2026-04-21 17:46:45.935385: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered WARNING: All log messages before absl::InitializeLog() is called are written to STDERR E0000 00:00:1776793605.945356 1669154 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776793605.948472 1669154 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776793605.958266 1669154 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776793605.958276 1669154 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776793605.958277 1669154 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776793605.958279 1669154 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} `rope_scaling`'s factor field must be a float >= 1, got 40 `rope_scaling`'s beta_fast field must be a float, got 32 `rope_scaling`'s beta_slow field must be a float, got 1 Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} 2026-04-21 17:46:49.611403: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0421 17:46:49.747709 140412100615296 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0421 17:47:17.752761 140412100615296 max_utils.py:800] System Information: Jax Version: 0.8.1 I0421 17:47:17.753050 140412100615296 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0421 17:47:17.753097 140412100615296 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709 I0421 17:47:17.753126 140412100615296 train_utils.py:378] WARNING: 'dataset_path' might be pointing your local file system I0421 17:47:17.753317 140412100615296 train.py:806] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0421 17:47:18.385749 140412100615296 maxtext_utils.py:1732] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 17:47:18.484998 140412100615296 checkpointing.py:688] Setting up checkpoint logger... I0421 17:47:18.485123 140412100615296 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 17:47:18.485162 140412100615296 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0421 17:47:18.485333 140412100615296 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7fb3f11ab200>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 17:47:20.869881 140412100615296 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0421 17:47:20.870087 140412100615296 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 0x7fb2491918b0>}, handler_registry=None I0421 17:47:20.870312 140412100615296 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fb2491918b0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0421 17:47:20.870352 140412100615296 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fb2491a24e0>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0421 17:47:20.870381 140412100615296 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 0x7fb2491918b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fb2491918b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fb2491a24e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fb2491a24e0>}). I0421 17:47:20.870651 140412100615296 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0421 17:47:20.870703 140412100615296 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7fb24a5bcc20> timeout: 600 secs and primary_host=0 for async checkpoint writes I0421 17:47:21.000671 140412100615296 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints I0421 17:47:21.000906 140412100615296 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=10), 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/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7fb2491a32c0> I0421 17:47:21.000991 140412100615296 checkpointing.py:302] Checkpoint manager created! I0421 17:47:21.077775 140412100615296 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0421 17:47:21.080675 140412100615296 reader.py:262] Creating a tf.data.Dataset reading 8 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0. I0421 17:47:21.132668 140412100615296 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split train, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0421 17:47:21.158322 140412100615296 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0421 17:47:21.158411 140412100615296 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0421 17:47:21.797934 140412100615296 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0421 17:47:21.798071 140412100615296 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0421 17:47:22.589576 140412100615296 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[8,128,2048]........................................ ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 17:47:22.589807 140412100615296 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[8,128,2048]........................................ ('fsdp', None, None). I0421 17:47:22.594995 140412100615296 nnx_decoders.py:465] Unknown Logical: bfloat16[8,128,2048]........................................ ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 17:47:22.595163 140412100615296 nnx_decoders.py:465] Unknown Physical: bfloat16[8,128,2048]........................................ ('fsdp', None, None). I0421 17:47:22.608375 140412100615296 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,128,2048]........................................ ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 17:47:22.608458 140412100615296 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,128,2048]........................................ ('fsdp', None, None). I0421 17:47:22.620882 140412100615296 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,128,2048]........................................ ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 17:47:22.620962 140412100615296 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,128,2048]........................................ ('fsdp', None, None). I0421 17:47:22.641597 140412100615296 attentions.py:1154] attentions/query Logical: bfloat16[8,128,16,128]...................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 17:47:22.641689 140412100615296 attentions.py:1154] attentions/query Physical: bfloat16[8,128,16,128]...................................... ('fsdp', None, None, None). I0421 17:47:22.654541 140412100615296 attentions.py:1155] attentions/key Logical: bfloat16[8,128,16,128]...................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 17:47:22.654626 140412100615296 attentions.py:1155] attentions/key Physical: bfloat16[8,128,16,128]...................................... ('fsdp', None, None, None). I0421 17:47:22.667176 140412100615296 attentions.py:1156] attentions/value Logical: bfloat16[8,128,16,128]...................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 17:47:22.667258 140412100615296 attentions.py:1156] attentions/value Physical: bfloat16[8,128,16,128]...................................... ('fsdp', None, None, None). I0421 17:47:22.690208 140412100615296 attentions.py:1198] attentions/out Logical: bfloat16[8,128,16,128]...................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 17:47:22.690299 140412100615296 attentions.py:1198] attentions/out Physical: bfloat16[8,128,16,128]...................................... ('fsdp', None, None, None). I0421 17:47:22.713135 140412100615296 linears.py:525] linears/x Logical: bfloat16[8,128,7168]........................................ ('activation_batch', 'activation_length', 'activation_mlp'). I0421 17:47:22.713232 140412100615296 linears.py:525] linears/x Physical: bfloat16[8,128,7168]........................................ ('fsdp', None, None). I0421 17:47:27.211492 140412100615296 max_utils.py:791] Total memory size: 2.5 GB, Output size: 1.5 GB, Temp size: 1.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0421 17:47:27.215080 140412100615296 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0421 17:47:27.216450 140412100615296 metric_logger.py:301] number parameters: 1.104 billion I0421 17:47:35.218696 140412100615296 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0421 17:47:35.291915 140412100615296 checkpointing.py:798] Waited 0.0732123851776123 seconds for step 0 to finish before starting checkpointing. I0421 17:47:35.292418 140412100615296 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 17:47:35.292573 140412100615296 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0421 17:47:35.292973 140412100615296 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0. I0421 17:47:35.414923 140412100615296 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0421 17:47:35.503498 140320566150720 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0 I0421 17:47:35.517160 140412100615296 jax_array_handlers.py:347] Scheduling D2H of 69 prioritized jax.Array. I0421 17:47:35.517252 140412100615296 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0421 17:47:36.170694 140320555664960 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0/items I0421 17:47:36.336026 140320520013376 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776793656091130549, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776793656091130549, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0/_CHECKPOINT_METADATA I0421 17:47:36.663871 140412100615296 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.148033s I0421 17:47:36.668316 1672358 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0421 17:47:36.668380 1672358 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0421 17:47:36.672008 140412100615296 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 9.830 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0421 17:47:36.672234 140412100615296 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.255649s (batch_requests_ready=0.091597s, total_serialization_initiated=1.162807s, others=0.001245s) I0421 17:47:36.672307 140412100615296 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.256862s (all_items=0.000029s, per_item={'items': '0.00002909'}, temp_paths=1.256833) I0421 17:47:36.673614 140320345949760 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 17:47:36.673729 140412100615296 async_checkpointer.py:561] Finished blocking save. Time taken: 1.381117s. Continuing background save to gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0. I0421 17:47:36.673964 140412100615296 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 17:47:36.674231 140320052348480 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 17:47:36.674310 140412100615296 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776793655.2924056, 'wait_for_prev_duration_secs': 3.981590270996094e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776793655.292595, 'checkpointer_blocking_duration_secs': 1.3812997341156006, 'get_old_steps_start_time': 1776793656.673911, 'get_old_steps_duration_secs': 1.9311904907226562e-05, 'checkpoint_manager_blocking_start_time': 1776793655.2923174, 'checkpoint_manager_blocking_duration_secs': 1.3819684982299805} I0421 17:47:36.674411 140412100615296 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0421 17:47:36.674481 140412100615296 max_utils.py:750] Memstats: After params initialized: I0421 17:47:36.674530 140412100615296 max_utils.py:756] Using (GB) 1.56 / 31.25 (4.992000%) on TPU_0(process=0,(0,0,0,0)) I0421 17:47:36.674555 140412100615296 max_utils.py:756] Using (GB) 1.56 / 31.25 (4.992000%) on TPU_1(process=0,(1,0,0,0)) I0421 17:47:36.674575 140412100615296 max_utils.py:756] Using (GB) 1.56 / 31.25 (4.992000%) on TPU_2(process=0,(0,1,0,0)) I0421 17:47:36.674595 140412100615296 max_utils.py:756] Using (GB) 1.56 / 31.25 (4.992000%) on TPU_3(process=0,(1,1,0,0)) I0421 17:47:36.674613 140412100615296 max_utils.py:756] Using (GB) 1.56 / 31.25 (4.992000%) on TPU_4(process=0,(0,2,0,0)) I0421 17:47:36.674631 140412100615296 max_utils.py:756] Using (GB) 1.56 / 31.25 (4.992000%) on TPU_5(process=0,(1,2,0,0)) I0421 17:47:36.674647 140412100615296 max_utils.py:756] Using (GB) 1.56 / 31.25 (4.992000%) on TPU_6(process=0,(0,3,0,0)) I0421 17:47:36.674664 140412100615296 max_utils.py:756] Using (GB) 1.56 / 31.25 (4.992000%) on TPU_7(process=0,(1,3,0,0)) I0421 17:47:37.115229 140412100615296 metric_logger.py:196] completed step: 0, seconds: 8.001, TFLOP/s/device: 0.100, Tokens/s/device: 15.999, total_weights: 963, loss: 10.836, lm_loss: 10.836, perplexity: 50825.188 I0421 17:47:37.116259 140412100615296 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/tensorboard/' I0421 17:47:37.173740 140412100615296 metric_logger.py:196] completed step: 1, seconds: 1.893, TFLOP/s/device: 0.423, Tokens/s/device: 67.631, total_weights: 945, loss: 10.847, lm_loss: 10.847, perplexity: 51361.996 I0421 17:47:37.237014 140412100615296 metric_logger.py:196] completed step: 2, seconds: 0.023, TFLOP/s/device: 34.912, Tokens/s/device: 5579.773, total_weights: 962, loss: 10.296, lm_loss: 10.296, perplexity: 29611.197 I0421 17:47:37.300415 140412100615296 metric_logger.py:196] completed step: 3, seconds: 0.050, TFLOP/s/device: 16.172, Tokens/s/device: 2584.762, total_weights: 975, loss: 9.777, lm_loss: 9.777, perplexity: 17618.064 I0421 17:47:46.490178 140412100615296 metric_logger.py:196] completed step: 4, seconds: 0.064, TFLOP/s/device: 12.592, Tokens/s/device: 2012.515, total_weights: 962, loss: 9.422, lm_loss: 9.422, perplexity: 12362.127 I0421 17:47:46.490934 140320398378560 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0/items/array_metadatas/process_0 I0421 17:47:46.523896 140412100615296 metric_logger.py:196] completed step: 5, seconds: 0.063, TFLOP/s/device: 12.693, Tokens/s/device: 2028.623, total_weights: 958, loss: 9.222, lm_loss: 9.222, perplexity: 10120.491 I0421 17:47:46.584592 140412100615296 metric_logger.py:196] completed step: 6, seconds: 9.212, TFLOP/s/device: 0.087, Tokens/s/device: 13.895, total_weights: 930, loss: 9.362, lm_loss: 9.362, perplexity: 11635.593 I0421 17:47:46.648303 140412100615296 metric_logger.py:196] completed step: 7, seconds: 0.010, TFLOP/s/device: 78.687, Tokens/s/device: 12576.145, total_weights: 935, loss: 9.333, lm_loss: 9.333, perplexity: 11305.932 I0421 17:47:46.711741 140412100615296 metric_logger.py:196] completed step: 8, seconds: 0.060, TFLOP/s/device: 13.366, Tokens/s/device: 2136.182, total_weights: 930, loss: 9.560, lm_loss: 9.560, perplexity: 14189.047 I0421 17:47:46.776078 140412100615296 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0421 17:47:46.777381 140412100615296 checkpointing.py:798] Waited 0.0013260841369628906 seconds for step 9 to finish before starting checkpointing. I0421 17:47:46.777743 140412100615296 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 17:48:13.923907 140320377407040 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.252367s (commit=36.827783s, array_metadata_write=0.424583s) I0421 17:48:13.925066 140320345949760 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 328.156 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 38 seconds) (per-host) I0421 17:48:13.925120 140320345949760 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.251350s. I0421 17:48:14.175624 140320345949760 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776793656091130549, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0/_CHECKPOINT_METADATA I0421 17:48:14.359888 140320345949760 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0421 17:48:14.551159 140320520013376 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776793656091130549, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0/_CHECKPOINT_METADATA I0421 17:48:14.762465 140320345949760 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 17:48:14.763170 140320345949760 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.546997s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0/items I0421 17:48:14.763946 140320345949760 atomicity.py:608] Finalizing gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0/items I0421 17:48:15.006178 140320345949760 atomicity.py:608] Finalizing gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0 I0421 17:48:15.671089 140320345949760 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0`. I0421 17:48:15.671844 140320345949760 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 40.379238s. directory=gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/0 I0421 17:48:15.671922 140320345949760 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 38.998152s. I0421 17:48:15.672073 140320052348480 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 17:48:15.672124 140320052348480 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 17:48:15.672179 140320052348480 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 17:48:15.672225 140320052348480 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 17:48:15.672394 140412100615296 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0421 17:48:15.672538 140412100615296 checkpoint_manager.py:1441] Waiting for previous save to complete took 28.894801 seconds. If this number is high, consider checkpointing less frequently. I0421 17:48:15.673300 140412100615296 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0421 17:48:15.673612 140412100615296 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9. I0421 17:48:15.856178 140412100615296 jax_array_handlers.py:347] Scheduling D2H of 69 prioritized jax.Array. I0421 17:48:15.856343 140320052348480 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9 I0421 17:48:15.856435 140412100615296 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0421 17:48:16.574027 140320377407040 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9/items I0421 17:48:31.670632 140412100615296 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 15.816223s I0421 17:48:31.678571 140412100615296 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 793.763 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 15 seconds) (per-host) I0421 17:48:31.678646 140412100615296 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 15.920225s (batch_requests_ready=0.088363s, total_serialization_initiated=15.824103s, others=0.007759s) I0421 17:48:31.678732 140412100615296 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 15.920836s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=15.920819) I0421 17:48:31.680482 140320356435520 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 17:48:31.680588 140412100615296 async_checkpointer.py:561] Finished blocking save. Time taken: 16.007241s. Continuing background save to gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9. I0421 17:48:31.680807 140412100615296 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 17:48:31.680993 140320408864320 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 17:48:31.681107 140412100615296 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776793666.7777092, 'wait_for_prev_duration_secs': 28.894801378250122, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776793695.6733265, 'checkpointer_blocking_duration_secs': 16.007373571395874, 'get_old_steps_start_time': 1776793711.6807277, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776793666.777659, 'checkpoint_manager_blocking_duration_secs': 44.903420209884644} I0421 17:48:31.681253 140412100615296 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0421 17:48:31.681289 140412100615296 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 17:48:32.422882 140320429835840 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9/items/array_metadatas/process_0 I0421 17:49:06.276415 140320377407040 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 34.596930s (commit=34.133524s, array_metadata_write=0.463405s) I0421 17:49:06.277746 140320356435520 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 250.138 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 50 seconds) (per-host) I0421 17:49:06.277875 140320356435520 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 34.597238s. I0421 17:49:06.715166 140320356435520 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0421 17:49:07.120732 140320356435520 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 17:49:07.121515 140320356435520 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.566426s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9/items I0421 17:49:07.122267 140320356435520 atomicity.py:608] Finalizing gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9/items I0421 17:49:07.383318 140320356435520 atomicity.py:608] Finalizing gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9 I0421 17:49:08.095541 140320356435520 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9`. I0421 17:49:08.096256 140320356435520 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 52.422915s. directory=gs://wanglance-maxtext/nnx_ckpt_feat_nnx_trainstate_and_training_loop_20260421_161934/nnx_feat_nnx_trainstate_and_training_loop_20260421_161934_03_dropout/checkpoints/9 I0421 17:49:08.096333 140320356435520 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 36.415701s. I0421 17:49:08.096462 140320408864320 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 17:49:08.096515 140320408864320 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 17:49:08.096585 140320408864320 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 17:49:08.096626 140320408864320 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 17:49:08.097299 140412100615296 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0421 17:49:08.097520 140412100615296 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 17:49:08.098952 140412100615296 metric_logger.py:196] completed step: 9, seconds: 0.065, TFLOP/s/device: 12.362, Tokens/s/device: 1975.705, total_weights: 948, loss: 9.098, lm_loss: 9.098, perplexity: 8940.747 Per train step: Total TFLOPs: 0.80 split as 99.60% learnable weight flops and 0.40% attention flops