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 | 5.9150 | 5.9150 | 0 |
| TFLOP/s | 114.096 | 313.017 | +198.921 |
| Tok/s | 17197.8 | 47181.3 | +29983.53 |
| Avg s/step | 1.952 | 2.220 | +0.268 |
| 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 14:50:21.406719: 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 00:00:1776783021.416939 1476398 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776783021.419953 1476398 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776783021.431068 1476398 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776783021.431079 1476398 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776783021.431081 1476398 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776783021.431082 1476398 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. 2026-04-21 14:50:24.832298: 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 14:50:24.948929 139470404086912 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0421 14:51:23.425797 139470404086912 max_utils.py:800] System Information: Jax Version: 0.8.1 I0421 14:51:23.426088 139470404086912 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0421 14:51:23.426137 139470404086912 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 14:51:23.426171 139470404086912 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system I0421 14:51:23.426198 139470404086912 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 14:51:23.426367 139470404086912 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0421 14:51:23.783014 139470404086912 maxtext_utils.py:1551] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 14:51:23.783405 139470404086912 checkpointing.py:677] Setting up checkpoint logger... I0421 14:51:23.783450 139470404086912 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 14:51:23.783544 139470404086912 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0421 14:51:23.784395 139470404086912 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 0x7ed8af99b050>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 14:51:26.174512 139470404086912 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0421 14:51:26.174784 139470404086912 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 0x7ed707d7ae70>}, handler_registry=None I0421 14:51:26.175178 139470404086912 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ed707d7ae70>` 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 14:51:26.175224 139470404086912 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ed707d7b3e0>` 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 14:51:26.175258 139470404086912 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 0x7ed707d7ae70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ed707d7ae70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ed707d7b3e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ed707d7b3e0>}). I0421 14:51:26.175552 139470404086912 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0421 14:51:26.175615 139470404086912 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7ed707da5120> timeout: 600 secs and primary_host=0 for async checkpoint writes I0421 14:51:26.330549 139470404086912 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints I0421 14:51:26.330823 139470404086912 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_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ed708d71340> I0421 14:51:26.330916 139470404086912 checkpointing.py:301] Checkpoint manager created! I0421 14:51:26.974039 139470404086912 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 14:51:26.974153 139470404086912 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 14:51:27.253425 139470404086912 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 14:51:27.253511 139470404086912 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 14:51:27.265333 139470404086912 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 14:51:27.265382 139470404086912 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 14:51:27.283895 139470404086912 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 14:51:27.283951 139470404086912 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 14:51:27.295920 139470404086912 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 14:51:27.295973 139470404086912 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 14:51:27.307706 139470404086912 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 14:51:27.307761 139470404086912 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 14:51:27.326590 139470404086912 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 14:51:27.326646 139470404086912 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 14:51:27.341589 139470404086912 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 14:51:27.341641 139470404086912 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0421 14:51:27.487978 139470404086912 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0421 14:51:27.488086 139470404086912 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 14:51:28.025551 139470404086912 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0421 14:51:28.025660 139470404086912 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 14:51:28.025701 139470404086912 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 14:51:28.025751 139470404086912 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 14:51:28.025791 139470404086912 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 14:51:28.025820 139470404086912 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 14:51:28.025858 139470404086912 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 14:51:28.025897 139470404086912 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 14:51:28.025926 139470404086912 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 14:51:28.025951 139470404086912 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 14:51:28.025985 139470404086912 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 14:51:28.026019 139470404086912 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 14:51:28.424408 139470404086912 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 14:51:28.424498 139470404086912 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0421 14:51:28.435726 139470404086912 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 14:51:28.435781 139470404086912 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0421 14:51:29.170669 139470404086912 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 14:51:29.171285 139470404086912 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0421 14:51:29.171438 139470404086912 metric_logger.py:301] number parameters: 1.104 billion I0421 14:51:31.321615 139470404086912 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0421 14:51:31.439513 139470404086912 checkpointing.py:776] Waited 0.11787891387939453 seconds for step 0 to finish before starting checkpointing. I0421 14:51:31.439973 139470404086912 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 14:51:31.440157 139470404086912 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0421 14:51:31.440564 139470404086912 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0. I0421 14:51:31.521576 139470404086912 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0421 14:51:31.603075 139390540056128 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0 I0421 14:51:31.611714 139470404086912 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 14:51:31.611809 139470404086912 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 14:51:32.358428 139390529570368 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0/items I0421 14:51:32.518836 139390498113088 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783092262090898, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776783092262090898, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0421 14:51:32.884377 1478458 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0421 14:51:32.884439 1478458 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0421 14:51:33.010640 139470404086912 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.399415s I0421 14:51:33.015156 139470404086912 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 8.270 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0421 14:51:33.015218 139470404086912 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.492237s (batch_requests_ready=0.084930s, total_serialization_initiated=1.402908s, others=0.004399s) I0421 14:51:33.015296 139470404086912 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.493153s (all_items=0.000021s, per_item={'items': '0.00002098'}, temp_paths=1.493132) I0421 14:51:33.016753 139390414227008 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 14:51:33.016845 139470404086912 async_checkpointer.py:561] Finished blocking save. Time taken: 1.576648s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0. I0421 14:51:33.017033 139470404086912 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 14:51:33.017251 139390550541888 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 14:51:33.017382 139470404086912 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776783091.4399617, 'wait_for_prev_duration_secs': 4.0531158447265625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776783091.4401793, 'checkpointer_blocking_duration_secs': 1.576775074005127, 'get_old_steps_start_time': 1776783093.0169759, 'get_old_steps_duration_secs': 2.1696090698242188e-05, 'checkpoint_manager_blocking_start_time': 1776783091.4398768, 'checkpoint_manager_blocking_duration_secs': 1.5774800777435303} I0421 14:51:33.017459 139470404086912 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0421 14:51:33.017499 139470404086912 max_utils.py:750] Memstats: After params initialized: I0421 14:51:33.017544 139470404086912 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0421 14:51:33.017570 139470404086912 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0421 14:51:33.017599 139470404086912 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0421 14:51:33.017617 139470404086912 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0421 14:51:33.017635 139470404086912 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0421 14:51:33.017652 139470404086912 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0421 14:51:33.017668 139470404086912 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0421 14:51:33.017684 139470404086912 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0421 14:51:33.391740 139470404086912 metric_logger.py:196] completed step: 0, seconds: 2.150, TFLOP/s/device: 6.321, Tokens/s/device: 952.716, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0421 14:51:33.392849 139470404086912 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/tensorboard/' I0421 14:51:33.508208 139470404086912 metric_logger.py:196] completed step: 1, seconds: 2.068, TFLOP/s/device: 6.570, Tokens/s/device: 990.368, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0421 14:51:33.627402 139470404086912 metric_logger.py:196] completed step: 2, seconds: 0.007, TFLOP/s/device: 1865.083, Tokens/s/device: 281125.601, total_weights: 16384, loss: 9.775, lm_loss: 9.775, perplexity: 17589.287 I0421 14:51:33.733648 139390445684288 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_02_synthetic/checkpoints/0/items/array_metadatas/process_0 I0421 14:51:33.746272 139470404086912 metric_logger.py:196] completed step: 3, seconds: 0.117, TFLOP/s/device: 115.944, Tokens/s/device: 17476.341, total_weights: 16384, loss: 8.838, lm_loss: 8.838, perplexity: 6892.933 I0421 14:51:48.625281 139470404086912 metric_logger.py:196] completed step: 4, seconds: 0.119, TFLOP/s/device: 114.266, Tokens/s/device: 17223.400, total_weights: 16384, loss: 7.965, lm_loss: 7.965, perplexity: 2878.644 I0421 14:51:48.647536 139470404086912 metric_logger.py:196] completed step: 5, seconds: 0.119, TFLOP/s/device: 114.219, Tokens/s/device: 17216.305, total_weights: 16384, loss: 7.233, lm_loss: 7.233, perplexity: 1383.750 I0421 14:51:48.765708 139470404086912 metric_logger.py:196] completed step: 6, seconds: 14.894, TFLOP/s/device: 0.912, Tokens/s/device: 137.505, total_weights: 16384, loss: 6.676, lm_loss: 6.676, perplexity: 793.531 I0421 14:51:48.884772 139470404086912 metric_logger.py:196] completed step: 7, seconds: 0.006, TFLOP/s/device: 2307.597, Tokens/s/device: 347826.087, total_weights: 16384, loss: 6.293, lm_loss: 6.293, perplexity: 540.860 I0421 14:51:49.004377 139470404086912 metric_logger.py:196] completed step: 8, seconds: 0.120, TFLOP/s/device: 113.265, Tokens/s/device: 17072.500, total_weights: 16384, loss: 6.055, lm_loss: 6.055, perplexity: 426.371 I0421 14:51:49.122195 139470404086912 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0421 14:51:49.123276 139470404086912 checkpointing.py:776] Waited 0.0012292861938476562 seconds for step 9 to finish before starting checkpointing. I0421 14:51:49.123701 139470404086912 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 14:52:20.358803 139390435198528 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 47.342803s (commit=46.881431s, array_metadata_write=0.461372s) I0421 14:52:20.360018 139390414227008 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 258.754 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 48 seconds) (per-host) I0421 14:52:20.360099 139390414227008 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 47.343205s. I0421 14:52:20.575084 139390414227008 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783092262090898, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0421 14:52:20.768987 139390414227008 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 14:52:20.957540 139390498113088 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776783092262090898, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0421 14:52:21.167703 139390414227008 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 14:52:21.168336 139390414227008 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.550745s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0/items I0421 14:52:21.169152 139390414227008 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0/items I0421 14:52:21.425019 139390414227008 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0 I0421 14:52:22.170253 139390414227008 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_02_synthetic/checkpoints/0`. I0421 14:52:22.170986 139390414227008 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 50.730789s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/0 I0421 14:52:22.171089 139390414227008 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 49.154200s. I0421 14:52:22.171324 139390550541888 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 14:52:22.171456 139390550541888 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 14:52:22.171522 139390550541888 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 14:52:22.171578 139390550541888 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 14:52:22.171776 139470404086912 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 14:52:22.171926 139470404086912 checkpoint_manager.py:1441] Waiting for previous save to complete took 33.048263 seconds. If this number is high, consider checkpointing less frequently. I0421 14:52:22.173165 139470404086912 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0421 14:52:22.173501 139470404086912 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/9. I0421 14:52:22.351083 139390550541888 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/9 I0421 14:52:22.357026 139470404086912 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 14:52:22.357145 139470404086912 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 14:52:23.065605 139390403741248 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/9/items I0421 14:52:30.314961 139470404086912 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 7.958417s I0421 14:52:30.319876 139470404086912 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.532 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 8 seconds) (per-host) I0421 14:52:30.319945 139470404086912 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 8.055632s (batch_requests_ready=0.088901s, total_serialization_initiated=7.961981s, others=0.004750s) I0421 14:52:30.320068 139470404086912 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 8.056275s (all_items=0.000026s, per_item={'items': '0.00002575'}, temp_paths=8.056249) I0421 14:52:30.321977 139390550541888 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 14:52:30.322093 139470404086912 async_checkpointer.py:561] Finished blocking save. Time taken: 8.148879s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/9. I0421 14:52:30.322326 139470404086912 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 14:52:30.322512 139390414227008 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 14:52:30.322643 139470404086912 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776783109.1236281, 'wait_for_prev_duration_secs': 33.04826307296753, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776783142.1731896, 'checkpointer_blocking_duration_secs': 8.149014234542847, 'get_old_steps_start_time': 1776783150.322229, 'get_old_steps_duration_secs': 4.553794860839844e-05, 'checkpoint_manager_blocking_start_time': 1776783109.1235456, 'checkpoint_manager_blocking_duration_secs': 41.19906759262085} I0421 14:52:30.322823 139470404086912 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0421 14:52:30.322860 139470404086912 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 14:52:31.067810 139390508598848 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_02_synthetic/checkpoints/9/items/array_metadatas/process_0 I0421 14:53:09.523707 139390403741248 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 39.202937s (commit=38.762783s, array_metadata_write=0.440154s) I0421 14:53:09.525023 139390550541888 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 267.385 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 47 seconds) (per-host) I0421 14:53:09.525159 139390550541888 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.203021s. I0421 14:53:09.964835 139390550541888 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 14:53:10.368963 139390550541888 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 14:53:10.369181 139390550541888 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.547580s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/9/items I0421 14:53:10.369910 139390550541888 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/9/items I0421 14:53:10.611632 139390550541888 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/9 I0421 14:53:11.274410 139390550541888 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_02_synthetic/checkpoints/9`. I0421 14:53:11.275147 139390550541888 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 49.101942s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_02_synthetic/checkpoints/9 I0421 14:53:11.275224 139390550541888 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 40.953087s. I0421 14:53:11.275460 139390414227008 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 14:53:11.275585 139390414227008 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 14:53:11.275650 139390414227008 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 14:53:11.275693 139390414227008 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 14:53:11.276161 139470404086912 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 14:53:11.276313 139470404086912 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 14:53:11.278000 139470404086912 metric_logger.py:196] completed step: 9, seconds: 0.119, TFLOP/s/device: 114.096, Tokens/s/device: 17197.800, total_weights: 16384, loss: 5.915, lm_loss: 5.915, perplexity: 370.653 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 17:47:58.346415: 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 00:00:1776880078.359420 2018428 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776880078.363469 2018428 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776880078.374906 2018428 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776880078.374923 2018428 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776880078.374925 2018428 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776880078.374926 2018428 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 17:48:02.052109: 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 17:48:02.187617 132284632323904 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0422 17:48:32.663387 132284632323904 max_utils.py:800] System Information: Jax Version: 0.8.1 I0422 17:48:32.663559 132284632323904 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0422 17:48:32.663601 132284632323904 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 17:48:32.663629 132284632323904 train_utils.py:348] WARNING: 'dataset_path' might be pointing your local file system I0422 17:48:32.663653 132284632323904 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 17:48:32.663766 132284632323904 train.py:706] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0422 17:48:33.331442 132284632323904 maxtext_utils.py:1604] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 17:48:33.331732 132284632323904 checkpointing.py:677] Setting up checkpoint logger... I0422 17:48:33.331776 132284632323904 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 17:48:33.331821 132284632323904 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0422 17:48:33.332266 132284632323904 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 0x784fa0faadb0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 17:48:35.808532 132284632323904 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 17:48:35.808923 132284632323904 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 0x784dd41548c0>}, handler_registry=None I0422 17:48:35.809166 132284632323904 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x784dd41548c0>` 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 17:48:35.809210 132284632323904 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x784e461f91f0>` 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 17:48:35.809239 132284632323904 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 0x784dd41548c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x784dd41548c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x784e461f91f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x784e461f91f0>}). I0422 17:48:35.809510 132284632323904 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0422 17:48:35.809561 132284632323904 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x784dd41787c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0422 17:48:35.967142 132284632323904 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints I0422 17:48:35.967379 132284632323904 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_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x784dd4154b00> I0422 17:48:35.967664 132284632323904 checkpointing.py:301] Checkpoint manager created! I0422 17:48:36.350138 132284632323904 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 17:48:36.350239 132284632323904 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 17:48:36.631590 132284632323904 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 17:48:36.631683 132284632323904 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 17:48:36.643887 132284632323904 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 17:48:36.643940 132284632323904 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 17:48:36.663198 132284632323904 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 17:48:36.663264 132284632323904 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 17:48:36.675705 132284632323904 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 17:48:36.675757 132284632323904 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 17:48:36.687936 132284632323904 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 17:48:36.687986 132284632323904 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 17:48:36.707465 132284632323904 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 17:48:36.707530 132284632323904 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 17:48:36.722865 132284632323904 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 17:48:36.722920 132284632323904 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0422 17:48:36.882792 132284632323904 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 17:48:36.882895 132284632323904 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 17:48:37.506795 132284632323904 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0422 17:48:37.506906 132284632323904 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 17:48:37.506947 132284632323904 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 17:48:37.506990 132284632323904 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 17:48:37.507029 132284632323904 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 17:48:37.507085 132284632323904 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 17:48:37.507131 132284632323904 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 17:48:37.507170 132284632323904 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 17:48:37.507198 132284632323904 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 17:48:37.507224 132284632323904 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 17:48:37.507257 132284632323904 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 17:48:37.507292 132284632323904 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 17:48:37.919742 132284632323904 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 17:48:37.919836 132284632323904 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0422 17:48:37.931463 132284632323904 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 17:48:37.931514 132284632323904 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0422 17:48:38.704857 132284632323904 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 17:48:38.705455 132284632323904 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0422 17:48:38.705611 132284632323904 metric_logger.py:301] number parameters: 1.104 billion I0422 17:48:41.715435 132284632323904 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 17:48:41.833328 132284632323904 checkpointing.py:776] Waited 0.1178748607635498 seconds for step 0 to finish before starting checkpointing. I0422 17:48:41.833973 132284632323904 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 17:48:41.834149 132284632323904 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0422 17:48:41.834556 132284632323904 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0. I0422 17:48:41.919108 132284632323904 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0422 17:48:42.025506 132284632323904 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0422 17:48:42.025607 132284632323904 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 17:48:42.031433 132254066017856 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0 I0422 17:48:42.813442 132253919217216 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0/items I0422 17:48:42.973693 132253682239040 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776880122712322420, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776880122712322420, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0422 17:48:43.359696 2020488 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0422 17:48:43.359756 2020488 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0422 17:48:43.638380 132284632323904 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.613340s I0422 17:48:43.643165 132284632323904 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 7.164 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0422 17:48:43.643229 132284632323904 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.722744s (batch_requests_ready=0.100683s, total_serialization_initiated=1.617385s, others=0.004677s) I0422 17:48:43.643303 132284632323904 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.723668s (all_items=0.000021s, per_item={'items': '0.00002074'}, temp_paths=1.723647) I0422 17:48:43.644529 132252866446912 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0422 17:48:43.644630 132284632323904 async_checkpointer.py:561] Finished blocking save. Time taken: 1.810441s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0. I0422 17:48:43.644830 132284632323904 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 17:48:43.645040 132252675606080 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 17:48:43.645157 132284632323904 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776880121.833961, 'wait_for_prev_duration_secs': 4.100799560546875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776880121.8341715, 'checkpointer_blocking_duration_secs': 1.8105790615081787, 'get_old_steps_start_time': 1776880123.644767, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1776880121.8338764, 'checkpoint_manager_blocking_duration_secs': 1.8112568855285645} I0422 17:48:43.645272 132284632323904 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 17:48:43.645312 132284632323904 max_utils.py:750] Memstats: After params initialized: I0422 17:48:43.645357 132284632323904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0422 17:48:43.645381 132284632323904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0422 17:48:43.645402 132284632323904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0422 17:48:43.645420 132284632323904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0422 17:48:43.645437 132284632323904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0422 17:48:43.645454 132284632323904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0422 17:48:43.645470 132284632323904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0422 17:48:43.645486 132284632323904 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0422 17:48:44.036506 132284632323904 metric_logger.py:196] completed step: 0, seconds: 3.009, TFLOP/s/device: 4.515, Tokens/s/device: 680.599, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0422 17:48:44.037624 132284632323904 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/tensorboard/' I0422 17:48:44.152284 132284632323904 metric_logger.py:196] completed step: 1, seconds: 2.318, TFLOP/s/device: 5.861, Tokens/s/device: 883.379, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0422 17:48:44.271144 132284632323904 metric_logger.py:196] completed step: 2, seconds: 0.008, TFLOP/s/device: 1656.361, Tokens/s/device: 249664.757, total_weights: 16384, loss: 9.775, lm_loss: 9.775, perplexity: 17589.287 I0422 17:48:44.390438 132284632323904 metric_logger.py:196] completed step: 3, seconds: 0.116, TFLOP/s/device: 116.726, Tokens/s/device: 17594.199, total_weights: 16384, loss: 8.838, lm_loss: 8.838, perplexity: 6892.933 I0422 17:48:44.429670 132253078259264 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_02_synthetic/checkpoints/0/items/array_metadatas/process_0 I0422 17:49:01.305165 132284632323904 metric_logger.py:196] completed step: 4, seconds: 0.119, TFLOP/s/device: 114.237, Tokens/s/device: 17219.055, total_weights: 16384, loss: 7.965, lm_loss: 7.965, perplexity: 2878.644 I0422 17:49:01.365134 132284632323904 metric_logger.py:196] completed step: 5, seconds: 0.119, TFLOP/s/device: 114.101, Tokens/s/device: 17198.522, total_weights: 16384, loss: 7.233, lm_loss: 7.233, perplexity: 1383.750 I0422 17:49:01.484381 132284632323904 metric_logger.py:196] completed step: 6, seconds: 16.951, TFLOP/s/device: 0.802, Tokens/s/device: 120.820, total_weights: 16384, loss: 6.676, lm_loss: 6.676, perplexity: 793.531 I0422 17:49:01.655437 132284632323904 metric_logger.py:196] completed step: 7, seconds: 0.077, TFLOP/s/device: 177.260, Tokens/s/device: 26718.503, total_weights: 16384, loss: 6.293, lm_loss: 6.293, perplexity: 540.860 I0422 17:49:01.789219 132284632323904 metric_logger.py:196] completed step: 8, seconds: 0.230, TFLOP/s/device: 59.011, Tokens/s/device: 8894.796, total_weights: 16384, loss: 6.055, lm_loss: 6.055, perplexity: 426.371 I0422 17:49:01.895977 132284632323904 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 17:49:01.907653 132284632323904 checkpointing.py:776] Waited 0.011889934539794922 seconds for step 9 to finish before starting checkpointing. I0422 17:49:01.917390 132284632323904 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 17:49:37.174206 132253067773504 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 53.530298s (commit=53.021641s, array_metadata_write=0.508657s) I0422 17:49:37.175443 132252866446912 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 228.700 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 55 seconds) (per-host) I0422 17:49:37.175574 132252866446912 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 53.530893s. I0422 17:49:37.449729 132252866446912 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776880122712322420, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0422 17:49:37.685428 132252866446912 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 17:49:37.901900 132253682239040 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776880122712322420, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0422 17:49:38.106511 132252866446912 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0422 17:49:38.106729 132252866446912 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.598517s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0/items I0422 17:49:38.107651 132252866446912 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0/items I0422 17:49:38.360952 132252866446912 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0 I0422 17:49:39.102900 132252866446912 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_02_synthetic/checkpoints/0`. I0422 17:49:39.103744 132252866446912 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 57.269556s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/0 I0422 17:49:39.103850 132252866446912 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 55.459172s. I0422 17:49:39.104199 132252675606080 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 17:49:39.104338 132252675606080 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 17:49:39.104421 132252675606080 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 17:49:39.104486 132252675606080 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 17:49:39.104615 132284632323904 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 17:49:39.104698 132284632323904 checkpoint_manager.py:1441] Waiting for previous save to complete took 37.187526 seconds. If this number is high, consider checkpointing less frequently. I0422 17:49:39.106215 132284632323904 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0422 17:49:39.106503 132284632323904 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/9. I0422 17:49:39.278013 132252866446912 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/9 I0422 17:49:39.311473 132284632323904 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0422 17:49:39.311584 132284632323904 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 17:49:39.734495 132284632323904 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.423526s I0422 17:49:39.735003 132284632323904 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 22.772 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 541 milliseconds) (per-host) I0422 17:49:39.735076 132284632323904 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.542050s (batch_requests_ready=0.114481s, total_serialization_initiated=0.427164s, others=0.000404s) I0422 17:49:39.735142 132284632323904 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.542680s (all_items=0.000027s, per_item={'items': '0.00002742'}, temp_paths=0.542653) I0422 17:49:39.736059 132253269100096 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0422 17:49:39.736165 132284632323904 async_checkpointer.py:561] Finished blocking save. Time taken: 0.629905s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/9. I0422 17:49:39.736363 132284632323904 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 17:49:39.736609 132252675606080 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 17:49:39.736741 132284632323904 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776880141.9171424, 'wait_for_prev_duration_secs': 37.18752598762512, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776880179.1062398, 'checkpointer_blocking_duration_secs': 0.6300210952758789, 'get_old_steps_start_time': 1776880179.7362745, 'get_old_steps_duration_secs': 4.57763671875e-05, 'checkpoint_manager_blocking_start_time': 1776880141.9170709, 'checkpoint_manager_blocking_duration_secs': 37.819637060165405} I0422 17:49:39.736870 132284632323904 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 17:49:39.736903 132284632323904 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 17:49:39.983253 132177503192640 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/9/items I0422 17:49:41.271376 132254269441600 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_02_synthetic/checkpoints/9/items/array_metadatas/process_0 I0422 17:50:21.242621 132253462038080 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 40.781229s (commit=40.276747s, array_metadata_write=0.504482s) I0422 17:50:21.243723 132253269100096 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 300.514 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 42 seconds) (per-host) I0422 17:50:21.243832 132253269100096 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.507622s. I0422 17:50:21.720017 132253269100096 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 17:50:22.146813 132253269100096 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0422 17:50:22.147037 132253269100096 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.601457s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/9/items I0422 17:50:22.147785 132253269100096 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/9/items I0422 17:50:22.399624 132253269100096 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/9 I0422 17:50:23.120522 132253269100096 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_02_synthetic/checkpoints/9`. I0422 17:50:23.121279 132253269100096 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 44.015024s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_02_synthetic/checkpoints/9 I0422 17:50:23.121359 132253269100096 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 43.385151s. I0422 17:50:23.121530 132252675606080 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 17:50:23.121639 132252675606080 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 17:50:23.121695 132252675606080 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 17:50:23.121737 132252675606080 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 17:50:23.121912 132284632323904 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 17:50:23.122071 132284632323904 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 17:50:23.122997 132284632323904 metric_logger.py:196] completed step: 9, seconds: 0.043, TFLOP/s/device: 313.017, Tokens/s/device: 47181.330, total_weights: 16384, loss: 5.915, lm_loss: 5.915, perplexity: 370.653 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops