main (b341212ce) vs fix/gate-fp8-nan-sanitization (cd5a4f898)| Metric | main b341212ce | fix/gate-fp8-nan-sanitization cd5a4f898 | Diff (fix/gate-fp8-nan-sanitization − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.7100 | 8.7100 | 0 |
| TFLOP/s | 112.656 | 121.669 | +9.013 |
| Tok/s | 16980.8 | 18339.3 | +1358.486 |
| Avg s/step | 1.821 | 2.174 | +0.353 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.8.1 | 0.8.1 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-21 15:18:23.723105: 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 00:00:1776784703.733053 1508339 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776784703.736147 1508339 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776784703.745907 1508339 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776784703.745918 1508339 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776784703.745919 1508339 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776784703.745921 1508339 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. 2026-04-21 15:18:27.144948: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0421 15:18:27.260849 134706785995904 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0421 15:19:05.773081 134706785995904 max_utils.py:800] System Information: Jax Version: 0.8.1 I0421 15:19:05.773363 134706785995904 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0421 15:19:05.773411 134706785995904 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709 I0421 15:19:05.773443 134706785995904 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system I0421 15:19:05.773608 134706785995904 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0421 15:19:06.128012 134706785995904 maxtext_utils.py:1551] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 15:19:06.128396 134706785995904 checkpointing.py:677] Setting up checkpoint logger... I0421 15:19:06.128444 134706785995904 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 15:19:06.128539 134706785995904 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0421 15:19:06.129378 134706785995904 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 0x7a8391da5af0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 15:19:08.504277 134706785995904 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0421 15:19:08.504548 134706785995904 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 0x7a81ec589f10>}, handler_registry=None I0421 15:19:08.504933 134706785995904 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a81ec589f10>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0421 15:19:08.504984 134706785995904 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a81ec58a5a0>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0421 15:19:08.505016 134706785995904 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 0x7a81ec589f10>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a81ec589f10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a81ec58a5a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a81ec58a5a0>}). I0421 15:19:08.505319 134706785995904 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0421 15:19:08.505380 134706785995904 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7a81ec5a5260> timeout: 600 secs and primary_host=0 for async checkpoint writes I0421 15:19:08.628855 134706785995904 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints I0421 15:19:08.629140 134706785995904 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/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7a81ec588620> I0421 15:19:08.629235 134706785995904 checkpointing.py:301] Checkpoint manager created! I0421 15:19:08.990144 134706785995904 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0421 15:19:08.993080 134706785995904 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 15:19:09.047316 134706785995904 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 15:19:09.069895 134706785995904 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0421 15:19:09.069958 134706785995904 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0421 15:19:09.668799 134706785995904 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 15:19:09.668900 134706785995904 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:19:09.988501 134706785995904 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 15:19:09.988654 134706785995904 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:19:10.001280 134706785995904 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 15:19:10.001330 134706785995904 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:19:10.021265 134706785995904 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:19:10.021328 134706785995904 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:19:10.033884 134706785995904 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:19:10.033936 134706785995904 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:19:10.046317 134706785995904 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:19:10.046367 134706785995904 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:19:10.066429 134706785995904 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 15:19:10.066492 134706785995904 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:19:10.082374 134706785995904 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 15:19:10.082428 134706785995904 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0421 15:19:10.237258 134706785995904 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0421 15:19:10.237375 134706785995904 checkpointing.py:665] 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 15:19:10.847341 134706785995904 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0421 15:19:10.847457 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0421 15:19:10.847498 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0421 15:19:10.847545 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0421 15:19:10.847584 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0421 15:19:10.847613 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0421 15:19:10.847652 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0421 15:19:10.847690 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: PartitionSpec('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0421 15:19:10.847726 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0421 15:19:10.847754 134706785995904 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0421 15:19:10.847792 134706785995904 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 15:19:10.847827 134706785995904 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 15:19:11.257157 134706785995904 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 15:19:11.257248 134706785995904 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0421 15:19:11.269537 134706785995904 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 15:19:11.269588 134706785995904 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0421 15:19:12.303934 134706785995904 max_utils.py:791] Total memory size: 3.6 GB, Output size: 1.5 GB, Temp size: 2.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0421 15:19:12.304512 134706785995904 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0421 15:19:12.304664 134706785995904 metric_logger.py:301] number parameters: 1.104 billion I0421 15:19:15.998209 134706785995904 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0421 15:19:16.119521 134706785995904 checkpointing.py:776] Waited 0.12129330635070801 seconds for step 0 to finish before starting checkpointing. I0421 15:19:16.119987 134706785995904 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 15:19:16.120154 134706785995904 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0421 15:19:16.120556 134706785995904 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0. I0421 15:19:16.201225 134706785995904 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0421 15:19:16.286984 134706785995904 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:19:16.287093 134706785995904 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0421 15:19:16.294829 134616161715776 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0 I0421 15:19:16.943724 134706785995904 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.657251s I0421 15:19:16.944269 134706785995904 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 16.641 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 741 milliseconds) (per-host) I0421 15:19:16.944332 134706785995904 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.741700s (batch_requests_ready=0.080470s, total_serialization_initiated=0.660819s, others=0.000412s) I0421 15:19:16.944408 134706785995904 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.742637s (all_items=0.000021s, per_item={'items': '0.00002050'}, temp_paths=0.742617) I0421 15:19:16.945272 134616094606912 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:19:16.945402 134706785995904 async_checkpointer.py:561] Finished blocking save. Time taken: 0.825202s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0. I0421 15:19:16.945631 134706785995904 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:19:16.945808 134616172201536 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:19:16.945913 134706785995904 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776784756.1199749, 'wait_for_prev_duration_secs': 4.2438507080078125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776784756.1201763, 'checkpointer_blocking_duration_secs': 0.8253543376922607, 'get_old_steps_start_time': 1776784756.9455543, 'get_old_steps_duration_secs': 2.7418136596679688e-05, 'checkpoint_manager_blocking_start_time': 1776784756.119886, 'checkpoint_manager_blocking_duration_secs': 0.8259885311126709} I0421 15:19:16.946018 134706785995904 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0421 15:19:16.946079 134706785995904 max_utils.py:750] Memstats: After params initialized: I0421 15:19:16.946157 134706785995904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0421 15:19:16.946184 134706785995904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0421 15:19:16.946207 134706785995904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0421 15:19:16.946224 134706785995904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0421 15:19:16.946241 134706785995904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0421 15:19:16.946260 134706785995904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0421 15:19:16.946278 134706785995904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0421 15:19:16.946295 134706785995904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0421 15:19:17.059308 134616151230016 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0/items I0421 15:19:17.199605 134616084121152 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776784756959869884, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776784756959869884, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0421 15:19:17.353816 134706785995904 metric_logger.py:196] completed step: 0, seconds: 3.693, TFLOP/s/device: 3.679, Tokens/s/device: 554.564, total_weights: 13328, loss: 10.828, lm_loss: 10.828, perplexity: 50404.086 I0421 15:19:17.354779 134706785995904 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/tensorboard/' I0421 15:19:17.469473 134706785995904 metric_logger.py:196] completed step: 1, seconds: 1.353, TFLOP/s/device: 10.042, Tokens/s/device: 1513.629, total_weights: 12332, loss: 10.828, lm_loss: 10.828, perplexity: 50391.758 I0421 15:19:17.587515 134706785995904 metric_logger.py:196] completed step: 2, seconds: 0.010, TFLOP/s/device: 1334.820, Tokens/s/device: 201198.546, total_weights: 15161, loss: 9.823, lm_loss: 9.823, perplexity: 18444.867 I0421 15:19:17.705390 134706785995904 metric_logger.py:196] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.366, Tokens/s/device: 17539.953, total_weights: 13327, loss: 9.405, lm_loss: 9.405, perplexity: 12149.331 I0421 15:19:17.735918 1510775 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0421 15:19:17.736002 1510775 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0421 15:19:32.114356 134706785995904 metric_logger.py:196] completed step: 4, seconds: 0.118, TFLOP/s/device: 114.977, Tokens/s/device: 17330.524, total_weights: 11939, loss: 9.045, lm_loss: 9.045, perplexity: 8471.842 I0421 15:19:32.141810 134706785995904 metric_logger.py:196] completed step: 5, seconds: 0.118, TFLOP/s/device: 115.318, Tokens/s/device: 17382.005, total_weights: 15502, loss: 8.953, lm_loss: 8.953, perplexity: 7728.546 I0421 15:19:32.259139 134706785995904 metric_logger.py:196] completed step: 6, seconds: 14.426, TFLOP/s/device: 0.942, Tokens/s/device: 141.963, total_weights: 13864, loss: 8.775, lm_loss: 8.775, perplexity: 6468.536 I0421 15:19:32.377118 134706785995904 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 2033.697, Tokens/s/device: 306540.937, total_weights: 12988, loss: 8.688, lm_loss: 8.688, perplexity: 5928.887 I0421 15:19:32.440796 134616115578432 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0/items/array_metadatas/process_0 I0421 15:19:32.494474 134706785995904 metric_logger.py:196] completed step: 8, seconds: 0.119, TFLOP/s/device: 114.006, Tokens/s/device: 17184.235, total_weights: 13820, loss: 8.740, lm_loss: 8.740, perplexity: 6250.123 I0421 15:19:32.611933 134706785995904 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0421 15:19:32.613085 134706785995904 checkpointing.py:776] Waited 0.00112152099609375 seconds for step 9 to finish before starting checkpointing. I0421 15:19:32.613590 134706785995904 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:19:54.123345 134616105092672 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 36.384325s (commit=35.935485s, array_metadata_write=0.448839s) I0421 15:19:54.124393 134616094606912 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 333.233 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 37 seconds) (per-host) I0421 15:19:54.124463 134616094606912 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.178998s. I0421 15:19:54.357200 134616094606912 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776784756959869884, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0421 15:19:54.531131 134616094606912 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0421 15:19:54.752862 134616084121152 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776784756959869884, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0421 15:19:54.969173 134616094606912 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 15:19:54.969798 134616094606912 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.573558s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0/items I0421 15:19:54.970535 134616094606912 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0/items I0421 15:19:55.212997 134616094606912 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0 I0421 15:19:55.910525 134616094606912 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0`. I0421 15:19:55.911304 134616094606912 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 39.791111s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/0 I0421 15:19:55.911395 134616094606912 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 38.965935s. I0421 15:19:55.911614 134616172201536 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:19:55.911745 134616172201536 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:19:55.911816 134616172201536 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:19:55.911872 134616172201536 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 15:19:55.911977 134706785995904 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 15:19:55.912065 134706785995904 checkpoint_manager.py:1441] Waiting for previous save to complete took 23.298493 seconds. If this number is high, consider checkpointing less frequently. I0421 15:19:55.913260 134706785995904 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0421 15:19:55.913589 134706785995904 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9. I0421 15:19:56.078416 134616172201536 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9 I0421 15:19:56.093628 134706785995904 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:19:56.093739 134706785995904 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0421 15:19:56.793914 134616105092672 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9/items I0421 15:20:00.435124 134706785995904 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 4.341891s I0421 15:20:00.440729 134706785995904 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.778 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 4 seconds) (per-host) I0421 15:20:00.440803 134706785995904 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 4.442689s (batch_requests_ready=0.092087s, total_serialization_initiated=4.345255s, others=0.005347s) I0421 15:20:00.440893 134706785995904 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 4.443325s (all_items=0.000027s, per_item={'items': '0.00002742'}, temp_paths=4.443297) I0421 15:20:00.442771 134616094606912 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:20:00.442883 134706785995904 async_checkpointer.py:561] Finished blocking save. Time taken: 4.529576s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9. I0421 15:20:00.443122 134706785995904 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:20:00.443289 134616172201536 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:20:00.443422 134706785995904 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776784772.6135335, 'wait_for_prev_duration_secs': 23.298493146896362, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776784795.9132862, 'checkpointer_blocking_duration_secs': 4.52970552444458, 'get_old_steps_start_time': 1776784800.4430203, 'get_old_steps_duration_secs': 5.435943603515625e-05, 'checkpoint_manager_blocking_start_time': 1776784772.613461, 'checkpoint_manager_blocking_duration_secs': 27.82992720603943} I0421 15:20:00.443543 134706785995904 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0421 15:20:00.443580 134706785995904 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:20:01.204812 134616140744256 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9/items/array_metadatas/process_0 I0421 15:20:39.725329 134616105092672 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 39.283604s (commit=38.848808s, array_metadata_write=0.434796s) I0421 15:20:39.726572 134616094606912 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 288.983 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 43 seconds) (per-host) I0421 15:20:39.726689 134616094606912 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.283758s. I0421 15:20:40.131621 134616094606912 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0421 15:20:40.549186 134616094606912 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 15:20:40.549385 134616094606912 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.571236s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9/items I0421 15:20:40.550137 134616094606912 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9/items I0421 15:20:40.811715 134616094606912 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9 I0421 15:20:41.509639 134616094606912 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9`. I0421 15:20:41.510368 134616094606912 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 45.597066s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_10_shardy_true/checkpoints/9 I0421 15:20:41.510443 134616094606912 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 41.067515s. I0421 15:20:41.510620 134616172201536 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:20:41.510730 134616172201536 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:20:41.510803 134616172201536 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:20:41.510850 134616172201536 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 15:20:41.511141 134706785995904 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 15:20:41.511306 134706785995904 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 15:20:41.512468 134706785995904 metric_logger.py:196] completed step: 9, seconds: 0.121, TFLOP/s/device: 112.656, Tokens/s/device: 16980.772, total_weights: 12300, loss: 8.710, lm_loss: 8.710, perplexity: 6065.008 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
cd5a4f898
· fix_gate_fp8_nan_sanitization_20260422_174509
· full log
2026-04-22 18:16:52.734247: 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 00:00:1776881812.747226 2051734 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776881812.751028 2051734 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776881812.762461 2051734 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776881812.762476 2051734 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776881812.762478 2051734 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776881812.762479 2051734 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'} 2026-04-22 18:16:56.653837: 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) I0422 18:16:56.790378 132224132290368 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0422 18:17:26.857482 132224132290368 max_utils.py:800] System Information: Jax Version: 0.8.1 I0422 18:17:26.857611 132224132290368 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0422 18:17:26.857647 132224132290368 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 I0422 18:17:26.857673 132224132290368 train_utils.py:348] WARNING: 'dataset_path' might be pointing your local file system I0422 18:17:26.857755 132224132290368 train.py:705] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0422 18:17:27.550187 132224132290368 maxtext_utils.py:1604] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 18:17:27.550466 132224132290368 checkpointing.py:677] Setting up checkpoint logger... I0422 18:17:27.550512 132224132290368 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 18:17:27.550548 132224132290368 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0422 18:17:27.551005 132224132290368 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 0x78418adb3080>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 18:17:30.067039 132224132290368 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 18:17:30.067523 132224132290368 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 0x783fc0b35a00>}, handler_registry=None I0422 18:17:30.067830 132224132290368 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x783fc0b35a00>` 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`. I0422 18:17:30.067872 132224132290368 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x783fc0b35fa0>` 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`. I0422 18:17:30.067905 132224132290368 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 0x783fc0b35a00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x783fc0b35a00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x783fc0b35fa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x783fc0b35fa0>}). I0422 18:17:30.068416 132224132290368 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0422 18:17:30.068484 132224132290368 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x783fc0b44900> timeout: 600 secs and primary_host=0 for async checkpoint writes I0422 18:17:30.192873 132224132290368 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints I0422 18:17:30.193150 132224132290368 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/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x783fc0b34920> I0422 18:17:30.193432 132224132290368 checkpointing.py:301] Checkpoint manager created! I0422 18:17:30.273949 132224132290368 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0422 18:17:30.277033 132224132290368 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. I0422 18:17:30.345959 132224132290368 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 I0422 18:17:30.372346 132224132290368 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0422 18:17:30.372416 132224132290368 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0422 18:17:30.981803 132224132290368 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 18:17:30.981906 132224132290368 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 18:17:31.285720 132224132290368 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 18:17:31.285810 132224132290368 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 18:17:31.298658 132224132290368 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 18:17:31.298709 132224132290368 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 18:17:31.318035 132224132290368 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:17:31.318109 132224132290368 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 18:17:31.330939 132224132290368 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:17:31.330995 132224132290368 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 18:17:31.343809 132224132290368 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:17:31.343859 132224132290368 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 18:17:31.364102 132224132290368 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 18:17:31.364167 132224132290368 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 18:17:31.380232 132224132290368 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 18:17:31.380287 132224132290368 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0422 18:17:31.536193 132224132290368 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 18:17:31.536296 132224132290368 checkpointing.py:665] 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 I0422 18:17:32.119383 132224132290368 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0422 18:17:32.119498 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 18:17:32.119541 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 18:17:32.119588 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 18:17:32.119629 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0422 18:17:32.119658 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0422 18:17:32.119696 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 18:17:32.119744 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: PartitionSpec('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0422 18:17:32.119775 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0422 18:17:32.119802 132224132290368 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 18:17:32.119838 132224132290368 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 18:17:32.119873 132224132290368 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 18:17:32.858087 132224132290368 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 18:17:32.858178 132224132290368 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0422 18:17:32.870293 132224132290368 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 18:17:32.870347 132224132290368 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0422 18:17:33.631276 132224132290368 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. I0422 18:17:33.631828 132224132290368 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0422 18:17:33.631980 132224132290368 metric_logger.py:301] number parameters: 1.104 billion I0422 18:17:38.418665 132224132290368 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 18:17:38.540170 132224132290368 checkpointing.py:776] Waited 0.12148690223693848 seconds for step 0 to finish before starting checkpointing. I0422 18:17:38.540893 132224132290368 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 18:17:38.541064 132224132290368 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0422 18:17:38.541476 132224132290368 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0. I0422 18:17:38.627840 132224132290368 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0422 18:17:38.714508 132099770156608 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0 I0422 18:17:38.721011 132224132290368 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0422 18:17:38.721129 132224132290368 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 I0422 18:17:39.547661 132099759670848 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0/items I0422 18:17:39.745170 132099728213568 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776881859453012072, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776881859453012072, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0422 18:17:40.049116 2054259 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0422 18:17:40.049171 2054259 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0422 18:17:40.334845 132224132290368 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.614256s I0422 18:17:40.339513 132224132290368 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 7.216 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0422 18:17:40.339578 132224132290368 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.710295s (batch_requests_ready=0.087834s, total_serialization_initiated=1.617921s, others=0.004540s) I0422 18:17:40.339655 132224132290368 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.711248s (all_items=0.000021s, per_item={'items': '0.00002074'}, temp_paths=1.711227) I0422 18:17:40.341096 132098929198656 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0422 18:17:40.341191 132224132290368 async_checkpointer.py:561] Finished blocking save. Time taken: 1.800086s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0. I0422 18:17:40.341390 132224132290368 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 18:17:40.341574 132099919054400 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 18:17:40.341673 132224132290368 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776881858.54088, 'wait_for_prev_duration_secs': 4.5299530029296875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776881858.5410864, 'checkpointer_blocking_duration_secs': 1.8002216815948486, 'get_old_steps_start_time': 1776881860.3413246, 'get_old_steps_duration_secs': 2.86102294921875e-05, 'checkpoint_manager_blocking_start_time': 1776881858.5407946, 'checkpoint_manager_blocking_duration_secs': 1.8008487224578857} I0422 18:17:40.341786 132224132290368 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 18:17:40.341829 132224132290368 max_utils.py:750] Memstats: After params initialized: I0422 18:17:40.341875 132224132290368 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0422 18:17:40.341899 132224132290368 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0422 18:17:40.341919 132224132290368 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0422 18:17:40.341937 132224132290368 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0422 18:17:40.341955 132224132290368 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0422 18:17:40.341974 132224132290368 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0422 18:17:40.341991 132224132290368 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0422 18:17:40.342008 132224132290368 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0422 18:17:40.698876 132224132290368 metric_logger.py:196] completed step: 0, seconds: 4.786, TFLOP/s/device: 2.839, Tokens/s/device: 427.905, total_weights: 13328, loss: 10.828, lm_loss: 10.828, perplexity: 50404.086 I0422 18:17:40.699955 132224132290368 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/tensorboard/' I0422 18:17:40.814825 132224132290368 metric_logger.py:196] completed step: 1, seconds: 2.278, TFLOP/s/device: 5.965, Tokens/s/device: 899.039, total_weights: 12332, loss: 10.828, lm_loss: 10.828, perplexity: 50391.758 I0422 18:17:40.932304 132224132290368 metric_logger.py:196] completed step: 2, seconds: 0.011, TFLOP/s/device: 1236.655, Tokens/s/device: 186402.112, total_weights: 15161, loss: 9.823, lm_loss: 9.823, perplexity: 18444.867 I0422 18:17:41.050269 132224132290368 metric_logger.py:196] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.200, Tokens/s/device: 17514.902, total_weights: 13327, loss: 9.405, lm_loss: 9.405, perplexity: 12149.331 I0422 18:17:57.684010 132224132290368 metric_logger.py:196] completed step: 4, seconds: 0.118, TFLOP/s/device: 114.857, Tokens/s/device: 17312.504, total_weights: 11939, loss: 9.045, lm_loss: 9.045, perplexity: 8471.842 I0422 18:17:57.684945 132099312977472 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0/items/array_metadatas/process_0 I0422 18:17:57.732759 132224132290368 metric_logger.py:196] completed step: 5, seconds: 0.117, TFLOP/s/device: 115.988, Tokens/s/device: 17482.906, total_weights: 15502, loss: 8.953, lm_loss: 8.953, perplexity: 7728.546 I0422 18:17:57.862922 132224132290368 metric_logger.py:196] completed step: 6, seconds: 16.671, TFLOP/s/device: 0.815, Tokens/s/device: 122.847, total_weights: 13864, loss: 8.775, lm_loss: 8.775, perplexity: 6468.536 I0422 18:17:57.973156 132224132290368 metric_logger.py:196] completed step: 7, seconds: 0.009, TFLOP/s/device: 1517.437, Tokens/s/device: 228724.592, total_weights: 12988, loss: 8.688, lm_loss: 8.688, perplexity: 5928.887 I0422 18:17:58.095802 132224132290368 metric_logger.py:196] completed step: 8, seconds: 0.132, TFLOP/s/device: 103.116, Tokens/s/device: 15542.822, total_weights: 13820, loss: 8.740, lm_loss: 8.740, perplexity: 6250.123 I0422 18:17:58.217966 132224132290368 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 18:17:58.228106 132224132290368 checkpointing.py:776] Waited 0.010037660598754883 seconds for step 9 to finish before starting checkpointing. I0422 18:17:58.232640 132224132290368 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 18:18:36.180933 132099130525248 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 55.840436s (commit=55.319180s, array_metadata_write=0.521256s) I0422 18:18:36.182055 132098929198656 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 219.569 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 57 seconds) (per-host) I0422 18:18:36.182124 132098929198656 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 55.840878s. I0422 18:18:36.452611 132098929198656 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776881859453012072, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0422 18:18:36.689867 132098929198656 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0422 18:18:36.883217 132099728213568 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776881859453012072, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA I0422 18:18:37.087968 132098929198656 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0422 18:18:37.088253 132098929198656 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.594558s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0/items I0422 18:18:37.088895 132098929198656 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0/items I0422 18:18:37.349127 132098929198656 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0 I0422 18:18:38.078177 132098929198656 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0`. I0422 18:18:38.078986 132098929198656 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 59.537884s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/0 I0422 18:18:38.079094 132098929198656 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 57.737853s. I0422 18:18:38.079270 132099919054400 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 18:18:38.079402 132099919054400 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 18:18:38.079468 132099919054400 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 18:18:38.079522 132099919054400 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 18:18:38.079683 132224132290368 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. W0422 18:18:38.079995 132224132290368 checkpoint_manager.py:1441] Waiting for previous save to complete took 39.849354 seconds. If this number is high, consider checkpointing less frequently. I0422 18:18:38.081300 132224132290368 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0422 18:18:38.081654 132224132290368 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9. I0422 18:18:38.251361 132224132290368 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0422 18:18:38.251461 132224132290368 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 I0422 18:18:38.263273 132098929198656 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9 I0422 18:18:39.064925 132099312977472 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9/items I0422 18:18:41.943182 132224132290368 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 3.692271s I0422 18:18:41.947685 132224132290368 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 3.268 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 3 seconds) (per-host) I0422 18:18:41.947754 132224132290368 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 3.776321s (batch_requests_ready=0.076650s, total_serialization_initiated=3.695292s, others=0.004378s) I0422 18:18:41.947839 132224132290368 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 3.777200s (all_items=0.000026s, per_item={'items': '0.00002646'}, temp_paths=3.777173) I0422 18:18:41.949618 132098929198656 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0422 18:18:41.949707 132224132290368 async_checkpointer.py:561] Finished blocking save. Time taken: 3.868356s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9. I0422 18:18:41.949929 132224132290368 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 18:18:41.950175 132099738699328 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 18:18:41.950303 132224132290368 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776881878.230466, 'wait_for_prev_duration_secs': 39.84935402870178, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776881918.081328, 'checkpointer_blocking_duration_secs': 3.8684961795806885, 'get_old_steps_start_time': 1776881921.9498405, 'get_old_steps_duration_secs': 4.4345855712890625e-05, 'checkpoint_manager_blocking_start_time': 1776881878.2286053, 'checkpoint_manager_blocking_duration_secs': 43.72166872024536} I0422 18:18:41.950425 132224132290368 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 18:18:41.950460 132224132290368 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 18:18:42.730894 132207630878272 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9/items/array_metadatas/process_0 I0422 18:19:14.851476 132099312977472 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.903035s (commit=32.451405s, array_metadata_write=0.451630s) I0422 18:19:14.852564 132098929198656 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 344.504 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 36 seconds) (per-host) I0422 18:19:14.852621 132098929198656 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.902864s. I0422 18:19:15.377007 132098929198656 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0422 18:19:15.776982 132098929198656 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0422 18:19:15.777186 132098929198656 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.574787s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9/items I0422 18:19:15.777972 132098929198656 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9/items I0422 18:19:16.030886 132098929198656 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9 I0422 18:19:16.744821 132098929198656 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9`. I0422 18:19:16.745563 132098929198656 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 38.664220s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_10_shardy_true/checkpoints/9 I0422 18:19:16.745647 132098929198656 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.795891s. I0422 18:19:16.745878 132099738699328 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 18:19:16.746002 132099738699328 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 18:19:16.746074 132099738699328 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 18:19:16.746117 132099738699328 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 18:19:16.748626 132224132290368 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. I0422 18:19:16.748783 132224132290368 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 18:19:16.749750 132224132290368 metric_logger.py:196] completed step: 9, seconds: 0.112, TFLOP/s/device: 121.669, Tokens/s/device: 18339.258, total_weights: 12300, loss: 8.710, lm_loss: 8.710, perplexity: 6065.008 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops