main (7f7822822) vs feat/nnx-trainstate-and-training-loop (8bb919bd5)| Metric | main 7f7822822 | feat/nnx-trainstate-and-training-loop 8bb919bd5 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 5.9150 | 5.9150 | 0 |
| TFLOP/s | 114.488 | 114.969 | +0.481 |
| Tok/s | 17256.9 | 17329.4 | +72.427 |
| Avg s/step | 2.430 | 2.409 | -0.021 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-17 12:55:41.356227: 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) I0417 12:55:41.472959 130953167060096 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0417 12:56:26.813307 130953167060096 max_utils.py:800] System Information: Jax Version: 0.9.2 I0417 12:56:26.813417 130953167060096 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0417 12:56:26.813453 130953167060096 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0417 12:56:26.813482 130953167060096 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system I0417 12:56:26.813503 130953167060096 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0417 12:56:26.813620 130953167060096 train.py:671] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0417 12:56:26.907287 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 12:56:27.308230 130953167060096 maxtext_utils.py:1548] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 12:56:27.308581 130953167060096 checkpointing.py:677] Setting up checkpoint logger... I0417 12:56:27.308634 130953167060096 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0417 12:56:27.308703 130953167060096 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0417 12:56:27.309370 130953167060096 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 0x7719521fbb00>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0417 12:56:29.699664 130953167060096 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0417 12:56:29.699861 130953167060096 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 0x7712d6339d30>}, handler_registry=None I0417 12:56:29.700193 130953167060096 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7712d6339d30>` 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`. I0417 12:56:29.700239 130953167060096 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7712d633a210>` 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`. I0417 12:56:29.700268 130953167060096 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 0x7712d6339d30>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7712d6339d30>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7712d633a210>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7712d633a210>}). I0417 12:56:29.700556 130953167060096 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0417 12:56:29.700611 130953167060096 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7712d632e0c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0417 12:56:29.819078 130953167060096 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints I0417 12:56:29.819311 130953167060096 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_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7712d6339820> I0417 12:56:29.819401 130953167060096 checkpointing.py:301] Checkpoint manager created! W0417 12:56:29.835746 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 12:56:30.367378 130953167060096 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0417 12:56:30.367476 130953167060096 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 12:56:30.449855 130953167060096 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 12:56:30.449932 130953167060096 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 12:56:30.463917 130953167060096 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 12:56:30.463964 130953167060096 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 12:56:30.484219 130953167060096 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 12:56:30.484273 130953167060096 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 12:56:30.498381 130953167060096 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 12:56:30.498428 130953167060096 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 12:56:30.512437 130953167060096 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 12:56:30.512483 130953167060096 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 12:56:30.533702 130953167060096 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0417 12:56:30.533759 130953167060096 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 12:56:30.551830 130953167060096 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0417 12:56:30.551882 130953167060096 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0417 12:56:30.740951 130953167060096 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0417 12:56:30.741103 130953167060096 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0417 12:56:30.789950 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [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 I0417 12:56:31.164805 130953167060096 maxtext_utils.py:1651] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0417 12:56:31.164916 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0417 12:56:31.164959 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0417 12:56:31.165004 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0417 12:56:31.165057 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0417 12:56:31.165089 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0417 12:56:31.165134 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0417 12:56:31.165173 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0417 12:56:31.165202 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0417 12:56:31.165228 130953167060096 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0417 12:56:31.165262 130953167060096 maxtext_utils.py:1651] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0417 12:56:31.165296 130953167060096 maxtext_utils.py:1651] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0417 12:56:31.582931 130953167060096 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0417 12:56:31.583015 130953167060096 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0417 12:56:31.596464 130953167060096 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0417 12:56:31.596513 130953167060096 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0417 12:56:31.848156 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 12:56:31.961794 130953167060096 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. I0417 12:56:31.962237 130953167060096 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0417 12:56:31.962373 130953167060096 metric_logger.py:301] number parameters: 1.104 billion W0417 12:56:32.781907 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 12:56:32.877689 130953167060096 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0417 12:56:32.994541 130953167060096 checkpointing.py:776] Waited 0.11684489250183105 seconds for step 0 to finish before starting checkpointing. I0417 12:56:32.994965 130953167060096 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 12:56:32.995150 130953167060096 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0417 12:56:32.995510 130953167060096 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0. I0417 12:56:33.072982 130953167060096 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0417 12:56:33.160307 130953167060096 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0417 12:56:33.160392 130953167060096 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 I0417 12:56:33.168601 130845675357760 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0 W0417 12:56:33.373372 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.384381 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.395992 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.403021 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.414982 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.422795 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.427821 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.432321 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 12:56:33.787348 130953167060096 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.627480s I0417 12:56:33.787865 130953167060096 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 17.286 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 713 milliseconds) (per-host) I0417 12:56:33.787927 130953167060096 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.714025s (batch_requests_ready=0.082681s, total_serialization_initiated=0.630985s, others=0.000359s) I0417 12:56:33.787997 130953167060096 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.714561s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.714543) I0417 12:56:33.788907 130845612443200 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0417 12:56:33.789023 130953167060096 async_checkpointer.py:561] Finished blocking save. Time taken: 0.793825s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0. I0417 12:56:33.789281 130953167060096 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0417 12:56:33.789560 130953167060096 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776430592.9949534, 'wait_for_prev_duration_secs': 4.100799560546875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776430592.995174, 'checkpointer_blocking_duration_secs': 0.7940042018890381, 'get_old_steps_start_time': 1776430593.7892053, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776430592.9948719, 'checkpoint_manager_blocking_duration_secs': 0.7946560382843018} I0417 12:56:33.789650 130953167060096 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0417 12:56:33.789698 130953167060096 max_utils.py:750] Memstats: After params initialized: I0417 12:56:33.789786 130953167060096 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0417 12:56:33.789815 130953167060096 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0417 12:56:33.789836 130953167060096 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0417 12:56:33.789855 130953167060096 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0417 12:56:33.789873 130953167060096 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0417 12:56:33.789890 130953167060096 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0417 12:56:33.789908 130953167060096 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0417 12:56:33.789925 130953167060096 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0417 12:56:33.795808 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.801455 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 12:56:33.806614 410140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 12:56:33.828307 130845664872000 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0/items I0417 12:56:34.003629 130845633414720 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776430593739877524, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776430593739877524, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0417 12:56:34.003865 130845685843520 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 12:56:34.153923 130953167060096 metric_logger.py:196] completed step: 0, seconds: 0.915, TFLOP/s/device: 14.852, Tokens/s/device: 2238.621, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0417 12:56:34.154786 130953167060096 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/tensorboard/' I0417 12:56:34.270287 130953167060096 metric_logger.py:196] completed step: 1, seconds: 1.274, TFLOP/s/device: 10.663, Tokens/s/device: 1607.192, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0417 12:56:34.388779 130953167060096 metric_logger.py:196] completed step: 2, seconds: 0.007, TFLOP/s/device: 2079.132, Tokens/s/device: 313389.441, total_weights: 16384, loss: 9.775, lm_loss: 9.775, perplexity: 17589.287 I0417 12:56:34.507242 130953167060096 metric_logger.py:196] completed step: 3, seconds: 0.117, TFLOP/s/device: 115.746, Tokens/s/device: 17446.566, total_weights: 16384, loss: 8.838, lm_loss: 8.838, perplexity: 6892.933 I0417 12:56:34.519318 412994 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0417 12:56:34.519378 412994 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0417 12:56:35.473221 130845675357760 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0/items/array_metadatas/process_0 I0417 12:56:54.498537 130953167060096 metric_logger.py:196] completed step: 4, seconds: 0.118, TFLOP/s/device: 114.861, Tokens/s/device: 17313.090, total_weights: 16384, loss: 7.965, lm_loss: 7.965, perplexity: 2878.644 I0417 12:56:54.506448 130953167060096 metric_logger.py:196] completed step: 5, seconds: 0.119, TFLOP/s/device: 113.777, Tokens/s/device: 17149.700, total_weights: 16384, loss: 7.233, lm_loss: 7.233, perplexity: 1383.750 I0417 12:56:54.623167 130953167060096 metric_logger.py:196] completed step: 6, seconds: 19.992, TFLOP/s/device: 0.680, Tokens/s/device: 102.443, total_weights: 16384, loss: 6.676, lm_loss: 6.676, perplexity: 793.531 I0417 12:56:54.741463 130953167060096 metric_logger.py:196] completed step: 7, seconds: 0.006, TFLOP/s/device: 2252.508, Tokens/s/device: 339522.546, total_weights: 16384, loss: 6.293, lm_loss: 6.293, perplexity: 540.860 I0417 12:56:54.859556 130953167060096 metric_logger.py:196] completed step: 8, seconds: 0.116, TFLOP/s/device: 116.833, Tokens/s/device: 17610.387, total_weights: 16384, loss: 6.055, lm_loss: 6.055, perplexity: 426.371 I0417 12:56:54.976671 130953167060096 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0417 12:56:54.977453 130953167060096 checkpointing.py:776] Waited 0.0007998943328857422 seconds for step 9 to finish before starting checkpointing. I0417 12:56:54.977729 130953167060096 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 12:57:05.822555 130845622928960 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 31.308750s (commit=30.781250s, array_metadata_write=0.527500s) I0417 12:57:05.823698 130845612443200 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 385.859 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 32 seconds) (per-host) I0417 12:57:05.823805 130845612443200 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.034712s. I0417 12:57:06.051628 130845612443200 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776430593739877524, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0417 12:57:06.276456 130845612443200 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0417 12:57:06.482634 130845633414720 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776430593739877524, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0417 12:57:06.673196 130845612443200 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0417 12:57:06.673358 130845612443200 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.563232s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0/items I0417 12:57:06.674005 130845612443200 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0/items I0417 12:57:06.914263 130845612443200 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0 I0417 12:57:07.578248 130845612443200 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0`. I0417 12:57:07.578890 130845612443200 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 34.583704s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/0 I0417 12:57:07.578963 130845612443200 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 33.789872s. I0417 12:57:07.579127 130845685843520 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 12:57:07.579230 130845685843520 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 12:57:07.579286 130845685843520 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0417 12:57:07.579332 130845685843520 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0417 12:57:07.579441 130953167060096 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. W0417 12:57:07.579555 130953167060096 checkpoint_manager.py:1441] Waiting for previous save to complete took 12.601835 seconds. If this number is high, consider checkpointing less frequently. I0417 12:57:07.580349 130953167060096 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0417 12:57:07.580627 130953167060096 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9. I0417 12:57:07.748890 130845685843520 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9 I0417 12:57:07.754767 130953167060096 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0417 12:57:07.754866 130953167060096 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 I0417 12:57:08.447959 130845664872000 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9/items I0417 12:57:13.236340 130953167060096 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 5.481979s I0417 12:57:13.241039 130953167060096 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.211 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 5 seconds) (per-host) I0417 12:57:13.241127 130953167060096 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 5.580446s (batch_requests_ready=0.090709s, total_serialization_initiated=5.485183s, others=0.004554s) I0417 12:57:13.241220 130953167060096 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 5.580990s (all_items=0.000019s, per_item={'items': '0.00001860'}, temp_paths=5.580972) I0417 12:57:13.242926 130845580985920 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0417 12:57:13.243038 130953167060096 async_checkpointer.py:561] Finished blocking save. Time taken: 5.662637s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9. I0417 12:57:13.243430 130953167060096 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0417 12:57:13.243753 130845612443200 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 12:57:13.243879 130953167060096 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776430614.9776924, 'wait_for_prev_duration_secs': 12.601835489273071, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776430627.5803747, 'checkpointer_blocking_duration_secs': 5.662820100784302, 'get_old_steps_start_time': 1776430633.2432199, 'get_old_steps_duration_secs': 0.0001232624053955078, 'checkpoint_manager_blocking_start_time': 1776430614.9776485, 'checkpoint_manager_blocking_duration_secs': 18.266196250915527} I0417 12:57:13.244015 130953167060096 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0417 12:57:13.244069 130953167060096 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 12:57:13.960224 130845675357760 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9/items/array_metadatas/process_0 I0417 12:57:53.051772 130845664872000 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 39.809793s (commit=39.368221s, array_metadata_write=0.441572s) I0417 12:57:53.052802 130845580985920 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 278.392 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 45 seconds) (per-host) I0417 12:57:53.052912 130845580985920 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.809790s. I0417 12:57:53.459896 130845580985920 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0417 12:57:53.928848 130845580985920 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0417 12:57:53.929031 130845580985920 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.610138s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9/items I0417 12:57:53.929660 130845580985920 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9/items I0417 12:57:54.212209 130845580985920 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9 I0417 12:57:54.922832 130845580985920 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9`. I0417 12:57:54.923470 130845580985920 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 47.343078s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_02_synthetic/checkpoints/9 I0417 12:57:54.923551 130845580985920 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 41.680438s. I0417 12:57:54.923738 130845612443200 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 12:57:54.923850 130845612443200 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 12:57:54.923938 130845612443200 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0417 12:57:54.923987 130845612443200 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0417 12:57:54.924155 130953167060096 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. I0417 12:57:54.924287 130953167060096 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 12:57:54.924973 130953167060096 metric_logger.py:196] completed step: 9, seconds: 0.119, TFLOP/s/device: 114.488, Tokens/s/device: 17256.924, 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
8bb919bd5
· feat_nnx_trainstate_and_training_loop_20260420_142345
· full log
2026-04-20 14:26:35.393368: 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) I0420 14:26:35.510761 123899151375488 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0420 14:27:19.500394 123899151375488 max_utils.py:800] System Information: Jax Version: 0.9.2 I0420 14:27:19.500528 123899151375488 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0420 14:27:19.500570 123899151375488 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0420 14:27:19.500596 123899151375488 train_utils.py:365] WARNING: 'dataset_path' might be pointing your local file system I0420 14:27:19.500619 123899151375488 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0420 14:27:19.500703 123899151375488 train.py:806] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0420 14:27:20.025189 123899151375488 maxtext_utils.py:1718] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 14:27:20.025365 123899151375488 maxtext_utils.py:1718] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 14:27:20.025612 123899151375488 checkpointing.py:688] Setting up checkpoint logger... I0420 14:27:20.025652 123899151375488 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0420 14:27:20.025729 123899151375488 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0420 14:27:20.026409 123899151375488 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 0x70aeedf358e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0420 14:27:22.395537 123899151375488 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0420 14:27:22.395792 123899151375488 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 0x70a870560440>}, handler_registry=None I0420 14:27:22.396181 123899151375488 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x70a870560440>` 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`. I0420 14:27:22.396225 123899151375488 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x70a86f904560>` 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`. I0420 14:27:22.396255 123899151375488 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 0x70a870560440>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x70a870560440>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x70a86f904560>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x70a86f904560>}). I0420 14:27:22.396554 123899151375488 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0420 14:27:22.396613 123899151375488 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x70a86f54efc0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0420 14:27:22.524000 123899151375488 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints I0420 14:27:22.524297 123899151375488 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_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x70a86f5368d0> I0420 14:27:22.524403 123899151375488 checkpointing.py:302] Checkpoint manager created! I0420 14:27:23.274884 123899151375488 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0420 14:27:23.274988 123899151375488 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0420 14:27:23.359119 123899151375488 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 14:27:23.359206 123899151375488 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0420 14:27:23.373529 123899151375488 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 14:27:23.373578 123899151375488 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0420 14:27:23.395062 123899151375488 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 14:27:23.395127 123899151375488 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0420 14:27:23.409643 123899151375488 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 14:27:23.409706 123899151375488 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0420 14:27:23.424165 123899151375488 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 14:27:23.424220 123899151375488 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0420 14:27:23.445995 123899151375488 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0420 14:27:23.446069 123899151375488 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0420 14:27:23.464482 123899151375488 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0420 14:27:23.464540 123899151375488 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0420 14:27:23.657853 123899151375488 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0420 14:27:23.657953 123899151375488 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0420 14:27:23.707727 1035834 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [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 I0420 14:27:24.091136 123899151375488 maxtext_utils.py:1821] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0420 14:27:24.091249 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0420 14:27:24.091290 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0420 14:27:24.091333 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0420 14:27:24.091372 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0420 14:27:24.091400 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0420 14:27:24.091439 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0420 14:27:24.091478 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0420 14:27:24.091506 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0420 14:27:24.091532 123899151375488 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0420 14:27:24.091567 123899151375488 maxtext_utils.py:1821] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0420 14:27:24.091600 123899151375488 maxtext_utils.py:1821] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0420 14:27:24.519724 123899151375488 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0420 14:27:24.519821 123899151375488 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0420 14:27:24.533517 123899151375488 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0420 14:27:24.533569 123899151375488 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0420 14:27:24.807665 1035834 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0420 14:27:24.930254 123899151375488 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. I0420 14:27:24.930813 123899151375488 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0420 14:27:24.930957 123899151375488 metric_logger.py:301] number parameters: 1.104 billion W0420 14:27:25.815650 1035834 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0420 14:27:25.911623 123899151375488 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0420 14:27:26.029013 123899151375488 checkpointing.py:798] Waited 0.11737203598022461 seconds for step 0 to finish before starting checkpointing. I0420 14:27:26.029496 123899151375488 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 14:27:26.029645 123899151375488 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0420 14:27:26.030056 123899151375488 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0. I0420 14:27:26.114208 123899151375488 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0420 14:27:26.201406 123899151375488 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0420 14:27:26.201499 123899151375488 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 I0420 14:27:26.207861 123791600518720 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0 I0420 14:27:26.940401 123791590032960 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0/items I0420 14:27:27.077471 123899151375488 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.876539s I0420 14:27:27.078019 123899151375488 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 12.824 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 962 milliseconds) (per-host) I0420 14:27:27.078092 123899151375488 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.962412s (batch_requests_ready=0.081850s, total_serialization_initiated=0.880150s, others=0.000412s) I0420 14:27:27.078163 123899151375488 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.963372s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=0.963353) I0420 14:27:27.079102 123791529215552 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0420 14:27:27.079232 123899151375488 async_checkpointer.py:561] Finished blocking save. Time taken: 1.049544s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0. I0420 14:27:27.079481 123899151375488 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0420 14:27:27.079687 123899151375488 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776695246.0294833, 'wait_for_prev_duration_secs': 4.00543212890625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776695246.029664, 'checkpointer_blocking_duration_secs': 1.0497121810913086, 'get_old_steps_start_time': 1776695247.0794053, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1776695246.0293994, 'checkpoint_manager_blocking_duration_secs': 1.0502569675445557} I0420 14:27:27.079773 123899151375488 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0420 14:27:27.079821 123899151375488 max_utils.py:750] Memstats: After params initialized: I0420 14:27:27.079918 123899151375488 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0420 14:27:27.079952 123899151375488 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0420 14:27:27.079973 123899151375488 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0420 14:27:27.079992 123899151375488 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0420 14:27:27.080011 123899151375488 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0420 14:27:27.080029 123899151375488 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0420 14:27:27.080058 123899151375488 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0420 14:27:27.080076 123899151375488 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) I0420 14:27:27.108622 123791550187072 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776695246846311626, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776695246846311626, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0420 14:27:27.108896 123791611004480 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0420 14:27:27.522447 1038178 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0420 14:27:27.522516 1038178 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0420 14:27:27.549096 123899151375488 metric_logger.py:196] completed step: 0, seconds: 0.980, TFLOP/s/device: 13.862, Tokens/s/device: 2089.414, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0420 14:27:27.550075 123899151375488 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/tensorboard/' I0420 14:27:27.648617 123899151375488 metric_logger.py:196] completed step: 1, seconds: 1.619, TFLOP/s/device: 8.394, Tokens/s/device: 1265.268, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0420 14:27:27.768205 123899151375488 metric_logger.py:196] completed step: 2, seconds: 0.024, TFLOP/s/device: 568.928, Tokens/s/device: 85754.962, total_weights: 16384, loss: 9.775, lm_loss: 9.775, perplexity: 17589.287 I0420 14:27:27.886689 123899151375488 metric_logger.py:196] completed step: 3, seconds: 0.100, TFLOP/s/device: 135.530, Tokens/s/device: 20428.520, total_weights: 16384, loss: 8.838, lm_loss: 8.838, perplexity: 6892.933 I0420 14:27:28.249502 123791600518720 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0/items/array_metadatas/process_0 I0420 14:27:47.347945 123899151375488 metric_logger.py:196] completed step: 4, seconds: 0.120, TFLOP/s/device: 113.668, Tokens/s/device: 17133.344, total_weights: 16384, loss: 7.965, lm_loss: 7.965, perplexity: 2878.644 I0420 14:27:47.356188 123899151375488 metric_logger.py:196] completed step: 5, seconds: 0.118, TFLOP/s/device: 114.736, Tokens/s/device: 17294.230, total_weights: 16384, loss: 7.233, lm_loss: 7.233, perplexity: 1383.750 I0420 14:27:47.473786 123899151375488 metric_logger.py:196] completed step: 6, seconds: 19.463, TFLOP/s/device: 0.698, Tokens/s/device: 105.228, total_weights: 16384, loss: 6.676, lm_loss: 6.676, perplexity: 793.531 I0420 14:27:47.591869 123899151375488 metric_logger.py:196] completed step: 7, seconds: 0.005, TFLOP/s/device: 2608.395, Tokens/s/device: 393165.675, total_weights: 16384, loss: 6.293, lm_loss: 6.293, perplexity: 540.860 I0420 14:27:47.710408 123899151375488 metric_logger.py:196] completed step: 8, seconds: 0.118, TFLOP/s/device: 114.777, Tokens/s/device: 17300.366, total_weights: 16384, loss: 6.055, lm_loss: 6.055, perplexity: 426.371 I0420 14:27:47.827736 123899151375488 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0420 14:27:47.828680 123899151375488 checkpointing.py:798] Waited 0.0009813308715820312 seconds for step 9 to finish before starting checkpointing. I0420 14:27:47.828971 123899151375488 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 14:27:50.893736 123791579547200 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 23.375698s (commit=22.961807s, array_metadata_write=0.413891s) I0420 14:27:50.894954 123791529215552 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 509.975 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 24 seconds) (per-host) I0420 14:27:50.895084 123791529215552 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 23.815794s. I0420 14:27:51.126703 123791529215552 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776695246846311626, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0420 14:27:51.321537 123791529215552 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0420 14:27:51.514357 123791550187072 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776695246846311626, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA I0420 14:27:51.729783 123791529215552 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0420 14:27:51.730119 123791529215552 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.555397s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0/items I0420 14:27:51.730838 123791529215552 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0/items I0420 14:27:51.967906 123791529215552 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0 I0420 14:27:52.652213 123791529215552 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0`. I0420 14:27:52.652964 123791529215552 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 26.623287s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/0 I0420 14:27:52.653040 123791529215552 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 25.573752s. I0420 14:27:52.653183 123791611004480 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 14:27:52.653233 123791611004480 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0420 14:27:52.653283 123791611004480 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0420 14:27:52.653324 123791611004480 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0420 14:27:52.653450 123899151375488 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. W0420 14:27:52.653596 123899151375488 checkpoint_manager.py:1441] Waiting for previous save to complete took 4.824627 seconds. If this number is high, consider checkpointing less frequently. I0420 14:27:52.654407 123899151375488 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0420 14:27:52.654704 123899151375488 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9. I0420 14:27:52.824331 123791611004480 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9 I0420 14:27:52.828783 123899151375488 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0420 14:27:52.828891 123899151375488 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 I0420 14:27:53.506857 123791397094976 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9/items I0420 14:27:57.835964 123899151375488 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 5.007651s I0420 14:27:57.840526 123899151375488 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.420 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 5 seconds) (per-host) I0420 14:27:57.840597 123899151375488 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 5.099656s (batch_requests_ready=0.084326s, total_serialization_initiated=5.010905s, others=0.004425s) I0420 14:27:57.840675 123899151375488 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 5.100231s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=5.100217) I0420 14:27:57.842169 123791579547200 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0420 14:27:57.842257 123899151375488 async_checkpointer.py:561] Finished blocking save. Time taken: 5.187805s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9. I0420 14:27:57.842453 123899151375488 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0420 14:27:57.842689 123791478883904 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0420 14:27:57.842821 123899151375488 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776695267.828938, 'wait_for_prev_duration_secs': 4.824627161026001, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776695272.65443, 'checkpointer_blocking_duration_secs': 5.18793249130249, 'get_old_steps_start_time': 1776695277.8423803, 'get_old_steps_duration_secs': 3.24249267578125e-05, 'checkpoint_manager_blocking_start_time': 1776695267.82889, 'checkpoint_manager_blocking_duration_secs': 10.013904809951782} I0420 14:27:57.842945 123899151375488 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0420 14:27:57.842981 123899151375488 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 14:27:58.558249 123791600518720 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9/items/array_metadatas/process_0 I0420 14:28:30.146999 123791397094976 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.305822s (commit=31.859357s, array_metadata_write=0.446465s) I0420 14:28:30.148186 123791579547200 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 337.817 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 37 seconds) (per-host) I0420 14:28:30.148268 123791579547200 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.305959s. I0420 14:28:30.566613 123791579547200 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0420 14:28:31.013205 123791579547200 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0420 14:28:31.013885 123791579547200 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.601956s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9/items I0420 14:28:31.014639 123791579547200 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9/items I0420 14:28:31.264717 123791579547200 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9 I0420 14:28:31.939778 123791579547200 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9`. I0420 14:28:31.940559 123791579547200 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 39.286113s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_02_synthetic/checkpoints/9 I0420 14:28:31.940637 123791579547200 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.098335s. I0420 14:28:31.940860 123791478883904 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 14:28:31.940990 123791478883904 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0420 14:28:31.941089 123791478883904 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0420 14:28:31.941139 123791478883904 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0420 14:28:31.941330 123899151375488 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. I0420 14:28:31.941479 123899151375488 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 14:28:31.942386 123899151375488 metric_logger.py:196] completed step: 9, seconds: 0.118, TFLOP/s/device: 114.969, Tokens/s/device: 17329.351, 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