main (0ba93e21a) vs feat/nnx-post-train-fixes (d8cde296b)| Metric | main 0ba93e21a | feat/nnx-post-train-fixes d8cde296b | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 1.4490 | 1.4490 | 0 |
| TFLOP/s | 114.202 | 113.047 | -1.155 |
| Tok/s | 17213.7 | 17039.7 | -174.013 |
| Avg s/step | 3.255 | 3.166 | -0.089 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-16 13:39:28.713365: 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 13:39:28.829719 126178799148160 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 13:40:28.254224 126178799148160 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 13:40:28.254331 126178799148160 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 13:40:28.254364 126178799148160 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 13:40:28.254389 126178799148160 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0416 13:40:28.254411 126178799148160 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 13:40:28.254482 126178799148160 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 13:40:28.346230 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:40:28.748885 126178799148160 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 13:40:28.749168 126178799148160 checkpointing.py:677] Setting up checkpoint logger... I0416 13:40:28.749211 126178799148160 checkpointing.py:233] Creating checkpoint manager with ocdbt=False and zarr3=True I0416 13:40:28.749253 126178799148160 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 13:40:28.749708 126178799148160 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x72c1b39009e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 13:40:30.373797 126178799148160 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0416 13:40:30.373975 126178799148160 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 0x72bb37b34440>}, handler_registry=None I0416 13:40:30.374259 126178799148160 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72bb37b34440>` 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 13:40:30.374301 126178799148160 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72bb37b36030>` 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 13:40:30.374336 126178799148160 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 0x72bb37b34440>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72bb37b34440>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72bb37b36030>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72bb37b36030>}). I0416 13:40:30.374743 126178799148160 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 13:40:30.374795 126178799148160 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x72bb37b31300> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 13:40:30.873232 126072664360512 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776346706402643483, 'commit_timestamp_nsecs': 1776346723048801452, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/5/_CHECKPOINT_METADATA I0416 13:40:30.959870 126072664360512 checkpoint_manager.py:1847] Missing metrics for step 5 I0416 13:40:31.009643 126178799148160 checkpoint_manager.py:1788] Found 3 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints I0416 13:40:31.010151 126178799148160 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_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x72bb37b34260> I0416 13:40:31.010241 126178799148160 checkpointing.py:301] Checkpoint manager created! W0416 13:40:31.026786 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:40:31.210190 126178799148160 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 13:40:31.210292 126178799148160 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:40:31.307913 126178799148160 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:40:31.307996 126178799148160 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:40:31.322203 126178799148160 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:40:31.322249 126178799148160 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:40:31.342579 126178799148160 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 13:40:31.342646 126178799148160 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:40:31.356994 126178799148160 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 13:40:31.357054 126178799148160 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:40:31.371152 126178799148160 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 13:40:31.371199 126178799148160 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:40:31.392673 126178799148160 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 13:40:31.392725 126178799148160 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:40:31.410865 126178799148160 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 13:40:31.410923 126178799148160 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 13:40:31.595246 126178799148160 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0416 13:40:31.595335 126178799148160 checkpointing.py:581] restoring from this run's directory step 9 I0416 13:40:31.754769 126178799148160 checkpointer.py:304] Restoring checkpoint from gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/9. I0416 13:40:32.557819 3883839 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 13:40:32.557867 3883839 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 13:40:38.653770 126178799148160 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/read/gbytes_per_sec: 1.887 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 6 seconds) (per-host) I0416 13:40:38.654159 126178799148160 checkpointer.py:318] Finished restoring checkpoint in 6.98 seconds from gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/9. I0416 13:40:38.654265 126178799148160 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints', 'checkpointer_start_time': 1776346831.675294, 'checkpointer_duration_secs': 6.978917837142944, 'checkpoint_manager_start_time': 1776346831.5955594, 'checkpoint_manager_duration_secs': 7.058655023574829} [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 13:40:39.237738 126178799148160 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 13:40:39.237844 126178799148160 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) I0416 13:40:39.237883 126178799148160 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) I0416 13:40:39.237928 126178799148160 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') I0416 13:40:39.237969 126178799148160 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) I0416 13:40:39.237998 126178799148160 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) I0416 13:40:39.238039 126178799148160 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) I0416 13:40:39.238095 126178799148160 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') I0416 13:40:39.238125 126178799148160 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) I0416 13:40:39.238152 126178799148160 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) I0416 13:40:39.238186 126178799148160 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 13:40:39.238220 126178799148160 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 13:40:40.124099 126178799148160 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:40:40.124179 126178799148160 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 13:40:40.137556 126178799148160 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:40:40.137621 126178799148160 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 13:40:40.365340 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:40:40.446218 126178799148160 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 13:40:40.446753 126178799148160 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 13:40:40.446895 126178799148160 metric_logger.py:289] number parameters: 1.104 billion W0416 13:40:41.266666 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:40:41.336904 126178799148160 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0416 13:40:41.453328 126178799148160 checkpointing.py:776] Waited 0.11641120910644531 seconds for step 10 to finish before starting checkpointing. I0416 13:40:41.453718 126178799148160 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:40:41.453874 126178799148160 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0416 13:40:41.454156 126178799148160 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10. I0416 13:40:41.543518 126178799148160 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 13:40:41.630842 126178799148160 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:40:41.630983 126071131342400 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10 I0416 13:40:41.631082 126178799148160 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 13:40:41.853008 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:41.879594 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:41.885608 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:41.890179 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:41.894018 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:41.897797 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:41.901615 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:41.905545 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:40:42.327794 126071120856640 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10/items I0416 13:40:42.481288 126071089399360 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776346842233673267, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776346842233673267, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10/_CHECKPOINT_METADATA I0416 13:40:51.075798 126178799148160 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.445660s I0416 13:40:51.079928 126178799148160 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.294 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0416 13:40:51.079989 126178799148160 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.535566s (batch_requests_ready=0.083109s, total_serialization_initiated=9.448448s, others=0.004009s) I0416 13:40:51.080076 126178799148160 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.536108s (all_items=0.000019s, per_item={'items': '0.00001860'}, temp_paths=9.536089) I0416 13:40:51.081423 126071141828160 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:40:51.081506 126178799148160 async_checkpointer.py:561] Finished blocking save. Time taken: 9.627592s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10. I0416 13:40:51.081693 126178799148160 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:40:51.081859 126071131342400 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:40:51.081970 126178799148160 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776346841.4537055, 'wait_for_prev_duration_secs': 3.910064697265625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776346841.4538963, 'checkpointer_blocking_duration_secs': 9.627716779708862, 'get_old_steps_start_time': 1776346851.0816307, 'get_old_steps_duration_secs': 2.4318695068359375e-05, 'checkpoint_manager_blocking_start_time': 1776346841.453621, 'checkpoint_manager_blocking_duration_secs': 9.628321409225464} I0416 13:40:51.082062 126178799148160 checkpointing.py:408] Started an asynchronous checkpoint save for step 10 I0416 13:40:51.082101 126178799148160 max_utils.py:750] Memstats: After params initialized: I0416 13:40:51.082151 126178799148160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0416 13:40:51.082179 126178799148160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 13:40:51.082202 126178799148160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 13:40:51.082222 126178799148160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 13:40:51.082240 126178799148160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 13:40:51.082258 126178799148160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 13:40:51.082276 126178799148160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 13:40:51.082293 126178799148160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 13:40:51.088332 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:51.094337 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:40:51.098916 3882234 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:40:51.286280 126178799148160 metric_logger.py:185] completed step: 10, seconds: 0.890, TFLOP/s/device: 15.274, Tokens/s/device: 2302.334, total_weights: 16384, loss: 3.602 I0416 13:40:51.402185 126178799148160 metric_logger.py:185] completed step: 11, seconds: 9.948, TFLOP/s/device: 1.366, Tokens/s/device: 205.876, total_weights: 16384, loss: 3.101 I0416 13:40:51.521476 126178799148160 metric_logger.py:185] completed step: 12, seconds: 0.006, TFLOP/s/device: 2097.103, Tokens/s/device: 316098.163, total_weights: 16384, loss: 2.684 I0416 13:40:51.640176 126178799148160 metric_logger.py:185] completed step: 13, seconds: 0.117, TFLOP/s/device: 116.624, Tokens/s/device: 17578.796, total_weights: 16384, loss: 2.348 I0416 13:40:51.647034 126178799148160 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0416 13:40:51.830298 126071099885120 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10/items/array_metadatas/process_0 I0416 13:40:51.879825 126178799148160 checkpointing.py:776] Waited 0.23272442817687988 seconds for step 15 to finish before starting checkpointing. I0416 13:40:51.880472 126178799148160 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:40:56.633841 126071120856640 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 5.553198s (commit=5.189881s, array_metadata_write=0.363317s) I0416 13:40:56.634765 126071141828160 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 837.411 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 15 seconds) (per-host) I0416 13:40:56.634818 126071141828160 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 5.553270s. W0416 13:40:56.927160 126071141828160 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10/items, I0416 13:40:57.011421 126071141828160 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 13:40:57.011528 126071141828160 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.133421s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10/items I0416 13:40:57.012177 126071141828160 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10/items I0416 13:40:57.189190 126071089399360 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776346842233673267, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10/_CHECKPOINT_METADATA I0416 13:40:57.236456 126071141828160 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10 I0416 13:40:57.970346 126071141828160 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10`. I0416 13:40:57.970959 126071141828160 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 16.517051s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/10 I0416 13:40:57.971031 126071141828160 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 6.889484s. I0416 13:40:57.971228 126071131342400 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:40:57.971339 126071131342400 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:40:57.971401 126071131342400 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:40:57.971445 126071131342400 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0416 13:40:57.971765 126178799148160 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 13:40:57.972725 126178799148160 checkpoint_manager.py:1441] Waiting for previous save to complete took 6.092265 seconds. If this number is high, consider checkpointing less frequently. I0416 13:40:57.974673 126178799148160 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 15 I0416 13:40:57.975113 126178799148160 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15. I0416 13:40:58.156451 126178799148160 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:40:58.156539 126178799148160 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 13:40:58.158788 126071131342400 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15 I0416 13:40:58.843977 126071120856640 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15/items I0416 13:41:12.744232 126178799148160 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 14.588208s I0416 13:41:12.748409 126178799148160 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 860.958 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 14 seconds) (per-host) I0416 13:41:12.748474 126178799148160 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 14.677683s (batch_requests_ready=0.082595s, total_serialization_initiated=14.591012s, others=0.004076s) I0416 13:41:12.748551 126178799148160 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 14.678203s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=14.678188) I0416 13:41:12.749820 126072328816192 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:41:12.749903 126178799148160 async_checkpointer.py:561] Finished blocking save. Time taken: 14.775172s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15. I0416 13:41:12.750113 126178799148160 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:41:12.750274 126071141828160 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:41:12.750370 126178799148160 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776346851.880434, 'wait_for_prev_duration_secs': 6.092265367507935, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776346857.9747021, 'checkpointer_blocking_duration_secs': 14.775307893753052, 'get_old_steps_start_time': 1776346872.750031, 'get_old_steps_duration_secs': 4.1484832763671875e-05, 'checkpoint_manager_blocking_start_time': 1776346851.8803346, 'checkpoint_manager_blocking_duration_secs': 20.87001085281372} I0416 13:41:12.750624 126178799148160 checkpointing.py:408] Started an asynchronous checkpoint save for step 15 I0416 13:41:12.751082 126178799148160 metric_logger.py:185] completed step: 14, seconds: 0.120, TFLOP/s/device: 112.956, Tokens/s/device: 17025.946, total_weights: 16384, loss: 2.083 I0416 13:41:12.758396 126178799148160 metric_logger.py:185] completed step: 15, seconds: 0.119, TFLOP/s/device: 114.296, Tokens/s/device: 17227.891, total_weights: 16384, loss: 1.880 I0416 13:41:12.873642 126178799148160 metric_logger.py:185] completed step: 16, seconds: 21.110, TFLOP/s/device: 0.644, Tokens/s/device: 97.018, total_weights: 16384, loss: 1.726 I0416 13:41:12.991931 126178799148160 metric_logger.py:185] completed step: 17, seconds: 0.006, TFLOP/s/device: 2404.802, Tokens/s/device: 362477.876, total_weights: 16384, loss: 1.610 I0416 13:41:13.110760 126178799148160 metric_logger.py:185] completed step: 18, seconds: 0.116, TFLOP/s/device: 116.954, Tokens/s/device: 17628.578, total_weights: 16384, loss: 1.521 I0416 13:41:13.229197 126178799148160 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0416 13:41:13.230326 126178799148160 checkpointing.py:776] Waited 0.0011534690856933594 seconds for step 19 to finish before starting checkpointing. I0416 13:41:13.230947 126178799148160 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:41:13.488577 126072213472832 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15/items/array_metadatas/process_0 I0416 13:41:18.263554 126070439282240 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 5.514412s (commit=5.176864s, array_metadata_write=0.337548s) I0416 13:41:18.264606 126072328816192 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 625.776 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 20 seconds) (per-host) I0416 13:41:18.264662 126072328816192 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 5.514714s. W0416 13:41:18.577062 126072328816192 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15/items, I0416 13:41:18.673567 126072328816192 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 13:41:18.673675 126072328816192 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.155672s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15/items I0416 13:41:18.674338 126072328816192 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15/items I0416 13:41:18.898426 126072328816192 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15 I0416 13:41:19.543576 126072328816192 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15`. I0416 13:41:19.544215 126072328816192 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 21.569493s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/15 I0416 13:41:19.544296 126072328816192 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 6.794349s. I0416 13:41:19.544490 126071141828160 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:41:19.544594 126071141828160 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:41:19.544668 126071141828160 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:41:19.544713 126071141828160 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0416 13:41:19.544896 126178799148160 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 13:41:19.545019 126178799148160 checkpoint_manager.py:1441] Waiting for previous save to complete took 6.314085 seconds. If this number is high, consider checkpointing less frequently. I0416 13:41:19.545279 126178799148160 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0416 13:41:19.545583 126178799148160 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19. I0416 13:41:19.719893 126178799148160 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:41:19.720033 126072328816192 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19 I0416 13:41:19.720121 126178799148160 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 13:41:20.374901 126070439282240 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19/items I0416 13:41:32.667591 126178799148160 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 12.948391s I0416 13:41:32.671509 126178799148160 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 969.440 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 13 seconds) (per-host) I0416 13:41:32.671566 126178799148160 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 13.035229s (batch_requests_ready=0.079258s, total_serialization_initiated=12.952155s, others=0.003816s) I0416 13:41:32.671638 126178799148160 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 13.035727s (all_items=0.000013s, per_item={'items': '0.00001311'}, temp_paths=13.035714) I0416 13:41:32.672918 126072328816192 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:41:32.673067 126178799148160 async_checkpointer.py:561] Finished blocking save. Time taken: 13.127730s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19. I0416 13:41:32.673269 126178799148160 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:41:32.673478 126070386853440 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:41:32.673586 126178799148160 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776346873.2309139, 'wait_for_prev_duration_secs': 6.314085483551025, 'time_between_consecutive_saves_sec': 15.259453296661377, 'checkpointer_blocking_start_time': 1776346879.5453026, 'checkpointer_blocking_duration_secs': 13.127891302108765, 'get_old_steps_start_time': 1776346892.6732056, 'get_old_steps_duration_secs': 2.384185791015625e-05, 'checkpoint_manager_blocking_start_time': 1776346873.2308676, 'checkpoint_manager_blocking_duration_secs': 19.44269585609436} I0416 13:41:32.673713 126178799148160 checkpointing.py:408] Started an asynchronous checkpoint save for step 19 I0416 13:41:32.673745 126178799148160 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:41:33.405965 126070816769600 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19/items/array_metadatas/process_0 I0416 13:41:38.575607 126070439282240 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 5.903408s (commit=5.503093s, array_metadata_write=0.400314s) I0416 13:41:38.576552 126072328816192 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 667.196 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 18 seconds) (per-host) I0416 13:41:38.576661 126072328816192 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 5.903551s. W0416 13:41:38.900669 126072328816192 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19/items, I0416 13:41:38.989841 126072328816192 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 13:41:38.989949 126072328816192 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.143416s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19/items I0416 13:41:38.990594 126072328816192 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19/items I0416 13:41:39.225821 126072328816192 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19 I0416 13:41:39.902017 126072328816192 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19`. I0416 13:41:39.902640 126072328816192 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 20.357321s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_false/checkpoints/19 I0416 13:41:39.902711 126072328816192 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 7.229604s. I0416 13:41:39.902846 126070386853440 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:41:39.902952 126070386853440 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:41:39.902992 126070386853440 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:41:39.903027 126070386853440 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0416 13:41:39.903166 126178799148160 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 13:41:39.903306 126178799148160 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:41:39.903869 126178799148160 metric_logger.py:185] completed step: 19, seconds: 0.119, TFLOP/s/device: 114.202, Tokens/s/device: 17213.700, 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 19:20:32.251256: 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 19:20:32.367595 126330773126272 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 19:21:28.486418 126330773126272 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 19:21:28.486522 126330773126272 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 19:21:28.486555 126330773126272 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 19:21:28.486579 126330773126272 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 19:21:28.486605 126330773126272 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 19:21:28.486679 126330773126272 train.py:811] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 19:21:28.580125 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:21:28.984113 126330773126272 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 19:21:28.984277 126330773126272 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 19:21:28.984457 126330773126272 checkpointing.py:688] Setting up checkpoint logger... I0416 19:21:28.984498 126330773126272 checkpointing.py:234] Creating checkpoint manager with ocdbt=False and zarr3=True I0416 19:21:28.984537 126330773126272 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 19:21:28.984995 126330773126272 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x72e516100dd0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 19:21:30.637920 126330773126272 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 19:21:30.638113 126330773126272 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 0x72de9a339640>}, handler_registry=None I0416 19:21:30.638376 126330773126272 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72de9a339640>` 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 19:21:30.638418 126330773126272 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72de9a33a540>` 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 19:21:30.638450 126330773126272 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 0x72de9a339640>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72de9a339640>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72de9a33a540>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72de9a33a540>}). I0416 19:21:30.638990 126330773126272 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 19:21:30.639055 126330773126272 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x72de9a335ee0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 19:21:31.129949 126224644965952 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776367185516390926, 'commit_timestamp_nsecs': 1776367202194173737, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/9/_CHECKPOINT_METADATA I0416 19:21:31.217119 126224644965952 checkpoint_manager.py:1847] Missing metrics for step 9 I0416 19:21:31.251032 126330773126272 checkpoint_manager.py:1788] Found 3 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints I0416 19:21:31.251504 126330773126272 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_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x72de9a338b30> I0416 19:21:31.251595 126330773126272 checkpointing.py:302] Checkpoint manager created! W0416 19:21:31.268080 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:21:31.458990 126330773126272 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 19:21:31.459149 126330773126272 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:21:31.552500 126330773126272 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 19:21:31.552574 126330773126272 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:21:31.566686 126330773126272 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 19:21:31.566731 126330773126272 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:21:31.587920 126330773126272 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 19:21:31.587985 126330773126272 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:21:31.602273 126330773126272 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 19:21:31.602321 126330773126272 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:21:31.616441 126330773126272 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 19:21:31.616488 126330773126272 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:21:31.638005 126330773126272 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 19:21:31.638069 126330773126272 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:21:31.656098 126330773126272 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 19:21:31.656148 126330773126272 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 19:21:31.846215 126330773126272 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 19:21:31.846304 126330773126272 checkpointing.py:582] restoring from this run's directory step 9 I0416 19:21:32.004298 126330773126272 checkpointer.py:304] Restoring checkpoint from gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/9. I0416 19:21:32.844464 4032744 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 19:21:32.844527 4032744 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 19:21:39.397827 126330773126272 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/read/gbytes_per_sec: 1.762 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 7 seconds) (per-host) I0416 19:21:39.398357 126330773126272 checkpointer.py:318] Finished restoring checkpoint in 7.47 seconds from gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/9. I0416 19:21:39.398460 126330773126272 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints', 'checkpointer_start_time': 1776367291.9258633, 'checkpointer_duration_secs': 7.472548484802246, 'checkpoint_manager_start_time': 1776367291.8465333, 'checkpoint_manager_duration_secs': 7.551882266998291} [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 19:21:39.936879 126330773126272 maxtext_utils.py:1796] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 19:21:39.936996 126330773126272 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 19:21:39.937037 126330773126272 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 19:21:39.937101 126330773126272 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 19:21:39.937142 126330773126272 maxtext_utils.py:1796] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 19:21:39.937168 126330773126272 maxtext_utils.py:1796] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 19:21:39.937205 126330773126272 maxtext_utils.py:1796] 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 19:21:39.937247 126330773126272 maxtext_utils.py:1796] 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 19:21:39.937276 126330773126272 maxtext_utils.py:1796] 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 19:21:39.937302 126330773126272 maxtext_utils.py:1796] 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 19:21:39.937337 126330773126272 maxtext_utils.py:1796] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 19:21:39.937374 126330773126272 maxtext_utils.py:1796] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 19:21:40.465053 126330773126272 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 19:21:40.465134 126330773126272 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 19:21:40.478920 126330773126272 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 19:21:40.478970 126330773126272 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 19:21:41.142742 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:21:41.219629 126330773126272 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 19:21:41.220124 126330773126272 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 19:21:41.220267 126330773126272 metric_logger.py:289] number parameters: 1.104 billion W0416 19:21:42.076553 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:21:42.149069 126330773126272 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0416 19:21:42.266301 126330773126272 checkpointing.py:798] Waited 0.11723542213439941 seconds for step 10 to finish before starting checkpointing. I0416 19:21:42.266678 126330773126272 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 19:21:42.266840 126330773126272 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0416 19:21:42.267152 126330773126272 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10. I0416 19:21:42.350764 126330773126272 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 19:21:42.437082 126330773126272 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:21:42.437435 126330773126272 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 19:21:42.443077 126222988215872 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10 W0416 19:21:42.665269 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:42.684552 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:42.697375 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:42.702107 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:42.706344 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:42.710514 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:42.714479 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:42.718482 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:21:43.178018 126222977730112 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10/items I0416 19:21:43.362103 126222946272832 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776367303086160651, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776367303086160651, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10/_CHECKPOINT_METADATA I0416 19:21:51.973602 126330773126272 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.536928s I0416 19:21:51.977649 126330773126272 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.282 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0416 19:21:51.977713 126330773126272 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.625769s (batch_requests_ready=0.081708s, total_serialization_initiated=9.540121s, others=0.003940s) I0416 19:21:51.977783 126330773126272 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.626559s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=9.626541) I0416 19:21:51.979206 126222935787072 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:21:51.979287 126330773126272 async_checkpointer.py:561] Finished blocking save. Time taken: 9.712407s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10. I0416 19:21:51.979467 126330773126272 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:21:51.979636 126222998701632 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:21:51.979751 126330773126272 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776367302.266666, 'wait_for_prev_duration_secs': 3.8623809814453125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776367302.2668607, 'checkpointer_blocking_duration_secs': 9.712533235549927, 'get_old_steps_start_time': 1776367311.979409, 'get_old_steps_duration_secs': 2.3126602172851562e-05, 'checkpoint_manager_blocking_start_time': 1776367302.2665854, 'checkpoint_manager_blocking_duration_secs': 9.713140726089478} I0416 19:21:51.979834 126330773126272 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0416 19:21:51.979871 126330773126272 max_utils.py:750] Memstats: After params initialized: I0416 19:21:51.979923 126330773126272 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0416 19:21:51.979948 126330773126272 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 19:21:51.979968 126330773126272 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 19:21:51.979986 126330773126272 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 19:21:51.980004 126330773126272 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 19:21:51.980020 126330773126272 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 19:21:51.980037 126330773126272 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 19:21:51.980070 126330773126272 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 19:21:51.986065 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:51.992376 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:21:51.998284 4031160 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:21:52.192362 126330773126272 metric_logger.py:185] completed step: 10, seconds: 0.928, TFLOP/s/device: 14.636, Tokens/s/device: 2206.124, total_weights: 16384, loss: 3.602 I0416 19:21:52.308902 126330773126272 metric_logger.py:185] completed step: 11, seconds: 10.042, TFLOP/s/device: 1.353, Tokens/s/device: 203.953, total_weights: 16384, loss: 3.101 I0416 19:21:52.427675 126330773126272 metric_logger.py:185] completed step: 12, seconds: 0.006, TFLOP/s/device: 2452.550, Tokens/s/device: 369675.090, total_weights: 16384, loss: 2.684 I0416 19:21:52.546391 126330773126272 metric_logger.py:185] completed step: 13, seconds: 0.119, TFLOP/s/device: 114.376, Tokens/s/device: 17239.928, total_weights: 16384, loss: 2.348 I0416 19:21:52.554802 126330773126272 checkpointing.py:794] Waiting for step 15 to finish before checkpoint... I0416 19:21:52.736155 126222956758592 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10/items/array_metadatas/process_0 I0416 19:21:52.784336 126330773126272 checkpointing.py:798] Waited 0.22947287559509277 seconds for step 15 to finish before starting checkpointing. I0416 19:21:52.784963 126330773126272 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:21:59.284029 126222977730112 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 7.305540s (commit=6.884423s, array_metadata_write=0.421117s) I0416 19:21:59.284988 126222935787072 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 746.282 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 16 seconds) (per-host) I0416 19:21:59.285105 126222935787072 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 7.305774s. W0416 19:21:59.584393 126222935787072 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10/items, I0416 19:21:59.682189 126222935787072 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 19:21:59.682300 126222935787072 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.147094s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10/items I0416 19:21:59.682908 126222935787072 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10/items I0416 19:21:59.891749 126222946272832 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776367303086160651, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10/_CHECKPOINT_METADATA I0416 19:21:59.918206 126222935787072 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10 I0416 19:22:00.605099 126222935787072 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10`. I0416 19:22:00.605726 126222935787072 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 18.338854s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/10 I0416 19:22:00.605801 126222935787072 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 8.626472s. I0416 19:22:00.605999 126222998701632 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:22:00.606116 126222998701632 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:22:00.606177 126222998701632 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:22:00.606222 126222998701632 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0416 19:22:00.606491 126330773126272 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 19:22:00.606976 126330773126272 checkpoint_manager.py:1441] Waiting for previous save to complete took 7.822032 seconds. If this number is high, consider checkpointing less frequently. I0416 19:22:00.607864 126330773126272 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 15 I0416 19:22:00.608157 126330773126272 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15. I0416 19:22:00.787801 126330773126272 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:22:00.787936 126222998701632 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15 I0416 19:22:00.788011 126330773126272 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 19:22:01.468922 126222988215872 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15/items I0416 19:22:12.624210 126330773126272 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 11.837097s I0416 19:22:12.624691 126330773126272 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.035 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 11 seconds) (per-host) I0416 19:22:12.624893 126330773126272 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 11.926336s (batch_requests_ready=0.085179s, total_serialization_initiated=11.840767s, others=0.000390s) I0416 19:22:12.629534 126330773126272 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 11.931445s (all_items=0.000013s, per_item={'items': '0.00001287'}, temp_paths=11.931432) I0416 19:22:12.630704 126222998701632 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:22:12.630801 126330773126272 async_checkpointer.py:561] Finished blocking save. Time taken: 12.022890s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15. I0416 19:22:12.630989 126330773126272 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:22:12.631167 126222935787072 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:22:12.631269 126330773126272 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776367312.784928, 'wait_for_prev_duration_secs': 7.82203221321106, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776367320.60789, 'checkpointer_blocking_duration_secs': 12.023015022277832, 'get_old_steps_start_time': 1776367332.6309226, 'get_old_steps_duration_secs': 2.2649765014648438e-05, 'checkpoint_manager_blocking_start_time': 1776367312.784831, 'checkpoint_manager_blocking_duration_secs': 19.846409797668457} I0416 19:22:12.631388 126330773126272 checkpointing.py:409] Started an asynchronous checkpoint save for step 15 I0416 19:22:12.631821 126330773126272 metric_logger.py:185] completed step: 14, seconds: 0.118, TFLOP/s/device: 114.819, Tokens/s/device: 17306.798, total_weights: 16384, loss: 2.083 I0416 19:22:12.639676 126330773126272 metric_logger.py:185] completed step: 15, seconds: 0.121, TFLOP/s/device: 111.990, Tokens/s/device: 16880.419, total_weights: 16384, loss: 1.880 I0416 19:22:12.755067 126330773126272 metric_logger.py:185] completed step: 16, seconds: 20.083, TFLOP/s/device: 0.677, Tokens/s/device: 101.978, total_weights: 16384, loss: 1.726 I0416 19:22:12.873601 126330773126272 metric_logger.py:185] completed step: 17, seconds: 0.006, TFLOP/s/device: 2247.664, Tokens/s/device: 338792.390, total_weights: 16384, loss: 1.610 I0416 19:22:12.994183 126330773126272 metric_logger.py:185] completed step: 18, seconds: 0.116, TFLOP/s/device: 117.047, Tokens/s/device: 17642.549, total_weights: 16384, loss: 1.521 I0416 19:22:13.111627 126330773126272 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0416 19:22:13.113168 126330773126272 checkpointing.py:798] Waited 0.0015759468078613281 seconds for step 19 to finish before starting checkpointing. I0416 19:22:13.113556 126330773126272 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:22:13.470948 126222652671552 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15/items/array_metadatas/process_0 I0416 19:22:19.830324 126224309421632 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 7.205077s (commit=6.804623s, array_metadata_write=0.400455s) I0416 19:22:19.831238 126222998701632 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 660.481 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 19 seconds) (per-host) I0416 19:22:19.831345 126222998701632 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 7.200500s. W0416 19:22:20.138599 126222998701632 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15/items, I0416 19:22:20.223816 126222998701632 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 19:22:20.223928 126222998701632 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.135656s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15/items I0416 19:22:20.224564 126222998701632 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15/items I0416 19:22:20.471522 126222998701632 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15 I0416 19:22:21.154687 126222998701632 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15`. I0416 19:22:21.155319 126222998701632 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 20.547418s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/15 I0416 19:22:21.155387 126222998701632 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 8.524544s. I0416 19:22:21.155555 126222935787072 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:22:21.155662 126222935787072 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:22:21.155714 126222935787072 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:22:21.155761 126222935787072 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0416 19:22:21.155954 126330773126272 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 19:22:21.156135 126330773126272 checkpoint_manager.py:1441] Waiting for previous save to complete took 8.042593 seconds. If this number is high, consider checkpointing less frequently. I0416 19:22:21.156418 126330773126272 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0416 19:22:21.156952 126330773126272 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19. I0416 19:22:21.312797 126330773126272 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:22:21.312882 126330773126272 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 19:22:21.324079 126222935787072 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19 I0416 19:22:22.004440 126224309421632 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19/items I0416 19:22:34.293900 126330773126272 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 12.981481s I0416 19:22:34.294452 126330773126272 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 967.730 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 13 seconds) (per-host) I0416 19:22:34.294513 126330773126272 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 13.058251s (batch_requests_ready=0.073266s, total_serialization_initiated=12.984567s, others=0.000418s) I0416 19:22:34.294581 126330773126272 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 13.058728s (all_items=0.000012s, per_item={'items': '0.00001192'}, temp_paths=13.058716) I0416 19:22:34.299463 126222726071872 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:22:34.299547 126330773126272 async_checkpointer.py:561] Finished blocking save. Time taken: 13.143085s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19. I0416 19:22:34.299728 126330773126272 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:22:34.299995 126224267478592 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:22:34.300226 126330773126272 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776367333.113521, 'wait_for_prev_duration_secs': 8.042593240737915, 'time_between_consecutive_saves_sec': 12.507284164428711, 'checkpointer_blocking_start_time': 1776367341.1564422, 'checkpointer_blocking_duration_secs': 13.143224000930786, 'get_old_steps_start_time': 1776367354.2996767, 'get_old_steps_duration_secs': 1.9311904907226562e-05, 'checkpoint_manager_blocking_start_time': 1776367333.113469, 'checkpoint_manager_blocking_duration_secs': 21.18673038482666} I0416 19:22:34.300343 126330773126272 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0416 19:22:34.300376 126330773126272 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:22:35.074142 126222694614592 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19/items/array_metadatas/process_0 I0416 19:22:41.828716 126223403451968 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 7.529916s (commit=7.118421s, array_metadata_write=0.411495s) I0416 19:22:41.829699 126222726071872 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 613.633 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 20 seconds) (per-host) I0416 19:22:41.829759 126222726071872 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 7.530164s. W0416 19:22:42.132089 126222726071872 ocdbt_utils.py:175] [process=0][thread=async_save] Skipping merge of OCDBT checkpoints: No per-process OCDBT checkpoint subdirs found in gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19/items, I0416 19:22:42.223961 126222726071872 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 19:22:42.224079 126222726071872 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.139418s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19/items I0416 19:22:42.224712 126222726071872 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19/items I0416 19:22:42.455476 126222726071872 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19 I0416 19:22:43.104531 126222726071872 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19`. I0416 19:22:43.105172 126222726071872 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 21.948718s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_false/checkpoints/19 I0416 19:22:43.105236 126222726071872 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 8.805641s. I0416 19:22:43.105397 126224267478592 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:22:43.105504 126224267478592 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:22:43.105561 126224267478592 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:22:43.105600 126224267478592 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0416 19:22:43.105752 126330773126272 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 19:22:43.105877 126330773126272 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 19:22:43.106421 126330773126272 metric_logger.py:185] completed step: 19, seconds: 0.120, TFLOP/s/device: 113.047, Tokens/s/device: 17039.687, 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