkpointer_blocking_duration_secs': 13.554329872131348, 'get_old_steps_start_time': 1776298875.0797718, 'get_old_steps_duration_secs': 2.09808349609375e-05, 'checkpoint_manager_blocking_start_time': 1776298827.3237038, 'checkpoint_manager_blocking_duration_secs': 47.756433963775635} I0416 00:21:15.080280 132748576648320 checkpointing.py:408] Started an asynchronous checkpoint save for step 19 I0416 00:21:15.080314 132748576648320 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 00:21:15.856247 132640896190016 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19/items/array_metadatas/process_0 I0416 00:21:54.643676 132640694863424 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 39.564920s (commit=39.131826s, array_metadata_write=0.433095s) I0416 00:21:54.644614 132642372585024 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 238.319 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 53 seconds) (per-host) I0416 00:21:54.644678 132642372585024 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.564972s. I0416 00:21:55.046011 132642372585024 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 00:21:55.453451 132642372585024 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 00:21:55.454133 132642372585024 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.548239s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19/items I0416 00:21:55.454771 132642372585024 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19/items I0416 00:21:55.703580 132642372585024 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19 I0416 00:21:56.399721 132642372585024 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19`. I0416 00:21:56.400349 132642372585024 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 54.874910s. directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_16_zarr3_true/checkpoints/19 I0416 00:21:56.400418 132642372585024 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 41.320721s. I0416 00:21:56.400597 132640757777984 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 00:21:56.400704 132640757777984 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 00:21:56.400773 132640757777984 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0416 00:21:56.400821 132640757777984 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0416 00:21:56.401203 132748576648320 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. I0416 00:21:56.401341 132748576648320 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 00:21:56.402009 132748576648320 metric_logger.py:185] completed step: 19, seconds: 0.118, TFLOP/s/device: 114.977, Tokens/s/device: 17330.524, total_weights: 16384, loss: 1.449 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops