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 | 8.8880 | 8.8880 | 0 |
| TFLOP/s | 65.417 | 62.978 | -2.439 |
| Tok/s | 9860.4 | 9492.7 | -367.644 |
| Avg s/step | 1.095 | 1.354 | +0.259 |
| Memory % | 5.06 | 5.06 | 0 |
| JAX | 0.8.1 | 0.8.1 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-21 15:15:29.794779: 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 00:00:1776784529.805295 1504621 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776784529.808335 1504621 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776784529.818183 1504621 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776784529.818195 1504621 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776784529.818197 1504621 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776784529.818199 1504621 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. 2026-04-21 15:15:33.225556: 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 15:15:33.341180 131342151302272 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0421 15:16:06.108695 131342151302272 max_utils.py:800] System Information: Jax Version: 0.8.1 I0421 15:16:06.108972 131342151302272 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0421 15:16:06.109025 131342151302272 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 15:16:06.109071 131342151302272 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system I0421 15:16:06.109241 131342151302272 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0421 15:16:06.464756 131342151302272 maxtext_utils.py:1551] Num_devices: 8, shape (1, 1, 1, 2, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0421 15:16:06.465147 131342151302272 checkpointing.py:677] Setting up checkpoint logger... I0421 15:16:06.465196 131342151302272 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 15:16:06.465288 131342151302272 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0421 15:16:06.466142 131342151302272 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 0x77742dfa6ed0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 15:16:08.825397 131342151302272 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0421 15:16:08.825658 131342151302272 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 0x777285b86480>}, handler_registry=None I0421 15:16:08.826057 131342151302272 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x777285b86480>` 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 15:16:08.826103 131342151302272 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x777285fb3cb0>` 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 15:16:08.826133 131342151302272 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 0x777285b86480>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x777285b86480>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x777285fb3cb0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x777285fb3cb0>}). I0421 15:16:08.826432 131342151302272 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0421 15:16:08.826493 131342151302272 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x777285ba5120> timeout: 600 secs and primary_host=0 for async checkpoint writes I0421 15:16:08.943644 131342151302272 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints I0421 15:16:08.943912 131342151302272 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_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x777285b85130> I0421 15:16:08.944007 131342151302272 checkpointing.py:301] Checkpoint manager created! I0421 15:16:09.293122 131342151302272 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0421 15:16:09.296077 131342151302272 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 15:16:09.349640 131342151302272 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 15:16:09.387728 131342151302272 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0421 15:16:09.387815 131342151302272 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0421 15:16:10.001526 131342151302272 nnx_wrappers.py:437] Unknown Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 15:16:10.001624 131342151302272 nnx_wrappers.py:437] Unknown Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0421 15:16:10.301680 131342151302272 attentions.py:1088] attentions/inputs_q Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 15:16:10.301766 131342151302272 attentions.py:1088] attentions/inputs_q Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0421 15:16:10.314231 131342151302272 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 15:16:10.314283 131342151302272 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0421 15:16:10.333035 131342151302272 attentions.py:1154] attentions/query Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:16:10.333109 131342151302272 attentions.py:1154] attentions/query Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0421 15:16:10.345644 131342151302272 attentions.py:1155] attentions/key Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:16:10.345696 131342151302272 attentions.py:1155] attentions/key Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0421 15:16:10.358168 131342151302272 attentions.py:1156] attentions/value Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:16:10.358216 131342151302272 attentions.py:1156] attentions/value Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0421 15:16:10.377793 131342151302272 attentions.py:1197] attentions/out Logical: bfloat16[2,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 15:16:10.377850 131342151302272 attentions.py:1197] attentions/out Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0421 15:16:10.393533 131342151302272 linears.py:525] linears/x Logical: bfloat16[2,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 15:16:10.393583 131342151302272 linears.py:525] linears/x Physical: bfloat16[2,2048,7168]....................................... ('fsdp', None, 'tensor'). I0421 15:16:10.547083 131342151302272 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0421 15:16:10.547181 131342151302272 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: 2 tensor: 4 I0421 15:16:12.218732 131342151302272 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: ('tensor',) I0421 15:16:12.218847 131342151302272 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, 'tensor') I0421 15:16:12.218889 131342151302272 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, 'tensor') I0421 15:16:12.218933 131342151302272 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0421 15:16:12.218973 131342151302272 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: ('tensor', None) I0421 15:16:12.219002 131342151302272 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: ('tensor', None) I0421 15:16:12.219040 131342151302272 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, 'tensor', None) I0421 15:16:12.219098 131342151302272 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: ('tensor', None, None, 'fsdp') I0421 15:16:12.219128 131342151302272 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, 'tensor', None) I0421 15:16:12.219154 131342151302272 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, 'tensor', None) I0421 15:16:12.219192 131342151302272 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0421 15:16:12.219226 131342151302272 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0421 15:16:12.629145 131342151302272 train.py:155] train/xent Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 15:16:12.629237 131342151302272 train.py:155] train/xent Physical: float32[2,2048]............................................. ('fsdp', None). I0421 15:16:12.641232 131342151302272 train.py:162] train/z_loss Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 15:16:12.641281 131342151302272 train.py:162] train/z_loss Physical: float32[2,2048]............................................. ('fsdp', None). I0421 15:16:21.574035 131342151302272 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 15:16:21.574909 131342151302272 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0421 15:16:21.575151 131342151302272 metric_logger.py:301] number parameters: 1.104 billion I0421 15:16:25.640388 131342151302272 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0421 15:16:25.697728 131342151302272 checkpointing.py:776] Waited 0.05733060836791992 seconds for step 0 to finish before starting checkpointing. I0421 15:16:25.698224 131342151302272 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 15:16:25.698390 131342151302272 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0421 15:16:25.698861 131342151302272 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0. I0421 15:16:25.781945 131342151302272 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0421 15:16:25.865370 131250490705472 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0 I0421 15:16:25.882123 131342151302272 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:16:25.882221 131342151302272 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 15:16:26.519967 131342151302272 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.638362s I0421 15:16:26.520371 131342151302272 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 16.746 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 736 milliseconds) (per-host) I0421 15:16:26.520431 131342151302272 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.737032s (batch_requests_ready=0.094693s, total_serialization_initiated=0.642073s, others=0.000265s) I0421 15:16:26.520496 131342151302272 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.737971s (all_items=0.000028s, per_item={'items': '0.00002837'}, temp_paths=0.737943) I0421 15:16:26.521293 131250427790912 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:16:26.521396 131342151302272 async_checkpointer.py:561] Finished blocking save. Time taken: 0.822963s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0. I0421 15:16:26.521594 131342151302272 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:16:26.521796 131342151302272 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776784585.6982102, 'wait_for_prev_duration_secs': 4.172325134277344e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776784585.6984115, 'checkpointer_blocking_duration_secs': 0.8230991363525391, 'get_old_steps_start_time': 1776784586.5215278, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776784585.6981137, 'checkpoint_manager_blocking_duration_secs': 0.8236572742462158} I0421 15:16:26.521878 131342151302272 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0421 15:16:26.521928 131342151302272 max_utils.py:750] Memstats: After params initialized: I0421 15:16:26.522005 131342151302272 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_0(process=0,(0,0,0,0)) I0421 15:16:26.522037 131342151302272 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_1(process=0,(1,0,0,0)) I0421 15:16:26.522077 131342151302272 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_2(process=0,(0,1,0,0)) I0421 15:16:26.522096 131342151302272 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_3(process=0,(1,1,0,0)) I0421 15:16:26.522113 131342151302272 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_4(process=0,(0,2,0,0)) I0421 15:16:26.522130 131342151302272 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_5(process=0,(1,2,0,0)) I0421 15:16:26.522146 131342151302272 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_6(process=0,(0,3,0,0)) I0421 15:16:26.522161 131342151302272 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_7(process=0,(1,3,0,0)) I0421 15:16:26.618141 131250480219712 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0/items I0421 15:16:26.757712 131250448762432 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776784586509214787, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776784586509214787, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0421 15:16:26.758039 131250417305152 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:16:26.934128 131342151302272 metric_logger.py:196] completed step: 0, seconds: 4.064, TFLOP/s/device: 0.836, Tokens/s/device: 125.969, total_weights: 3650, loss: 10.896, lm_loss: 10.896, perplexity: 53934.855 I0421 15:16:26.935154 131342151302272 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/tensorboard/' I0421 15:16:26.984990 131342151302272 metric_logger.py:196] completed step: 1, seconds: 1.292, TFLOP/s/device: 2.630, Tokens/s/device: 396.429, total_weights: 2676, loss: 10.833, lm_loss: 10.833, perplexity: 50674.457 I0421 15:16:27.037610 131342151302272 metric_logger.py:196] completed step: 2, seconds: 0.014, TFLOP/s/device: 249.672, Tokens/s/device: 37633.223, total_weights: 3783, loss: 10.136, lm_loss: 10.136, perplexity: 25223.543 I0421 15:16:27.090226 131342151302272 metric_logger.py:196] completed step: 3, seconds: 0.052, TFLOP/s/device: 65.623, Tokens/s/device: 9891.426, total_weights: 3680, loss: 9.403, lm_loss: 9.403, perplexity: 12120.116 I0421 15:16:27.124267 1507746 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0421 15:16:27.124341 1507746 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0421 15:16:35.348189 131342151302272 metric_logger.py:196] completed step: 4, seconds: 0.053, TFLOP/s/device: 64.658, Tokens/s/device: 9745.884, total_weights: 3736, loss: 9.171, lm_loss: 9.171, perplexity: 9614.685 I0421 15:16:35.378094 131342151302272 metric_logger.py:196] completed step: 5, seconds: 0.053, TFLOP/s/device: 64.529, Tokens/s/device: 9726.444, total_weights: 4096, loss: 9.075, lm_loss: 9.075, perplexity: 8729.922 I0421 15:16:35.430404 131342151302272 metric_logger.py:196] completed step: 6, seconds: 8.274, TFLOP/s/device: 0.411, Tokens/s/device: 61.877, total_weights: 2931, loss: 8.899, lm_loss: 8.899, perplexity: 7322.795 I0421 15:16:35.483174 131342151302272 metric_logger.py:196] completed step: 7, seconds: 0.010, TFLOP/s/device: 341.214, Tokens/s/device: 51431.441, total_weights: 3526, loss: 8.818, lm_loss: 8.818, perplexity: 6751.396 I0421 15:16:35.535824 131342151302272 metric_logger.py:196] completed step: 8, seconds: 0.055, TFLOP/s/device: 61.646, Tokens/s/device: 9292.027, total_weights: 3012, loss: 8.750, lm_loss: 8.750, perplexity: 6308.487 I0421 15:16:35.587918 131342151302272 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0421 15:16:35.588910 131342151302272 checkpointing.py:776] Waited 0.0010066032409667969 seconds for step 9 to finish before starting checkpointing. I0421 15:16:35.589248 131342151302272 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:16:35.687933 131250490705472 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_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_0 I0421 15:17:14.805306 131250469733952 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 47.678940s (commit=47.250642s, array_metadata_write=0.428298s) I0421 15:17:14.806533 131250427790912 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 257.772 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 49 seconds) (per-host) I0421 15:17:14.806591 131250427790912 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 48.285147s. I0421 15:17:15.112177 131250427790912 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776784586509214787, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0421 15:17:15.284545 131250427790912 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 15:17:15.508149 131250448762432 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776784586509214787, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0421 15:17:15.747336 131250427790912 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 15:17:15.747519 131250427790912 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.600177s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0/items I0421 15:17:15.748287 131250427790912 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0/items I0421 15:17:15.996137 131250427790912 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0 I0421 15:17:16.675294 131250427790912 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_09_pdb_lt_1/checkpoints/0`. I0421 15:17:16.676064 131250427790912 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 50.977637s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/0 I0421 15:17:16.676145 131250427790912 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 50.154702s. I0421 15:17:16.676279 131250417305152 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:17:16.676331 131250417305152 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:17:16.676384 131250417305152 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:17:16.676426 131250417305152 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 15:17:16.676568 131342151302272 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 15:17:16.676728 131342151302272 checkpoint_manager.py:1441] Waiting for previous save to complete took 41.087477 seconds. If this number is high, consider checkpointing less frequently. I0421 15:17:16.677764 131342151302272 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0421 15:17:16.678102 131342151302272 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/9. I0421 15:17:16.856967 131342151302272 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:17:16.857141 131250417305152 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/9 I0421 15:17:16.857225 131342151302272 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 15:17:17.657746 131250333419072 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/9/items I0421 15:17:24.561482 131342151302272 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 7.705305s I0421 15:17:24.566247 131342151302272 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.583 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 7 seconds) (per-host) I0421 15:17:24.566312 131342151302272 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 7.795976s (batch_requests_ready=0.082476s, total_serialization_initiated=7.708859s, others=0.004640s) I0421 15:17:24.566391 131342151302272 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 7.796577s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=7.796559) I0421 15:17:24.567927 131250417305152 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:17:24.568016 131342151302272 async_checkpointer.py:561] Finished blocking save. Time taken: 7.890201s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/9. I0421 15:17:24.568233 131342151302272 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:17:24.568433 131250427790912 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:17:24.568560 131342151302272 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776784595.5892084, 'wait_for_prev_duration_secs': 41.08747720718384, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776784636.6777918, 'checkpointer_blocking_duration_secs': 7.890353202819824, 'get_old_steps_start_time': 1776784644.5681639, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776784595.589156, 'checkpoint_manager_blocking_duration_secs': 48.97937297821045} I0421 15:17:24.568692 131342151302272 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0421 15:17:24.568732 131342151302272 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:17:25.341668 131250490705472 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_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_0 I0421 15:17:59.875818 131250469733952 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 35.308761s (commit=34.838017s, array_metadata_write=0.470744s) I0421 15:17:59.876976 131250417305152 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 293.152 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 43 seconds) (per-host) I0421 15:17:59.877036 131250417305152 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.308959s. I0421 15:18:00.298905 131250417305152 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 15:18:00.723446 131250417305152 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 15:18:00.723627 131250417305152 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.588320s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/9/items I0421 15:18:00.724348 131250417305152 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/9/items I0421 15:18:00.993792 131250417305152 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/9 I0421 15:18:01.688182 131250417305152 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_09_pdb_lt_1/checkpoints/9`. I0421 15:18:01.688919 131250417305152 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 45.011112s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_09_pdb_lt_1/checkpoints/9 I0421 15:18:01.689006 131250417305152 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 37.120929s. I0421 15:18:01.689259 131250427790912 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:18:01.689384 131250427790912 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:18:01.689439 131250427790912 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:18:01.689483 131250427790912 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 15:18:01.689802 131342151302272 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 15:18:01.689866 131342151302272 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 15:18:01.690787 131342151302272 metric_logger.py:196] completed step: 9, seconds: 0.052, TFLOP/s/device: 65.417, Tokens/s/device: 9860.376, total_weights: 3004, loss: 8.888, lm_loss: 8.888, perplexity: 7242.487 Per train step: Total TFLOPs: 3.40 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 18:14:09.342307: 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 00:00:1776881649.355325 2047543 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776881649.359168 2047543 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776881649.370596 2047543 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776881649.370610 2047543 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776881649.370611 2047543 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776881649.370613 2047543 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 18:14:13.049800: 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 18:14:13.186304 128117644035904 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0422 18:14:42.207057 128117644035904 max_utils.py:800] System Information: Jax Version: 0.8.1 I0422 18:14:42.207318 128117644035904 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0422 18:14:42.207368 128117644035904 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 18:14:42.207399 128117644035904 train_utils.py:348] WARNING: 'dataset_path' might be pointing your local file system I0422 18:14:42.207584 128117644035904 train.py:705] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0422 18:14:42.866603 128117644035904 maxtext_utils.py:1604] Num_devices: 8, shape (1, 1, 1, 2, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0422 18:14:42.866881 128117644035904 checkpointing.py:677] Setting up checkpoint logger... I0422 18:14:42.866926 128117644035904 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 18:14:42.866962 128117644035904 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0422 18:14:42.867399 128117644035904 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 0x74856d1b1820>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 18:14:45.354719 128117644035904 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 18:14:45.355130 128117644035904 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 0x7483a3bbde80>}, handler_registry=None I0422 18:14:45.355354 128117644035904 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7483a3bbde80>` 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 18:14:45.355396 128117644035904 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7483a3bbe390>` 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 18:14:45.355425 128117644035904 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 0x7483a3bbde80>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7483a3bbde80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7483a3bbe390>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7483a3bbe390>}). I0422 18:14:45.355700 128117644035904 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0422 18:14:45.355755 128117644035904 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7483a3be07c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0422 18:14:45.483083 128117644035904 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints I0422 18:14:45.483317 128117644035904 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_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7483a3bbe330> I0422 18:14:45.483593 128117644035904 checkpointing.py:301] Checkpoint manager created! I0422 18:14:45.561838 128117644035904 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0422 18:14:45.564743 128117644035904 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 18:14:45.625579 128117644035904 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 18:14:45.648149 128117644035904 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0422 18:14:45.648211 128117644035904 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0422 18:14:46.267155 128117644035904 nnx_wrappers.py:437] Unknown Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 18:14:46.267257 128117644035904 nnx_wrappers.py:437] Unknown Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 18:14:46.568942 128117644035904 attentions.py:1088] attentions/inputs_q Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 18:14:46.569056 128117644035904 attentions.py:1088] attentions/inputs_q Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 18:14:46.582027 128117644035904 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 18:14:46.582090 128117644035904 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 18:14:46.601092 128117644035904 attentions.py:1154] attentions/query Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:14:46.601151 128117644035904 attentions.py:1154] attentions/query Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 18:14:46.613847 128117644035904 attentions.py:1155] attentions/key Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:14:46.613898 128117644035904 attentions.py:1155] attentions/key Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 18:14:46.626514 128117644035904 attentions.py:1156] attentions/value Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:14:46.626564 128117644035904 attentions.py:1156] attentions/value Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 18:14:46.646424 128117644035904 attentions.py:1198] attentions/out Logical: bfloat16[2,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 18:14:46.646486 128117644035904 attentions.py:1198] attentions/out Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 18:14:46.662286 128117644035904 linears.py:525] linears/x Logical: bfloat16[2,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 18:14:46.662343 128117644035904 linears.py:525] linears/x Physical: bfloat16[2,2048,7168]....................................... ('fsdp', None, 'tensor'). I0422 18:14:46.836749 128117644035904 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 18:14:46.836878 128117644035904 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: 2 tensor: 4 I0422 18:14:47.415966 128117644035904 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: ('tensor',) I0422 18:14:47.416099 128117644035904 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, 'tensor') I0422 18:14:47.416145 128117644035904 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, 'tensor') I0422 18:14:47.416192 128117644035904 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0422 18:14:47.416233 128117644035904 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: ('tensor', None) I0422 18:14:47.416262 128117644035904 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: ('tensor', None) I0422 18:14:47.416302 128117644035904 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, 'tensor', None) I0422 18:14:47.416342 128117644035904 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: ('tensor', None, None, 'fsdp') I0422 18:14:47.416372 128117644035904 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, 'tensor', None) I0422 18:14:47.416399 128117644035904 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, 'tensor', None) I0422 18:14:47.416434 128117644035904 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0422 18:14:47.416468 128117644035904 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0422 18:14:48.139958 128117644035904 train.py:155] train/xent Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 18:14:48.140208 128117644035904 train.py:155] train/xent Physical: float32[2,2048]............................................. ('fsdp', None). I0422 18:14:48.152276 128117644035904 train.py:162] train/z_loss Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 18:14:48.152327 128117644035904 train.py:162] train/z_loss Physical: float32[2,2048]............................................. ('fsdp', None). I0422 18:14:57.741766 128117644035904 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. I0422 18:14:57.742405 128117644035904 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0422 18:14:57.742609 128117644035904 metric_logger.py:301] number parameters: 1.104 billion I0422 18:15:01.333056 128117644035904 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 18:15:01.390541 128117644035904 checkpointing.py:776] Waited 0.05748748779296875 seconds for step 0 to finish before starting checkpointing. I0422 18:15:01.391234 128117644035904 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 18:15:01.391418 128117644035904 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0422 18:15:01.391843 128117644035904 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0. I0422 18:15:01.520117 128117644035904 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0422 18:15:01.604230 128117644035904 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0422 18:15:01.604330 128117644035904 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 18:15:01.615753 127992598627904 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0 I0422 18:15:02.373624 127992588142144 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0/items I0422 18:15:02.527873 127992068048448 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776881702270395679, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776881702270395679, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0422 18:15:02.897806 2050339 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0422 18:15:02.897864 2050339 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0422 18:15:03.985551 128117644035904 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 2.381833s I0422 18:15:03.990399 128117644035904 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.999 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 2 seconds) (per-host) I0422 18:15:03.990467 128117644035904 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 2.468956s (batch_requests_ready=0.078671s, total_serialization_initiated=2.385549s, others=0.004736s) I0422 18:15:03.990546 128117644035904 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.469866s (all_items=0.000020s, per_item={'items': '0.00002027'}, temp_paths=2.469846) I0422 18:15:03.991935 127992609113664 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0422 18:15:03.992022 128117644035904 async_checkpointer.py:561] Finished blocking save. Time taken: 2.600566s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0. I0422 18:15:03.992225 128117644035904 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 18:15:03.992423 127991531177536 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 18:15:03.992535 128117644035904 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776881701.3912227, 'wait_for_prev_duration_secs': 4.315376281738281e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776881701.3914402, 'checkpointer_blocking_duration_secs': 2.6007070541381836, 'get_old_steps_start_time': 1776881703.9921632, 'get_old_steps_duration_secs': 2.6702880859375e-05, 'checkpoint_manager_blocking_start_time': 1776881701.3911386, 'checkpoint_manager_blocking_duration_secs': 2.6013710498809814} I0422 18:15:03.992656 128117644035904 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 18:15:03.992696 128117644035904 max_utils.py:750] Memstats: After params initialized: I0422 18:15:03.992741 128117644035904 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_0(process=0,(0,0,0,0)) I0422 18:15:03.992769 128117644035904 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_1(process=0,(1,0,0,0)) I0422 18:15:03.992793 128117644035904 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_2(process=0,(0,1,0,0)) I0422 18:15:03.992812 128117644035904 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_3(process=0,(1,1,0,0)) I0422 18:15:03.992831 128117644035904 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_4(process=0,(0,2,0,0)) I0422 18:15:03.992849 128117644035904 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_5(process=0,(1,2,0,0)) I0422 18:15:03.992867 128117644035904 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_6(process=0,(0,3,0,0)) I0422 18:15:03.992884 128117644035904 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_7(process=0,(1,3,0,0)) I0422 18:15:04.395465 128117644035904 metric_logger.py:196] completed step: 0, seconds: 3.590, TFLOP/s/device: 0.946, Tokens/s/device: 142.624, total_weights: 3650, loss: 10.896, lm_loss: 10.896, perplexity: 53934.855 I0422 18:15:04.396510 128117644035904 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/tensorboard/' I0422 18:15:04.446634 128117644035904 metric_logger.py:196] completed step: 1, seconds: 3.060, TFLOP/s/device: 1.110, Tokens/s/device: 167.309, total_weights: 2676, loss: 10.833, lm_loss: 10.833, perplexity: 50674.457 I0422 18:15:04.499307 128117644035904 metric_logger.py:196] completed step: 2, seconds: 0.016, TFLOP/s/device: 218.106, Tokens/s/device: 32875.305, total_weights: 3783, loss: 10.136, lm_loss: 10.136, perplexity: 25223.543 I0422 18:15:04.551758 128117644035904 metric_logger.py:196] completed step: 3, seconds: 0.053, TFLOP/s/device: 63.758, Tokens/s/device: 9610.331, total_weights: 3680, loss: 9.403, lm_loss: 9.403, perplexity: 12120.116 I0422 18:15:13.365624 127992361649728 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_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_0 I0422 18:15:13.366735 128117644035904 metric_logger.py:196] completed step: 4, seconds: 0.051, TFLOP/s/device: 66.024, Tokens/s/device: 9951.796, total_weights: 3736, loss: 9.171, lm_loss: 9.171, perplexity: 9614.685 I0422 18:15:13.400837 128117644035904 metric_logger.py:196] completed step: 5, seconds: 0.052, TFLOP/s/device: 65.167, Tokens/s/device: 9822.730, total_weights: 4096, loss: 9.075, lm_loss: 9.075, perplexity: 8729.922 I0422 18:15:13.453314 128117644035904 metric_logger.py:196] completed step: 6, seconds: 8.834, TFLOP/s/device: 0.385, Tokens/s/device: 57.960, total_weights: 2931, loss: 8.899, lm_loss: 8.899, perplexity: 7322.795 I0422 18:15:13.505890 128117644035904 metric_logger.py:196] completed step: 7, seconds: 0.013, TFLOP/s/device: 253.662, Tokens/s/device: 38234.635, total_weights: 3526, loss: 8.818, lm_loss: 8.818, perplexity: 6751.396 I0422 18:15:13.558366 128117644035904 metric_logger.py:196] completed step: 8, seconds: 0.054, TFLOP/s/device: 63.353, Tokens/s/device: 9549.210, total_weights: 3012, loss: 8.750, lm_loss: 8.750, perplexity: 6308.487 I0422 18:15:13.610534 128117644035904 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 18:15:13.612426 128117644035904 checkpointing.py:776] Waited 0.0017657279968261719 seconds for step 9 to finish before starting checkpointing. I0422 18:15:13.615126 128117644035904 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 18:15:43.709283 127991732504128 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 39.718074s (commit=39.167333s, array_metadata_write=0.550741s) I0422 18:15:43.710396 127992609113664 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 299.529 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 42 seconds) (per-host) I0422 18:15:43.710464 127992609113664 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.718376s. I0422 18:15:43.970409 127992609113664 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776881702270395679, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0422 18:15:44.212784 127992609113664 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 18:15:44.418468 127992068048448 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776881702270395679, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0422 18:15:44.627119 127992609113664 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0422 18:15:44.627355 127992609113664 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.613757s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0/items I0422 18:15:44.628147 127992609113664 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0/items I0422 18:15:44.894810 127992609113664 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0 I0422 18:15:45.649785 127992609113664 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_09_pdb_lt_1/checkpoints/0`. I0422 18:15:45.650535 127992609113664 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 44.259079s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/0 I0422 18:15:45.650620 127992609113664 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 41.658535s. I0422 18:15:45.650849 127991531177536 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 18:15:45.650970 127991531177536 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 18:15:45.651057 127991531177536 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 18:15:45.651115 127991531177536 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 18:15:45.651387 128117644035904 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 18:15:45.651542 128117644035904 checkpoint_manager.py:1441] Waiting for previous save to complete took 32.038258 seconds. If this number is high, consider checkpointing less frequently. I0422 18:15:45.652706 128117644035904 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0422 18:15:45.653039 128117644035904 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/9. I0422 18:15:45.815803 127992609113664 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/9 I0422 18:15:45.821144 128117644035904 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0422 18:15:45.821252 128117644035904 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 18:15:46.559154 127992361649728 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/9/items I0422 18:15:50.902392 128117644035904 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 5.081716s I0422 18:15:50.907090 128117644035904 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.385 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 5 seconds) (per-host) I0422 18:15:50.907153 128117644035904 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 5.174350s (batch_requests_ready=0.084665s, total_serialization_initiated=5.085114s, others=0.004572s) I0422 18:15:50.907229 128117644035904 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 5.174936s (all_items=0.000032s, per_item={'items': '0.00003242'}, temp_paths=5.174903) I0422 18:15:50.908990 127991732504128 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0422 18:15:50.909089 128117644035904 async_checkpointer.py:561] Finished blocking save. Time taken: 5.256331s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/9. I0422 18:15:50.909294 128117644035904 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 18:15:50.909479 127991531177536 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 18:15:50.909583 128117644035904 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776881713.6132562, 'wait_for_prev_duration_secs': 32.03825759887695, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776881745.6527364, 'checkpointer_blocking_duration_secs': 5.256458282470703, 'get_old_steps_start_time': 1776881750.9092128, 'get_old_steps_duration_secs': 3.933906555175781e-05, 'checkpoint_manager_blocking_start_time': 1776881713.6126812, 'checkpoint_manager_blocking_duration_secs': 37.29687809944153} I0422 18:15:50.909720 128117644035904 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 18:15:50.909761 128117644035904 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 18:15:51.704015 128100941694528 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_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_0 I0422 18:16:29.316644 128100931208768 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 38.408783s (commit=37.947471s, array_metadata_write=0.461312s) I0422 18:16:29.317918 127991732504128 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 289.934 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 43 seconds) (per-host) I0422 18:16:29.318058 127991732504128 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.408909s. I0422 18:16:29.841650 127991732504128 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 18:16:30.220001 127991732504128 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0422 18:16:30.220243 127991732504128 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.557694s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/9/items I0422 18:16:30.220998 127991732504128 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/9/items I0422 18:16:30.514623 127991732504128 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/9 I0422 18:16:31.259596 127991732504128 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_09_pdb_lt_1/checkpoints/9`. I0422 18:16:31.260357 127991732504128 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 45.607605s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_09_pdb_lt_1/checkpoints/9 I0422 18:16:31.260431 127991732504128 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 40.351298s. I0422 18:16:31.260652 127991531177536 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 18:16:31.260781 127991531177536 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 18:16:31.260853 127991531177536 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 18:16:31.260896 127991531177536 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 18:16:31.261104 128117644035904 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 18:16:31.261264 128117644035904 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 18:16:31.262526 128117644035904 metric_logger.py:196] completed step: 9, seconds: 0.054, TFLOP/s/device: 62.978, Tokens/s/device: 9492.732, total_weights: 3004, loss: 8.888, lm_loss: 8.888, perplexity: 7242.487 Per train step: Total TFLOPs: 3.40 split as 93.93% learnable weight flops and 6.07% attention flops