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 | 114.421 | 113.472 | -0.949 |
| Tok/s | 17246.8 | 17103.7 | -143.026 |
| Avg s/step | 7.907 | 9.023 | +1.116 |
| 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 21:57:11.739230: 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 21:57:11.856290 131993809726592 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 21:57:56.856376 131993809726592 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 21:57:56.856489 131993809726592 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 21:57:56.856523 131993809726592 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 21:57:56.856559 131993809726592 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 21:57:56.856583 131993809726592 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0415 21:57:56.856690 131993809726592 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 21:57:56.948543 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:57:57.346002 131993809726592 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0415 21:57:57.346321 131993809726592 checkpointing.py:677] Setting up checkpoint logger... I0415 21:57:57.346365 131993809726592 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 21:57:57.346410 131993809726592 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 21:57:57.346931 131993809726592 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 0x780b9c903b00>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 21:57:59.002804 131993809726592 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 21:57:59.002992 131993809726592 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 0x78052239f620>}, handler_registry=None I0415 21:57:59.003274 131993809726592 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78052239f620>` 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 21:57:59.003317 131993809726592 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78052113a2d0>` 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 21:57:59.003347 131993809726592 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 0x78052239f620>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78052239f620>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78052113a2d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78052113a2d0>}). I0415 21:57:59.003653 131993809726592 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 21:57:59.386978 131887706146368 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776290092648198647, 'commit_timestamp_nsecs': 1776290127897450198, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/0/_CHECKPOINT_METADATA I0415 21:57:59.471681 131887706146368 checkpoint_manager.py:1847] Missing metrics for step 0 I0415 21:57:59.565236 131993809726592 checkpoint_manager.py:1788] Found 3 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints I0415 21:57:59.565745 131993809726592 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=False, 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_14_async_ckpt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x780521139cd0> I0415 21:57:59.565831 131993809726592 checkpointing.py:301] Checkpoint manager created! W0415 21:57:59.582573 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:57:59.764981 131993809726592 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 21:57:59.765101 131993809726592 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:57:59.854565 131993809726592 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:57:59.854645 131993809726592 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:57:59.868992 131993809726592 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:57:59.869039 131993809726592 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:57:59.889507 131993809726592 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 21:57:59.889566 131993809726592 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:57:59.903906 131993809726592 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 21:57:59.903954 131993809726592 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:57:59.918211 131993809726592 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 21:57:59.918261 131993809726592 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:57:59.939906 131993809726592 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 21:57:59.939963 131993809726592 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:57:59.958338 131993809726592 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 21:57:59.958389 131993809726592 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0415 21:58:00.142020 131993809726592 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 21:58:00.142201 131993809726592 checkpointing.py:581] restoring from this run's directory step 9 I0415 21:58:00.305421 131993809726592 checkpointer.py:304] Restoring checkpoint from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/9. I0415 21:58:00.855263 3277122 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 21:58:00.855327 3277122 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 21:58:07.420892 131993809726592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/read/gbytes_per_sec: 1.823 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 6 seconds) (per-host) I0415 21:58:07.421162 131993809726592 checkpointer.py:318] Finished restoring checkpoint in 7.20 seconds from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/9. I0415 21:58:07.421234 131993809726592 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints', 'checkpointer_start_time': 1776290280.2229471, 'checkpointer_duration_secs': 7.198258876800537, 'checkpoint_manager_start_time': 1776290280.1424315, 'checkpoint_manager_duration_secs': 7.278776407241821} [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 21:58:07.795490 131993809726592 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0415 21:58:07.795594 131993809726592 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 21:58:07.795634 131993809726592 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 21:58:07.795677 131993809726592 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 21:58:07.795716 131993809726592 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 21:58:07.795743 131993809726592 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 21:58:07.795781 131993809726592 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 21:58:07.795820 131993809726592 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 21:58:07.795847 131993809726592 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 21:58:07.795875 131993809726592 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 21:58:07.795918 131993809726592 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0415 21:58:07.795954 131993809726592 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0415 21:58:08.215910 131993809726592 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:58:08.215992 131993809726592 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0415 21:58:08.229661 131993809726592 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:58:08.229712 131993809726592 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0415 21:58:08.866636 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:58:08.929831 131993809726592 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 21:58:08.930559 131993809726592 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 21:58:08.930712 131993809726592 metric_logger.py:289] number parameters: 1.104 billion W0415 21:58:09.746987 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:58:09.816917 131993809726592 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0415 21:58:09.933501 131993809726592 checkpointing.py:776] Waited 0.11657452583312988 seconds for step 10 to finish before starting checkpointing. I0415 21:58:09.933900 131993809726592 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:58:09.934099 131993809726592 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0415 21:58:09.934165 131993809726592 checkpointer.py:241] [process=0] Started saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10. I0415 21:58:10.092360 131993809726592 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10 I0415 21:58:10.879230 131993809726592 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776290290643437069, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776290290643437069, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10/_CHECKPOINT_METADATA I0415 21:58:10.880146 131993809726592 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient W0415 21:58:10.880456 131993809726592 future.py:328] [process=0][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 21:58:10.967469 131886198294080 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10/items I0415 21:58:11.495870 131993809726592 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:58:11.495976 131993809726592 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 W0415 21:58:11.729396 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:11.759497 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:11.764769 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:11.769112 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:11.773274 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:11.777228 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:11.781127 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:11.785004 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:58:16.426450 131993809726592 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 4.930941s I0415 21:58:16.430808 131993809726592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.457 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 5 seconds) (per-host) I0415 21:58:16.430873 131993809726592 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 5.023098s (batch_requests_ready=0.085109s, total_serialization_initiated=4.933730s, others=0.004260s) I0415 21:58:16.430949 131993809726592 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 5.550868s (all_items=0.000029s, per_item={'items': '0.00002933'}, temp_paths=5.550839) W0415 21:58:16.430985 131993809726592 future.py:328] [process=0][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 21:58:17.180000 131886208779840 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_14_async_ckpt_false/checkpoints/10/items/array_metadatas/process_0 I0415 21:58:48.194980 131886573684288 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 31.763762s (commit=31.299990s, array_metadata_write=0.463772s) I0415 21:58:48.196084 131993809726592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 343.501 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 36 seconds) (per-host) I0415 21:58:48.842190 131993809726592 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776290290643437069, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10/_CHECKPOINT_METADATA I0415 21:58:48.982028 131993809726592 array_metadata_store.py:367] [process=0][thread=MainThread] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 21:58:49.395003 131993809726592 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:58:49.395188 131993809726592 base_pytree_checkpoint_handler.py:1342] [process=0][thread=MainThread] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.552620s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10/items I0415 21:58:49.395869 131993809726592 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10/items I0415 21:58:49.640710 131993809726592 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10 I0415 21:58:50.301972 131993809726592 atomicity.py:794] [process=0][thread=MainThread] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10`. I0415 21:58:50.302628 131993809726592 checkpointer.py:290] Finished synchronous save in 40.37 seconds to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/10 I0415 21:58:50.302825 131993809726592 checkpoint_manager.py:2103] [process=0][thread=MainThread][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:58:50.302881 131993809726592 checkpoint_manager.py:2112] [process=0][thread=MainThread][step=10] CheckpointManager Save Finalize is done on all hosts. I0415 21:58:50.302906 131993809726592 checkpoint_manager.py:1570] [process=0][thread=MainThread][step=10] Finished synchronous save. I0415 21:58:50.302944 131993809726592 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776290289.9338877, 'wait_for_prev_duration_secs': 3.981590270996094e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776290289.934122, 'checkpointer_blocking_duration_secs': 40.36855459213257, 'get_old_steps_start_time': 1776290330.302705, 'get_old_steps_duration_secs': 3.790855407714844e-05, 'checkpoint_manager_blocking_start_time': 1776290289.9338043, 'checkpoint_manager_blocking_duration_secs': 40.36911678314209} I0415 21:58:50.303068 131993809726592 checkpointing.py:410] Saved a checkpoint at step 10. I0415 21:58:50.303111 131993809726592 max_utils.py:750] Memstats: After params initialized: I0415 21:58:50.303167 131993809726592 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0415 21:58:50.303192 131993809726592 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0415 21:58:50.303213 131993809726592 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0415 21:58:50.303232 131993809726592 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0415 21:58:50.303251 131993809726592 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0415 21:58:50.303269 131993809726592 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0415 21:58:50.303287 131993809726592 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0415 21:58:50.303304 131993809726592 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0415 21:58:50.309797 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:50.315920 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:58:50.320649 3275529 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:58:50.515680 131993809726592 metric_logger.py:185] completed step: 10, seconds: 0.886, TFLOP/s/device: 15.341, Tokens/s/device: 2312.311, total_weights: 16384, loss: 3.602 I0415 21:58:50.629705 131993809726592 metric_logger.py:185] completed step: 11, seconds: 40.695, TFLOP/s/device: 0.334, Tokens/s/device: 50.325, total_weights: 16384, loss: 3.101 I0415 21:58:50.747831 131993809726592 metric_logger.py:185] completed step: 12, seconds: 0.007, TFLOP/s/device: 1919.086, Tokens/s/device: 289265.537, total_weights: 16384, loss: 2.684 I0415 21:58:50.865785 131993809726592 metric_logger.py:185] completed step: 13, seconds: 0.115, TFLOP/s/device: 118.553, Tokens/s/device: 17869.608, total_weights: 16384, loss: 2.348 I0415 21:58:50.870487 131993809726592 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0415 21:58:51.102278 131993809726592 checkpointing.py:776] Waited 0.2317638397216797 seconds for step 15 to finish before starting checkpointing. I0415 21:58:51.102660 131993809726592 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:58:51.102807 131993809726592 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 15 I0415 21:58:51.102867 131993809726592 checkpointer.py:241] [process=0] Started saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/15. I0415 21:58:51.283343 131993809726592 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/15 W0415 21:58:52.121914 131993809726592 future.py:328] [process=0][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 21:58:52.203243 131886573684288 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/15/items I0415 21:58:52.756185 131993809726592 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:58:52.756293 131993809726592 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 21:58:55.488718 131993809726592 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 2.732868s I0415 21:58:55.492771 131993809726592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.383 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 2 seconds) (per-host) I0415 21:58:55.492830 131993809726592 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 2.815740s (batch_requests_ready=0.075772s, total_serialization_initiated=2.736027s, others=0.003941s) I0415 21:58:55.492898 131993809726592 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 3.371320s (all_items=0.000030s, per_item={'items': '0.00003028'}, temp_paths=3.371289) W0415 21:58:55.492930 131993809726592 future.py:328] [process=0][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 21:58:56.253896 131886198294080 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_14_async_ckpt_false/checkpoints/15/items/array_metadatas/process_0 I0415 21:59:25.679255 131886552712768 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 30.185982s (commit=29.777135s, array_metadata_write=0.408848s) I0415 21:59:25.680318 131993809726592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 382.896 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 33 seconds) (per-host) I0415 21:59:26.418927 131993809726592 array_metadata_store.py:367] [process=0][thread=MainThread] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 21:59:26.832029 131993809726592 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:59:26.832461 131993809726592 base_pytree_checkpoint_handler.py:1342] [process=0][thread=MainThread] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.551964s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/15/items I0415 21:59:26.833131 131993809726592 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/15/items I0415 21:59:27.069664 131993809726592 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/15 I0415 21:59:27.751036 131993809726592 atomicity.py:794] [process=0][thread=MainThread] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/15`. I0415 21:59:27.751713 131993809726592 checkpointer.py:290] Finished synchronous save in 36.65 seconds to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/15 I0415 21:59:27.751889 131993809726592 checkpoint_manager.py:2103] [process=0][thread=MainThread][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:59:27.751934 131993809726592 checkpoint_manager.py:2112] [process=0][thread=MainThread][step=15] CheckpointManager Save Finalize is done on all hosts. I0415 21:59:27.751958 131993809726592 checkpoint_manager.py:1570] [process=0][thread=MainThread][step=15] Finished synchronous save. I0415 21:59:27.751998 131993809726592 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776290331.1026332, 'wait_for_prev_duration_secs': 5.245208740234375e-05, 'time_between_consecutive_saves_sec': 0.7997355461120605, 'checkpointer_blocking_start_time': 1776290331.1028266, 'checkpointer_blocking_duration_secs': 36.648937940597534, 'get_old_steps_start_time': 1776290367.751795, 'get_old_steps_duration_secs': 3.266334533691406e-05, 'checkpoint_manager_blocking_start_time': 1776290331.1025627, 'checkpoint_manager_blocking_duration_secs': 36.64941048622131} I0415 21:59:27.752121 131993809726592 checkpointing.py:410] Saved a checkpoint at step 15. I0415 21:59:27.752944 131993809726592 metric_logger.py:185] completed step: 14, seconds: 0.118, TFLOP/s/device: 114.871, Tokens/s/device: 17314.553, total_weights: 16384, loss: 2.083 I0415 21:59:27.765092 131993809726592 metric_logger.py:185] completed step: 15, seconds: 0.118, TFLOP/s/device: 115.139, Tokens/s/device: 17355.050, total_weights: 16384, loss: 1.880 I0415 21:59:27.877573 131993809726592 metric_logger.py:185] completed step: 16, seconds: 36.890, TFLOP/s/device: 0.368, Tokens/s/device: 55.517, total_weights: 16384, loss: 1.726 I0415 21:59:27.996573 131993809726592 metric_logger.py:185] completed step: 17, seconds: 0.008, TFLOP/s/device: 1616.553, Tokens/s/device: 243664.485, total_weights: 16384, loss: 1.610 I0415 21:59:28.114763 131993809726592 metric_logger.py:185] completed step: 18, seconds: 0.114, TFLOP/s/device: 119.657, Tokens/s/device: 18035.949, total_weights: 16384, loss: 1.521 I0415 21:59:28.232842 131993809726592 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0415 21:59:28.233614 131993809726592 checkpointing.py:776] Waited 0.0007996559143066406 seconds for step 19 to finish before starting checkpointing. I0415 21:59:28.233854 131993809726592 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:59:28.233996 131993809726592 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0415 21:59:28.234077 131993809726592 checkpointer.py:241] [process=0] Started saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/19. I0415 21:59:28.404707 131993809726592 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/19 W0415 21:59:29.225137 131993809726592 future.py:328] [process=0][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 21:59:29.313645 131886573684288 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/19/items I0415 21:59:29.859277 131993809726592 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:59:29.859386 131993809726592 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 21:59:30.988761 131993809726592 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.129838s I0415 21:59:30.992779 131993809726592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 10.047 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0415 21:59:30.992837 131993809726592 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.228438s (batch_requests_ready=0.091685s, total_serialization_initiated=1.132841s, others=0.003912s) I0415 21:59:30.992911 131993809726592 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.768115s (all_items=0.000027s, per_item={'items': '0.00002742'}, temp_paths=1.768088) W0415 21:59:30.992943 131993809726592 future.py:328] [process=0][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 21:59:31.783295 131886552712768 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_14_async_ckpt_false/checkpoints/19/items/array_metadatas/process_0 I0415 22:00:05.505312 131887653717568 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 34.512007s (commit=34.025522s, array_metadata_write=0.486485s) I0415 22:00:05.506546 131993809726592 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 353.555 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 35 seconds) (per-host) I0415 22:00:06.264228 131993809726592 array_metadata_store.py:367] [process=0][thread=MainThread] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 22:00:06.738245 131993809726592 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 22:00:06.738450 131993809726592 base_pytree_checkpoint_handler.py:1342] [process=0][thread=MainThread] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.613410s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/19/items I0415 22:00:06.739136 131993809726592 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/19/items I0415 22:00:06.977102 131993809726592 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/19 I0415 22:00:07.680433 131993809726592 atomicity.py:794] [process=0][thread=MainThread] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/19`. I0415 22:00:07.681078 131993809726592 checkpointer.py:290] Finished synchronous save in 39.45 seconds to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints/19 I0415 22:00:07.681226 131993809726592 checkpoint_manager.py:2103] [process=0][thread=MainThread][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0415 22:00:07.681267 131993809726592 checkpoint_manager.py:2112] [process=0][thread=MainThread][step=19] CheckpointManager Save Finalize is done on all hosts. I0415 22:00:07.681291 131993809726592 checkpoint_manager.py:1570] [process=0][thread=MainThread][step=19] Finished synchronous save. I0415 22:00:07.681326 131993809726592 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776290368.233834, 'wait_for_prev_duration_secs': 4.315376281738281e-05, 'time_between_consecutive_saves_sec': 0.4818856716156006, 'checkpointer_blocking_start_time': 1776290368.2340157, 'checkpointer_blocking_duration_secs': 39.4471070766449, 'get_old_steps_start_time': 1776290407.6811452, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1776290368.233801, 'checkpoint_manager_blocking_duration_secs': 39.4475040435791} I0415 22:00:07.681412 131993809726592 checkpointing.py:410] Saved a checkpoint at step 19. I0415 22:00:07.681437 131993809726592 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 22:00:07.681457 131993809726592 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 22:00:07.682186 131993809726592 metric_logger.py:185] completed step: 19, seconds: 0.119, TFLOP/s/device: 114.421, Tokens/s/device: 17246.751, 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
2026-04-15 23:58:01.723162: 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 23:58:01.838528 133849967262848 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 23:59:01.087951 133849967262848 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 23:59:01.088077 133849967262848 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 23:59:01.088114 133849967262848 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 23:59:01.088142 133849967262848 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 23:59:01.088170 133849967262848 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0415 23:59:01.088248 133849967262848 train.py:715] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 23:59:01.179870 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:59:01.576450 133849967262848 maxtext_utils.py:1520] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0415 23:59:01.576772 133849967262848 checkpointing.py:677] Setting up checkpoint logger... I0415 23:59:01.576817 133849967262848 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 23:59:01.576857 133849967262848 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 23:59:01.577429 133849967262848 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 0x79bbc8df8ad0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 23:59:03.259343 133849967262848 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 23:59:03.259512 133849967262848 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 0x79b54cb31400>}, handler_registry=None I0415 23:59:03.259776 133849967262848 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79b54cb31400>` 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 23:59:03.259816 133849967262848 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79b54cb324e0>` 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 23:59:03.259846 133849967262848 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 0x79b54cb31400>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79b54cb31400>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79b54cb324e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79b54cb324e0>}). I0415 23:59:03.260288 133849967262848 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 23:59:03.666837 133743863924288 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776297359680267563, 'commit_timestamp_nsecs': 1776297384950928769, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/0/_CHECKPOINT_METADATA I0415 23:59:03.743977 133743863924288 checkpoint_manager.py:1847] Missing metrics for step 0 I0415 23:59:03.807432 133849967262848 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_14_async_ckpt_false/checkpoints I0415 23:59:03.807894 133849967262848 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=False, 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_14_async_ckpt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x79b54cb32240> I0415 23:59:03.807978 133849967262848 checkpointing.py:301] Checkpoint manager created! W0415 23:59:03.824184 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:59:04.005795 133849967262848 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 23:59:04.005892 133849967262848 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 23:59:04.098010 133849967262848 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 23:59:04.098101 133849967262848 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 23:59:04.112148 133849967262848 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 23:59:04.112195 133849967262848 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 23:59:04.132481 133849967262848 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 23:59:04.132535 133849967262848 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 23:59:04.146555 133849967262848 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 23:59:04.146606 133849967262848 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 23:59:04.160640 133849967262848 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 23:59:04.160692 133849967262848 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 23:59:04.182069 133849967262848 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 23:59:04.182125 133849967262848 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 23:59:04.200145 133849967262848 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 23:59:04.200195 133849967262848 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0415 23:59:04.379397 133849967262848 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 23:59:04.379487 133849967262848 checkpointing.py:581] restoring from this run's directory step 9 I0415 23:59:04.563686 133849967262848 checkpointer.py:304] Restoring checkpoint from gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/9. I0415 23:59:05.097121 3406093 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 23:59:05.097189 3406093 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 23:59:11.079179 133849967262848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/read/gbytes_per_sec: 1.994 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 6 seconds) (per-host) I0415 23:59:11.079475 133849967262848 checkpointer.py:318] Finished restoring checkpoint in 6.59 seconds from gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/9. I0415 23:59:11.079560 133849967262848 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_14_async_ckpt_false/checkpoints', 'checkpointer_start_time': 1776297544.4885297, 'checkpointer_duration_secs': 6.590992450714111, 'checkpoint_manager_start_time': 1776297544.3797123, 'checkpoint_manager_duration_secs': 6.699812889099121} [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 23:59:11.638895 133849967262848 maxtext_utils.py:1623] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0415 23:59:11.639008 133849967262848 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) I0415 23:59:11.639066 133849967262848 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) I0415 23:59:11.639115 133849967262848 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') I0415 23:59:11.639155 133849967262848 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) I0415 23:59:11.639183 133849967262848 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) I0415 23:59:11.639222 133849967262848 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) I0415 23:59:11.639259 133849967262848 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') I0415 23:59:11.639287 133849967262848 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) I0415 23:59:11.639315 133849967262848 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) I0415 23:59:11.639350 133849967262848 maxtext_utils.py:1623] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed', 'vocab') Physical: ('fsdp', None) I0415 23:59:11.639383 133849967262848 maxtext_utils.py:1623] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed') Physical: (None, 'fsdp') I0415 23:59:12.185546 133849967262848 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 23:59:12.185627 133849967262848 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0415 23:59:12.199322 133849967262848 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 23:59:12.199373 133849967262848 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0415 23:59:12.913282 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:59:12.977369 133849967262848 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 23:59:13.940183 133849967262848 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 23:59:13.940434 133849967262848 metric_logger.py:289] number parameters: 1.104 billion W0415 23:59:15.341344 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:59:15.413549 133849967262848 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0415 23:59:15.530349 133849967262848 checkpointing.py:776] Waited 0.11679220199584961 seconds for step 10 to finish before starting checkpointing. I0415 23:59:15.530747 133849967262848 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 23:59:15.530911 133849967262848 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0415 23:59:15.530974 133849967262848 checkpointer.py:241] [process=0] Started saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/10. I0415 23:59:15.704735 133849967262848 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/10 I0415 23:59:16.548439 133849967262848 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776297556291338155, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776297556291338155, "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_14_async_ckpt_false/checkpoints/10/_CHECKPOINT_METADATA I0415 23:59:16.549370 133849967262848 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient W0415 23:59:16.549694 133849967262848 future.py:328] [process=0][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 23:59:16.634305 133742267991616 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/10/items I0415 23:59:17.171669 133849967262848 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 23:59:17.171770 133849967262848 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 W0415 23:59:17.420243 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:17.434187 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:17.438626 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:17.442651 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:17.446593 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:17.450412 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:17.454148 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:17.458006 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:59:27.585422 133849967262848 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.414101s I0415 23:59:27.589900 133849967262848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.174 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 10 seconds) (per-host) I0415 23:59:27.589963 133849967262848 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 10.510503s (batch_requests_ready=0.088850s, total_serialization_initiated=10.417284s, others=0.004368s) I0415 23:59:27.590055 133849967262848 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 11.040735s (all_items=0.000030s, per_item={'items': '0.00003004'}, temp_paths=11.040705) W0415 23:59:27.590094 133849967262848 future.py:328] [process=0][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 23:59:28.342172 133742278477376 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_14_async_ckpt_false/checkpoints/10/items/array_metadatas/process_0 I0415 23:59:54.725353 133742697907776 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 27.135043s (commit=26.663899s, array_metadata_write=0.471143s) I0415 23:59:54.726487 133849967262848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 335.665 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 37 seconds) (per-host) I0415 23:59:55.305768 133849967262848 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776297556291338155, '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_14_async_ckpt_false/checkpoints/10/_CHECKPOINT_METADATA I0415 23:59:55.442571 133849967262848 array_metadata_store.py:367] [process=0][thread=MainThread] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 23:59:55.858390 133849967262848 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 23:59:55.858552 133849967262848 base_pytree_checkpoint_handler.py:1342] [process=0][thread=MainThread] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.552382s. 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_14_async_ckpt_false/checkpoints/10/items I0415 23:59:55.859238 133849967262848 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/10/items I0415 23:59:56.122362 133849967262848 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/10 I0415 23:59:56.786684 133849967262848 atomicity.py:794] [process=0][thread=MainThread] 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_14_async_ckpt_false/checkpoints/10`. I0415 23:59:56.787334 133849967262848 checkpointer.py:290] Finished synchronous save in 41.26 seconds to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/10 I0415 23:59:56.787539 133849967262848 checkpoint_manager.py:2103] [process=0][thread=MainThread][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0415 23:59:56.787585 133849967262848 checkpoint_manager.py:2112] [process=0][thread=MainThread][step=10] CheckpointManager Save Finalize is done on all hosts. I0415 23:59:56.787610 133849967262848 checkpoint_manager.py:1570] [process=0][thread=MainThread][step=10] Finished synchronous save. I0415 23:59:56.787649 133849967262848 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_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776297555.5307348, 'wait_for_prev_duration_secs': 3.9577484130859375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776297555.5309324, 'checkpointer_blocking_duration_secs': 41.25645041465759, 'get_old_steps_start_time': 1776297596.7874174, 'get_old_steps_duration_secs': 3.504753112792969e-05, 'checkpoint_manager_blocking_start_time': 1776297555.5306501, 'checkpoint_manager_blocking_duration_secs': 41.25697445869446} I0415 23:59:56.787775 133849967262848 checkpointing.py:410] Saved a checkpoint at step 10. I0415 23:59:56.787820 133849967262848 max_utils.py:750] Memstats: After params initialized: I0415 23:59:56.787885 133849967262848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0415 23:59:56.787914 133849967262848 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0415 23:59:56.787934 133849967262848 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0415 23:59:56.787952 133849967262848 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0415 23:59:56.787971 133849967262848 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0415 23:59:56.787990 133849967262848 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0415 23:59:56.788007 133849967262848 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0415 23:59:56.788025 133849967262848 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0415 23:59:56.794724 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:56.801196 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:59:56.806154 3404437 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:59:57.003576 133849967262848 metric_logger.py:185] completed step: 10, seconds: 1.473, TFLOP/s/device: 9.227, Tokens/s/device: 1390.725, total_weights: 16384, loss: 3.602 I0415 23:59:57.117619 133849967262848 metric_logger.py:185] completed step: 11, seconds: 41.586, TFLOP/s/device: 0.327, Tokens/s/device: 49.247, total_weights: 16384, loss: 3.101 I0415 23:59:57.236489 133849967262848 metric_logger.py:185] completed step: 12, seconds: 0.008, TFLOP/s/device: 1750.693, Tokens/s/device: 263883.520, total_weights: 16384, loss: 2.684 I0415 23:59:57.354897 133849967262848 metric_logger.py:185] completed step: 13, seconds: 0.115, TFLOP/s/device: 118.230, Tokens/s/device: 17820.938, total_weights: 16384, loss: 2.348 I0415 23:59:57.359989 133849967262848 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0415 23:59:57.590677 133849967262848 checkpointing.py:776] Waited 0.23065996170043945 seconds for step 15 to finish before starting checkpointing. I0415 23:59:57.591082 133849967262848 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 23:59:57.591237 133849967262848 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 15 I0415 23:59:57.591298 133849967262848 checkpointer.py:241] [process=0] Started saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/15. I0415 23:59:57.757934 133849967262848 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/15 W0415 23:59:58.592442 133849967262848 future.py:328] [process=0][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0415 23:59:58.674633 133742697907776 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/15/items I0415 23:59:59.239381 133849967262848 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 23:59:59.239487 133849967262848 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 00:00:00.759868 133849967262848 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.520853s I0416 00:00:00.763987 133849967262848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 7.601 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 00:00:00.764062 133849967262848 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.623686s (batch_requests_ready=0.095573s, total_serialization_initiated=1.524099s, others=0.004014s) I0416 00:00:00.764310 133849967262848 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.172182s (all_items=0.000027s, per_item={'items': '0.00002694'}, temp_paths=2.172155) W0416 00:00:00.764348 133849967262848 future.py:328] [process=0][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0416 00:00:01.522690 133742267991616 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_14_async_ckpt_false/checkpoints/15/items/array_metadatas/process_0 I0416 00:00:41.751756 133742676936256 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 40.987229s (commit=40.549276s, array_metadata_write=0.437953s) I0416 00:00:41.752778 133849967262848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 296.552 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 42 seconds) (per-host) I0416 00:00:42.501898 133849967262848 array_metadata_store.py:367] [process=0][thread=MainThread] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 00:00:42.972613 133849967262848 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 00:00:42.972781 133849967262848 base_pytree_checkpoint_handler.py:1342] [process=0][thread=MainThread] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.600975s. 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_14_async_ckpt_false/checkpoints/15/items I0416 00:00:42.973483 133849967262848 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/15/items I0416 00:00:43.241167 133849967262848 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/15 I0416 00:00:43.920969 133849967262848 atomicity.py:794] [process=0][thread=MainThread] 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_14_async_ckpt_false/checkpoints/15`. I0416 00:00:43.921600 133849967262848 checkpointer.py:290] Finished synchronous save in 46.33 seconds to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/15 I0416 00:00:43.921730 133849967262848 checkpoint_manager.py:2103] [process=0][thread=MainThread][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0416 00:00:43.921766 133849967262848 checkpoint_manager.py:2112] [process=0][thread=MainThread][step=15] CheckpointManager Save Finalize is done on all hosts. I0416 00:00:43.921787 133849967262848 checkpoint_manager.py:1570] [process=0][thread=MainThread][step=15] Finished synchronous save. I0416 00:00:43.921820 133849967262848 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_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776297597.5910294, 'wait_for_prev_duration_secs': 7.963180541992188e-05, 'time_between_consecutive_saves_sec': 0.8034281730651855, 'checkpointer_blocking_start_time': 1776297597.5912576, 'checkpointer_blocking_duration_secs': 46.33038330078125, 'get_old_steps_start_time': 1776297643.9216592, 'get_old_steps_duration_secs': 2.193450927734375e-05, 'checkpoint_manager_blocking_start_time': 1776297597.5909643, 'checkpoint_manager_blocking_duration_secs': 46.33083653450012} I0416 00:00:43.921902 133849967262848 checkpointing.py:410] Saved a checkpoint at step 15. I0416 00:00:43.922478 133849967262848 metric_logger.py:185] completed step: 14, seconds: 0.119, TFLOP/s/device: 114.213, Tokens/s/device: 17215.437, total_weights: 16384, loss: 2.083 I0416 00:00:43.932307 133849967262848 metric_logger.py:185] completed step: 15, seconds: 0.118, TFLOP/s/device: 114.794, Tokens/s/device: 17302.997, total_weights: 16384, loss: 1.880 I0416 00:00:44.046827 133849967262848 metric_logger.py:185] completed step: 16, seconds: 46.568, TFLOP/s/device: 0.292, Tokens/s/device: 43.978, total_weights: 16384, loss: 1.726 I0416 00:00:44.165864 133849967262848 metric_logger.py:185] completed step: 17, seconds: 0.008, TFLOP/s/device: 1679.497, Tokens/s/device: 253152.040, total_weights: 16384, loss: 1.610 I0416 00:00:44.284231 133849967262848 metric_logger.py:185] completed step: 18, seconds: 0.115, TFLOP/s/device: 117.776, Tokens/s/device: 17752.505, total_weights: 16384, loss: 1.521 I0416 00:00:44.401941 133849967262848 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0416 00:00:44.402804 133849967262848 checkpointing.py:776] Waited 0.0008809566497802734 seconds for step 19 to finish before starting checkpointing. I0416 00:00:44.403032 133849967262848 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 00:00:44.403186 133849967262848 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0416 00:00:44.403257 133849967262848 checkpointer.py:241] [process=0] Started saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/19. I0416 00:00:44.582629 133849967262848 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/19 W0416 00:00:45.406236 133849967262848 future.py:328] [process=0][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0416 00:00:45.496725 133742697907776 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/19/items I0416 00:00:46.065355 133849967262848 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 00:00:46.065456 133849967262848 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 00:00:47.846223 133849967262848 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.781198s I0416 00:00:47.850528 133849967262848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 6.565 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 00:00:47.850590 133849967262848 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.879980s (batch_requests_ready=0.091531s, total_serialization_initiated=1.784276s, others=0.004173s) I0416 00:00:47.850692 133849967262848 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.444765s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=2.444747) W0416 00:00:47.850730 133849967262848 future.py:328] [process=0][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0416 00:00:49.598723 133742676936256 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_14_async_ckpt_false/checkpoints/19/items/array_metadatas/process_0 I0416 00:01:19.771547 133742267991616 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 31.920495s (commit=31.480982s, array_metadata_write=0.439512s) I0416 00:01:19.772622 133849967262848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 373.848 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 33 seconds) (per-host) I0416 00:01:20.511188 133849967262848 array_metadata_store.py:367] [process=0][thread=MainThread] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 00:01:20.956175 133849967262848 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 00:01:20.956708 133849967262848 base_pytree_checkpoint_handler.py:1342] [process=0][thread=MainThread] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.590138s. 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_14_async_ckpt_false/checkpoints/19/items I0416 00:01:20.957406 133849967262848 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/19/items I0416 00:01:21.206185 133849967262848 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/19 I0416 00:01:21.886306 133849967262848 atomicity.py:794] [process=0][thread=MainThread] 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_14_async_ckpt_false/checkpoints/19`. I0416 00:01:21.886949 133849967262848 checkpointer.py:290] Finished synchronous save in 37.48 seconds to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_14_async_ckpt_false/checkpoints/19 I0416 00:01:21.887124 133849967262848 checkpoint_manager.py:2103] [process=0][thread=MainThread][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0416 00:01:21.887166 133849967262848 checkpoint_manager.py:2112] [process=0][thread=MainThread][step=19] CheckpointManager Save Finalize is done on all hosts. I0416 00:01:21.887191 133849967262848 checkpoint_manager.py:1570] [process=0][thread=MainThread][step=19] Finished synchronous save. I0416 00:01:21.887223 133849967262848 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_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776297644.4030187, 'wait_for_prev_duration_secs': 5.125999450683594e-05, 'time_between_consecutive_saves_sec': 0.48123860359191895, 'checkpointer_blocking_start_time': 1776297644.4032075, 'checkpointer_blocking_duration_secs': 37.483792781829834, 'get_old_steps_start_time': 1776297681.8870182, 'get_old_steps_duration_secs': 4.100799560546875e-05, 'checkpoint_manager_blocking_start_time': 1776297644.4029865, 'checkpoint_manager_blocking_duration_secs': 37.48421835899353} I0416 00:01:21.887324 133849967262848 checkpointing.py:410] Saved a checkpoint at step 19. I0416 00:01:21.887348 133849967262848 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 00:01:21.887367 133849967262848 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 00:01:21.887994 133849967262848 metric_logger.py:185] completed step: 19, seconds: 0.120, TFLOP/s/device: 113.472, Tokens/s/device: 17103.725, 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