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 | 66.099 | 66.633 | +0.534 |
| Tok/s | 9963.2 | 10043.6 | +80.474 |
| Avg s/step | 2.031 | 2.154 | +0.123 |
| Memory % | 1.79 | 1.79 | 0 |
| JAX | 0.8.1 | 0.8.1 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-21 15:26:37.651584: 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 00:00:1776785197.662120 1517749 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776785197.665200 1517749 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776785197.675086 1517749 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776785197.675099 1517749 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776785197.675101 1517749 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776785197.675102 1517749 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. 2026-04-21 15:26:41.079857: 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 15:26:41.195642 139303774743680 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0421 15:27:09.172411 139303774743680 max_utils.py:800] System Information: Jax Version: 0.8.1 I0421 15:27:09.172713 139303774743680 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0421 15:27:09.172765 139303774743680 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 15:27:09.172798 139303774743680 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system I0421 15:27:09.172826 139303774743680 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 15:27:09.173000 139303774743680 train.py:703] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0421 15:27:09.532153 139303774743680 maxtext_utils.py:1551] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 15:27:09.532520 139303774743680 checkpointing.py:677] Setting up checkpoint logger... I0421 15:27:09.532570 139303774743680 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 15:27:09.532660 139303774743680 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0421 15:27:09.533518 139303774743680 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 0x7eb1e3ba6fc0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 15:27:11.917417 139303774743680 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0421 15:27:11.917696 139303774743680 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 0x7eb03b73d0a0>}, handler_registry=None I0421 15:27:11.918094 139303774743680 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7eb03b73d0a0>` 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 15:27:11.918139 139303774743680 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eb03b382b40>` 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 15:27:11.918170 139303774743680 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 0x7eb03b73d0a0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7eb03b73d0a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eb03b382b40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eb03b382b40>}). I0421 15:27:11.918463 139303774743680 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0421 15:27:11.918526 139303774743680 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7eb03b3a5080> timeout: 600 secs and primary_host=0 for async checkpoint writes I0421 15:27:12.033767 139303774743680 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints I0421 15:27:12.034060 139303774743680 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_11_optimizer_offload_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7eb03bb46a50> I0421 15:27:12.034158 139303774743680 checkpointing.py:301] Checkpoint manager created! I0421 15:27:12.678892 139303774743680 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 15:27:12.678990 139303774743680 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:27:12.960174 139303774743680 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 15:27:12.960263 139303774743680 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:27:12.972341 139303774743680 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 15:27:12.972392 139303774743680 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0421 15:27:12.991253 139303774743680 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:27:12.991312 139303774743680 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:27:13.003434 139303774743680 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:27:13.003483 139303774743680 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:27:13.015433 139303774743680 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 15:27:13.015483 139303774743680 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:27:13.034619 139303774743680 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 15:27:13.034680 139303774743680 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0421 15:27:13.049758 139303774743680 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 15:27:13.049810 139303774743680 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0421 15:27:13.197080 139303774743680 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0421 15:27:13.197179 139303774743680 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0421 15:27:18.062701 139303774743680 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0421 15:27:18.062813 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0421 15:27:18.062854 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0421 15:27:18.062898 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0421 15:27:18.062937 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0421 15:27:18.062966 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0421 15:27:18.063013 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0421 15:27:18.063075 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: PartitionSpec('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0421 15:27:18.063109 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0421 15:27:18.063137 139303774743680 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0421 15:27:18.063175 139303774743680 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 15:27:18.063212 139303774743680 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 15:27:18.468445 139303774743680 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 15:27:18.468536 139303774743680 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0421 15:27:18.480494 139303774743680 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0421 15:27:18.480548 139303774743680 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0421 15:27:26.389032 139303774743680 max_utils.py:791] Total memory size: 2.4 GB, Output size: 0.5 GB, Temp size: 1.9 GB, Argument size: 0.5 GB, Host temp size: 0.0 GB. I0421 15:27:26.389926 139303774743680 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0421 15:27:26.390174 139303774743680 metric_logger.py:301] number parameters: 1.104 billion I0421 15:27:35.295631 139303774743680 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0421 15:27:35.502516 139303774743680 checkpointing.py:776] Waited 0.20686030387878418 seconds for step 0 to finish before starting checkpointing. I0421 15:27:35.503095 139303774743680 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 15:27:35.503270 139303774743680 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0421 15:27:35.503738 139303774743680 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0. I0421 15:27:35.608802 139303774743680 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0421 15:27:35.693651 139218877679168 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0 I0421 15:27:35.705154 139303774743680 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:27:35.705256 139303774743680 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 15:27:36.709460 139303774743680 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.004799s I0421 15:27:36.709933 139303774743680 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 11.225 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0421 15:27:36.710002 139303774743680 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.099561s (batch_requests_ready=0.090471s, total_serialization_initiated=1.008748s, others=0.000341s) I0421 15:27:36.710086 139303774743680 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.100700s (all_items=0.000029s, per_item={'items': '0.00002933'}, temp_paths=1.100671) I0421 15:27:36.710881 139218810570304 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:27:36.711002 139303774743680 async_checkpointer.py:561] Finished blocking save. Time taken: 1.207684s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0. I0421 15:27:36.711255 139303774743680 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:27:36.711483 139303774743680 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776785255.50308, 'wait_for_prev_duration_secs': 4.363059997558594e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776785255.5032914, 'checkpointer_blocking_duration_secs': 1.2078642845153809, 'get_old_steps_start_time': 1776785256.7111807, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1776785255.502952, 'checkpoint_manager_blocking_duration_secs': 1.208498477935791} I0421 15:27:36.711578 139303774743680 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0421 15:27:36.711626 139303774743680 max_utils.py:750] Memstats: After params initialized: I0421 15:27:36.711731 139303774743680 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_0(process=0,(0,0,0,0)) I0421 15:27:36.711760 139303774743680 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_1(process=0,(1,0,0,0)) I0421 15:27:36.711782 139303774743680 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_2(process=0,(0,1,0,0)) I0421 15:27:36.711802 139303774743680 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_3(process=0,(1,1,0,0)) I0421 15:27:36.711820 139303774743680 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_4(process=0,(0,2,0,0)) I0421 15:27:36.711838 139303774743680 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_5(process=0,(1,2,0,0)) I0421 15:27:36.711856 139303774743680 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_6(process=0,(0,3,0,0)) I0421 15:27:36.711873 139303774743680 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_7(process=0,(1,3,0,0)) I0421 15:27:36.799420 139218867193408 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0/items I0421 15:27:36.948191 139218831541824 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776785256685780976, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776785256685780976, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0/_CHECKPOINT_METADATA I0421 15:27:36.948478 139218800084544 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:27:37.123735 139303774743680 metric_logger.py:196] completed step: 0, seconds: 8.905, TFLOP/s/device: 1.526, Tokens/s/device: 229.990, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.840 I0421 15:27:37.124471 139303774743680 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/tensorboard/' I0421 15:27:37.317178 139303774743680 metric_logger.py:196] completed step: 1, seconds: 1.825, TFLOP/s/device: 7.444, Tokens/s/device: 1121.982, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.840 I0421 15:27:37.506403 1520663 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0421 15:27:37.506492 1520663 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0421 15:27:37.514985 139303774743680 metric_logger.py:196] completed step: 2, seconds: 0.008, TFLOP/s/device: 1712.519, Tokens/s/device: 258129.569, total_weights: 16384, loss: 9.775, lm_loss: 9.775, perplexity: 17589.299 I0421 15:27:37.710835 139303774743680 metric_logger.py:196] completed step: 3, seconds: 0.194, TFLOP/s/device: 69.876, Tokens/s/device: 10532.433, total_weights: 16384, loss: 8.838, lm_loss: 8.838, perplexity: 6892.454 I0421 15:27:53.137274 139218877679168 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_11_optimizer_offload_true/checkpoints/0/items/array_metadatas/process_0 I0421 15:27:53.137799 139303774743680 metric_logger.py:196] completed step: 4, seconds: 0.210, TFLOP/s/device: 64.811, Tokens/s/device: 9768.988, total_weights: 16384, loss: 7.965, lm_loss: 7.965, perplexity: 2879.114 I0421 15:27:53.163673 139303774743680 metric_logger.py:196] completed step: 5, seconds: 0.184, TFLOP/s/device: 73.989, Tokens/s/device: 11152.437, total_weights: 16384, loss: 7.233, lm_loss: 7.233, perplexity: 1383.691 I0421 15:27:53.363461 139303774743680 metric_logger.py:196] completed step: 6, seconds: 15.445, TFLOP/s/device: 0.880, Tokens/s/device: 132.596, total_weights: 16384, loss: 6.677, lm_loss: 6.677, perplexity: 793.605 I0421 15:27:53.568382 139303774743680 metric_logger.py:196] completed step: 7, seconds: 0.006, TFLOP/s/device: 2191.119, Tokens/s/device: 330269.311, total_weights: 16384, loss: 6.293, lm_loss: 6.293, perplexity: 540.987 I0421 15:27:53.773086 139303774743680 metric_logger.py:196] completed step: 8, seconds: 0.201, TFLOP/s/device: 67.493, Tokens/s/device: 10173.314, total_weights: 16384, loss: 6.055, lm_loss: 6.055, perplexity: 426.449 I0421 15:27:53.974343 139303774743680 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0421 15:27:53.975390 139303774743680 checkpointing.py:776] Waited 0.0011684894561767578 seconds for step 9 to finish before starting checkpointing. I0421 15:27:53.975724 139303774743680 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:28:25.663869 139218856707648 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 48.155145s (commit=47.699853s, array_metadata_write=0.455292s) I0421 15:28:25.665279 139218810570304 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 252.459 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 50 seconds) (per-host) I0421 15:28:25.665400 139218810570304 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 48.954327s. I0421 15:28:25.913483 139218810570304 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776785256685780976, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0/_CHECKPOINT_METADATA I0421 15:28:26.108586 139218810570304 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 15:28:26.305177 139218831541824 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776785256685780976, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0/_CHECKPOINT_METADATA I0421 15:28:26.541881 139218810570304 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 15:28:26.542482 139218810570304 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.586750s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0/items I0421 15:28:26.543255 139218810570304 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0/items I0421 15:28:26.797082 139218810570304 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0 I0421 15:28:27.487957 139218810570304 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_11_optimizer_offload_true/checkpoints/0`. I0421 15:28:27.488715 139218810570304 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 51.985409s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/0 I0421 15:28:27.488787 139218810570304 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 50.777715s. I0421 15:28:27.488926 139218800084544 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:28:27.488979 139218800084544 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:28:27.489032 139218800084544 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:28:27.489087 139218800084544 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 15:28:27.489216 139303774743680 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 15:28:27.489442 139303774743680 checkpoint_manager.py:1441] Waiting for previous save to complete took 33.513743 seconds. If this number is high, consider checkpointing less frequently. I0421 15:28:27.490430 139303774743680 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0421 15:28:27.490670 139303774743680 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/9. I0421 15:28:27.659669 139303774743680 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0421 15:28:27.659768 139303774743680 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 15:28:27.728067 139218800084544 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/9 I0421 15:28:28.027032 139303774743680 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.367885s I0421 15:28:28.027528 139303774743680 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 27.388 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 450 milliseconds) (per-host) I0421 15:28:28.027582 139303774743680 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.450688s (batch_requests_ready=0.079483s, total_serialization_initiated=0.370831s, others=0.000374s) I0421 15:28:28.027640 139303774743680 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.451280s (all_items=0.000019s, per_item={'items': '0.00001931'}, temp_paths=0.451260) I0421 15:28:28.028793 139218739267136 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0421 15:28:28.028895 139303774743680 async_checkpointer.py:561] Finished blocking save. Time taken: 0.538424s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/9. I0421 15:28:28.029085 139303774743680 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 15:28:28.029244 139218728781376 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0421 15:28:28.029339 139303774743680 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776785273.975676, 'wait_for_prev_duration_secs': 33.51374292373657, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776785307.4904532, 'checkpointer_blocking_duration_secs': 0.5385372638702393, 'get_old_steps_start_time': 1776785308.0290074, 'get_old_steps_duration_secs': 2.7418136596679688e-05, 'checkpoint_manager_blocking_start_time': 1776785273.9756274, 'checkpoint_manager_blocking_duration_secs': 34.05367612838745} I0421 15:28:28.029457 139303774743680 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0421 15:28:28.029490 139303774743680 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 15:28:28.491726 139218718295616 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/9/items I0421 15:28:29.824206 139218877679168 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_11_optimizer_offload_true/checkpoints/9/items/array_metadatas/process_0 I0421 15:29:01.352599 139218777015872 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.383963s (commit=31.903358s, array_metadata_write=0.480604s) I0421 15:29:01.353772 139218739267136 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 374.126 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 33 seconds) (per-host) I0421 15:29:01.353898 139218739267136 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.324957s. I0421 15:29:01.757969 139218739267136 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 15:29:02.232974 139218739267136 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0421 15:29:02.233178 139218739267136 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.611454s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/9/items I0421 15:29:02.233908 139218739267136 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/9/items I0421 15:29:02.487545 139218739267136 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/9 I0421 15:29:03.251506 139218739267136 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_11_optimizer_offload_true/checkpoints/9`. I0421 15:29:03.252263 139218739267136 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 35.761797s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260421_144657/linen_main_20260421_144657_11_optimizer_offload_true/checkpoints/9 I0421 15:29:03.252347 139218739267136 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 35.223409s. I0421 15:29:03.252459 139218728781376 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 15:29:03.252511 139218728781376 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0421 15:29:03.252571 139218728781376 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 15:29:03.252609 139218728781376 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 15:29:03.252747 139303774743680 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 15:29:03.252894 139303774743680 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 15:29:03.253805 139303774743680 metric_logger.py:196] completed step: 9, seconds: 0.206, TFLOP/s/device: 66.099, Tokens/s/device: 9963.173, total_weights: 16384, loss: 5.915, lm_loss: 5.915, perplexity: 370.686 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
cd5a4f898
· fix_gate_fp8_nan_sanitization_20260422_174509
· full log
2026-04-22 18:25:54.244098: 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 00:00:1776882354.257366 2061756 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776882354.261207 2061756 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776882354.272683 2061756 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776882354.272697 2061756 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776882354.272699 2061756 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776882354.272700 2061756 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 18:25:57.971557: 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 18:25:58.105818 125986646951744 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0422 18:26:56.295407 125986646951744 max_utils.py:800] System Information: Jax Version: 0.8.1 I0422 18:26:56.295675 125986646951744 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0422 18:26:56.295737 125986646951744 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 18:26:56.295765 125986646951744 train_utils.py:348] WARNING: 'dataset_path' might be pointing your local file system I0422 18:26:56.295789 125986646951744 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 18:26:56.295959 125986646951744 train.py:705] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0422 18:26:56.951247 125986646951744 maxtext_utils.py:1604] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 18:26:56.951516 125986646951744 checkpointing.py:677] Setting up checkpoint logger... I0422 18:26:56.951562 125986646951744 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 18:26:56.951599 125986646951744 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0422 18:26:56.952022 125986646951744 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 0x7295465a30e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 18:26:59.389694 125986646951744 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 18:26:59.390090 125986646951744 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 0x7293781459d0>}, handler_registry=None I0422 18:26:59.390333 125986646951744 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7293781459d0>` 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 18:26:59.390373 125986646951744 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x729378146060>` 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 18:26:59.390404 125986646951744 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 0x7293781459d0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7293781459d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x729378146060>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x729378146060>}). I0422 18:26:59.390676 125986646951744 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0422 18:26:59.390729 125986646951744 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x729378158720> timeout: 600 secs and primary_host=0 for async checkpoint writes I0422 18:26:59.524196 125986646951744 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints I0422 18:26:59.524432 125986646951744 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_11_optimizer_offload_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x729378145ee0> I0422 18:26:59.524704 125986646951744 checkpointing.py:301] Checkpoint manager created! I0422 18:26:59.927409 125986646951744 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 18:26:59.927513 125986646951744 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 18:27:00.209162 125986646951744 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 18:27:00.209260 125986646951744 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 18:27:00.221716 125986646951744 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 18:27:00.221781 125986646951744 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0422 18:27:00.241020 125986646951744 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:27:00.241096 125986646951744 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 18:27:00.253667 125986646951744 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:27:00.253719 125986646951744 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 18:27:00.266155 125986646951744 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 18:27:00.266210 125986646951744 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 18:27:00.285893 125986646951744 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 18:27:00.285952 125986646951744 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0422 18:27:00.301987 125986646951744 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 18:27:00.302060 125986646951744 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0422 18:27:00.453967 125986646951744 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 18:27:00.454090 125986646951744 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0422 18:27:03.904558 125986646951744 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0422 18:27:03.904676 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 18:27:03.904719 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 18:27:03.904765 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 18:27:03.904804 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0422 18:27:03.904841 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0422 18:27:03.904882 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 18:27:03.904922 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: PartitionSpec('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0422 18:27:03.904951 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0422 18:27:03.904977 125986646951744 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 18:27:03.905011 125986646951744 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 18:27:03.905062 125986646951744 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 18:27:04.316492 125986646951744 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 18:27:04.316586 125986646951744 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0422 18:27:04.328292 125986646951744 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 18:27:04.328346 125986646951744 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0422 18:27:11.593033 125986646951744 max_utils.py:791] Total memory size: 2.4 GB, Output size: 0.5 GB, Temp size: 1.9 GB, Argument size: 0.5 GB, Host temp size: 0.0 GB. I0422 18:27:11.593925 125986646951744 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0422 18:27:11.594092 125986646951744 metric_logger.py:301] number parameters: 1.104 billion I0422 18:27:19.307763 125986646951744 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 18:27:19.516747 125986646951744 checkpointing.py:776] Waited 0.20896387100219727 seconds for step 0 to finish before starting checkpointing. I0422 18:27:19.517558 125986646951744 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 18:27:19.517724 125986646951744 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0422 18:27:19.518143 125986646951744 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0. I0422 18:27:19.618335 125986646951744 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0422 18:27:19.697929 125948670248512 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0 I0422 18:27:19.732966 125986646951744 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0422 18:27:19.733087 125986646951744 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 18:27:20.362761 125986646951744 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.630264s I0422 18:27:20.363136 125986646951744 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 16.601 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 743 milliseconds) (per-host) I0422 18:27:20.363189 125986646951744 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.743477s (batch_requests_ready=0.108917s, total_serialization_initiated=0.634297s, others=0.000262s) I0422 18:27:20.363257 125986646951744 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.744448s (all_items=0.000021s, per_item={'items': '0.00002074'}, temp_paths=0.744427) I0422 18:27:20.364096 125953894254144 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0422 18:27:20.364200 125986646951744 async_checkpointer.py:561] Finished blocking save. Time taken: 0.846436s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0. I0422 18:27:20.364377 125986646951744 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 18:27:20.364563 125948659762752 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 18:27:20.364680 125986646951744 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776882439.517546, 'wait_for_prev_duration_secs': 5.4836273193359375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776882439.517746, 'checkpointer_blocking_duration_secs': 0.8465566635131836, 'get_old_steps_start_time': 1776882440.3643186, 'get_old_steps_duration_secs': 2.5272369384765625e-05, 'checkpoint_manager_blocking_start_time': 1776882439.5174644, 'checkpoint_manager_blocking_duration_secs': 0.8471887111663818} I0422 18:27:20.364801 125986646951744 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 18:27:20.364840 125986646951744 max_utils.py:750] Memstats: After params initialized: I0422 18:27:20.364912 125986646951744 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_0(process=0,(0,0,0,0)) I0422 18:27:20.364941 125986646951744 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_1(process=0,(1,0,0,0)) I0422 18:27:20.364966 125986646951744 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_2(process=0,(0,1,0,0)) I0422 18:27:20.364984 125986646951744 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_3(process=0,(1,1,0,0)) I0422 18:27:20.365005 125986646951744 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_4(process=0,(0,2,0,0)) I0422 18:27:20.365032 125986646951744 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_5(process=0,(1,2,0,0)) I0422 18:27:20.365066 125986646951744 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_6(process=0,(0,3,0,0)) I0422 18:27:20.365083 125986646951744 max_utils.py:756] Using (GB) 0.56 / 31.25 (1.792000%) on TPU_7(process=0,(1,3,0,0)) I0422 18:27:20.514462 125954299004480 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0/items I0422 18:27:20.732982 125986646951744 metric_logger.py:196] completed step: 0, seconds: 7.713, TFLOP/s/device: 1.762, Tokens/s/device: 265.521, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.840 I0422 18:27:20.733735 125986646951744 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/tensorboard/' I0422 18:27:20.854997 125949406348864 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776882440403448806, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776882440403448806, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0/_CHECKPOINT_METADATA I0422 18:27:20.929817 125986646951744 metric_logger.py:196] completed step: 1, seconds: 1.422, TFLOP/s/device: 9.552, Tokens/s/device: 1439.830, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.840 I0422 18:27:21.099485 2064413 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0422 18:27:21.099550 2064413 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0422 18:27:21.129318 125986646951744 metric_logger.py:196] completed step: 2, seconds: 0.008, TFLOP/s/device: 1757.714, Tokens/s/device: 264941.785, total_weights: 16384, loss: 9.775, lm_loss: 9.775, perplexity: 17589.299 I0422 18:27:21.327649 125986646951744 metric_logger.py:196] completed step: 3, seconds: 0.198, TFLOP/s/device: 68.542, Tokens/s/device: 10331.433, total_weights: 16384, loss: 8.838, lm_loss: 8.838, perplexity: 6892.454 I0422 18:27:38.257376 125954095580736 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_11_optimizer_offload_true/checkpoints/0/items/array_metadatas/process_0 I0422 18:27:38.257676 125986646951744 metric_logger.py:196] completed step: 4, seconds: 0.199, TFLOP/s/device: 68.247, Tokens/s/device: 10286.857, total_weights: 16384, loss: 7.965, lm_loss: 7.965, perplexity: 2879.114 I0422 18:27:38.285493 125986646951744 metric_logger.py:196] completed step: 5, seconds: 0.198, TFLOP/s/device: 68.491, Tokens/s/device: 10323.777, total_weights: 16384, loss: 7.233, lm_loss: 7.233, perplexity: 1383.691 I0422 18:27:38.486308 125986646951744 metric_logger.py:196] completed step: 6, seconds: 16.950, TFLOP/s/device: 0.802, Tokens/s/device: 120.825, total_weights: 16384, loss: 6.677, lm_loss: 6.677, perplexity: 793.605 I0422 18:27:38.690722 125986646951744 metric_logger.py:196] completed step: 7, seconds: 0.006, TFLOP/s/device: 2238.776, Tokens/s/device: 337452.628, total_weights: 16384, loss: 6.293, lm_loss: 6.293, perplexity: 540.987 I0422 18:27:38.892167 125986646951744 metric_logger.py:196] completed step: 8, seconds: 0.204, TFLOP/s/device: 66.456, Tokens/s/device: 10016.923, total_weights: 16384, loss: 6.055, lm_loss: 6.055, perplexity: 426.449 I0422 18:27:39.094505 125986646951744 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 18:27:39.096192 125986646951744 checkpointing.py:776] Waited 0.001764059066772461 seconds for step 9 to finish before starting checkpointing. I0422 18:27:39.096605 125986646951744 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 18:28:27.710416 125953904739904 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 66.617180s (commit=66.141587s, array_metadata_write=0.475593s) I0422 18:28:27.711487 125953894254144 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 185.585 MiB/s (total gbytes: 12.3 GiB) (time elapsed: a minute) (per-host) I0422 18:28:27.711542 125953894254144 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 67.347295s. I0422 18:28:28.004262 125953894254144 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776882440403448806, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0/_CHECKPOINT_METADATA I0422 18:28:28.235081 125953894254144 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 18:28:28.430808 125949406348864 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776882440403448806, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0/_CHECKPOINT_METADATA I0422 18:28:28.678120 125953894254144 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0422 18:28:28.678313 125953894254144 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.632534s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0/items I0422 18:28:28.679128 125953894254144 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0/items I0422 18:28:28.923983 125953894254144 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0 I0422 18:28:29.629219 125953894254144 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_11_optimizer_offload_true/checkpoints/0`. I0422 18:28:29.629925 125953894254144 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 70.112167s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/0 I0422 18:28:29.630000 125953894254144 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 69.265755s. I0422 18:28:29.630211 125948659762752 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 18:28:29.630340 125948659762752 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 18:28:29.630403 125948659762752 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 18:28:29.630451 125948659762752 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 18:28:29.630631 125986646951744 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 18:28:29.630890 125986646951744 checkpoint_manager.py:1441] Waiting for previous save to complete took 50.534218 seconds. If this number is high, consider checkpointing less frequently. I0422 18:28:29.631863 125986646951744 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0422 18:28:29.632226 125986646951744 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/9. I0422 18:28:29.802241 125953894254144 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/9 I0422 18:28:29.808534 125986646951744 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0422 18:28:29.808639 125986646951744 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 18:28:30.196326 125986646951744 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.388300s I0422 18:28:30.196792 125986646951744 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 25.754 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 479 milliseconds) (per-host) I0422 18:28:30.196843 125986646951744 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.479277s (batch_requests_ready=0.087580s, total_serialization_initiated=0.391345s, others=0.000352s) I0422 18:28:30.196899 125986646951744 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.479856s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.479841) I0422 18:28:30.197743 125949062415936 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0422 18:28:30.197854 125986646951744 async_checkpointer.py:561] Finished blocking save. Time taken: 0.565944s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/9. I0422 18:28:30.198040 125986646951744 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 18:28:30.198319 125948659762752 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 18:28:30.198456 125986646951744 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776882459.096546, 'wait_for_prev_duration_secs': 50.534217834472656, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776882509.6318896, 'checkpointer_blocking_duration_secs': 0.5660629272460938, 'get_old_steps_start_time': 1776882510.1979704, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776882459.0964782, 'checkpoint_manager_blocking_duration_secs': 51.101940870285034} I0422 18:28:30.198565 125986646951744 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 18:28:30.198599 125986646951744 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 18:28:30.551105 125954095580736 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/9/items I0422 18:28:31.760384 125949395863104 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_11_optimizer_offload_true/checkpoints/9/items/array_metadatas/process_0 I0422 18:29:01.050735 125949072901696 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 30.045703s (commit=29.594824s, array_metadata_write=0.450879s) I0422 18:29:01.059777 125949062415936 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 403.188 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 31 seconds) (per-host) I0422 18:29:01.059839 125949062415936 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 30.861940s. I0422 18:29:01.564074 125949062415936 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 18:29:01.960392 125949062415936 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0422 18:29:01.960573 125949062415936 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.574394s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/9/items I0422 18:29:01.961339 125949062415936 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/9/items I0422 18:29:02.198876 125949062415936 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/9 I0422 18:29:02.888124 125949062415936 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_11_optimizer_offload_true/checkpoints/9`. I0422 18:29:02.888880 125949062415936 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 33.256977s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260422_174509/linen_main_20260422_174509_11_optimizer_offload_true/checkpoints/9 I0422 18:29:02.888961 125949062415936 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 32.691064s. I0422 18:29:02.889124 125948659762752 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 18:29:02.889248 125948659762752 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 18:29:02.889300 125948659762752 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 18:29:02.889338 125948659762752 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 18:29:02.889492 125986646951744 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 18:29:02.889644 125986646951744 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 18:29:02.890592 125986646951744 metric_logger.py:196] completed step: 9, seconds: 0.204, TFLOP/s/device: 66.633, Tokens/s/device: 10043.647, total_weights: 16384, loss: 5.915, lm_loss: 5.915, perplexity: 370.686 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops