main (51c7f2b8e) vs feat/nnx-trainstate-and-training-loop (a2a803184)| Metric | main 51c7f2b8e | feat/nnx-trainstate-and-training-loop a2a803184 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1750 | 8.1750 | 0 |
| TFLOP/s | 128.416 | 125.098 | -3.318 |
| Tok/s | 19356.3 | 18856.1 | -500.204 |
| Avg s/step | 4.523 | 4.410 | -0.113 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-15 21:23:47.601893: 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) I0415 21:23:47.718806 137006241836160 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 21:24:33.745581 137006241836160 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 21:24:33.745698 137006241836160 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 21:24:33.745733 137006241836160 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 I0415 21:24:33.745761 137006241836160 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 21:24:33.745784 137006241836160 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0415 21:24:33.745858 137006241836160 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 21:24:33.836388 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:24:34.235151 137006241836160 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0415 21:24:34.235420 137006241836160 checkpointing.py:677] Setting up checkpoint logger... I0415 21:24:34.235466 137006241836160 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 21:24:34.235506 137006241836160 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 21:24:34.235965 137006241836160 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 0x7c9aa9700950>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 21:24:36.605519 137006241836160 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 21:24:36.605703 137006241836160 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 0x7c942c13dfa0>}, handler_registry=None I0415 21:24:36.605956 137006241836160 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c942c13dfa0>` 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`. I0415 21:24:36.605997 137006241836160 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c942c13cb60>` 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`. I0415 21:24:36.606027 137006241836160 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 0x7c942c13dfa0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c942c13dfa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c942c13cb60>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c942c13cb60>}). I0415 21:24:36.606327 137006241836160 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 21:24:36.606379 137006241836160 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7c942c1313a0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0415 21:24:36.733365 137006241836160 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints I0415 21:24:36.733587 137006241836160 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_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c942c5782f0> I0415 21:24:36.733672 137006241836160 checkpointing.py:301] Checkpoint manager created! I0415 21:24:37.587486 137006241836160 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 21:24:37.587588 137006241836160 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:24:37.677780 137006241836160 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:24:37.677857 137006241836160 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:24:37.692060 137006241836160 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:24:37.692108 137006241836160 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:24:37.712496 137006241836160 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:24:37.712552 137006241836160 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:24:37.726788 137006241836160 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:24:37.726837 137006241836160 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:24:37.740969 137006241836160 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:24:37.741019 137006241836160 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:24:37.762494 137006241836160 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 21:24:37.762550 137006241836160 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:24:37.780781 137006241836160 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 21:24:37.780832 137006241836160 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0415 21:24:37.971321 137006241836160 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 21:24:37.971537 137006241836160 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0415 21:24:38.023709 3239479 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 I0415 21:24:38.416252 137006241836160 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0415 21:24:38.416363 137006241836160 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0415 21:24:38.416403 137006241836160 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0415 21:24:38.416446 137006241836160 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0415 21:24:38.416486 137006241836160 maxtext_utils.py:1620] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0415 21:24:38.416512 137006241836160 maxtext_utils.py:1620] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0415 21:24:38.416550 137006241836160 maxtext_utils.py:1620] 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) I0415 21:24:38.416587 137006241836160 maxtext_utils.py:1620] 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') I0415 21:24:38.416615 137006241836160 maxtext_utils.py:1620] 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) I0415 21:24:38.416647 137006241836160 maxtext_utils.py:1620] 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) I0415 21:24:38.416682 137006241836160 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0415 21:24:38.416714 137006241836160 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0415 21:24:38.434952 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0415 21:24:38.435004 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0415 21:24:38.448259 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0415 21:24:38.448304 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0415 21:24:38.474001 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0415 21:24:38.474064 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0415 21:24:38.486953 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0415 21:24:38.486999 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0415 21:24:38.512638 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0415 21:24:38.512691 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0415 21:24:38.525598 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0415 21:24:38.525650 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0415 21:24:38.564270 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0415 21:24:38.564330 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0415 21:24:38.577224 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0415 21:24:38.577270 137006241836160 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0415 21:24:39.159214 137006241836160 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:24:39.159297 137006241836160 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0415 21:24:39.172790 137006241836160 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:24:39.172838 137006241836160 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0415 21:24:40.026971 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:24:40.165036 137006241836160 max_utils.py:791] Total memory size: 4.5 GB, Output size: 1.5 GB, Temp size: 2.9 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0415 21:24:40.165559 137006241836160 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 21:24:40.165722 137006241836160 metric_logger.py:289] number parameters: 1.104 billion W0415 21:24:41.519308 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:24:41.623704 137006241836160 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0415 21:24:42.056651 137006241836160 checkpointing.py:776] Waited 0.43293309211730957 seconds for step 0 to finish before starting checkpointing. I0415 21:24:42.057229 137006241836160 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:24:42.057415 137006241836160 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0415 21:24:42.057830 137006241836160 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0. I0415 21:24:42.144493 137006241836160 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0415 21:24:42.234752 136898748089920 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0 I0415 21:24:42.240766 137006241836160 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:24:42.240861 137006241836160 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 W0415 21:24:42.465916 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:42.477519 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:42.492491 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:42.511515 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:42.522285 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:42.528868 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:42.534428 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:42.539392 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:24:42.942122 136898737604160 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0/items I0415 21:24:43.078327 136898706146880 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776288282851810969, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776288282851810969, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0415 21:24:43.081861 137006241836160 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.841517s I0415 21:24:43.082258 137006241836160 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 13.178 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 936 milliseconds) (per-host) I0415 21:24:43.082317 137006241836160 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.936554s (batch_requests_ready=0.090828s, total_serialization_initiated=0.845416s, others=0.000311s) I0415 21:24:43.082374 137006241836160 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.937403s (all_items=0.000032s, per_item={'items': '0.00003171'}, temp_paths=0.937371) I0415 21:24:43.083137 136898685175360 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:24:43.083234 137006241836160 async_checkpointer.py:561] Finished blocking save. Time taken: 1.025777s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0. I0415 21:24:43.083431 137006241836160 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:24:43.083609 136898758575680 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:24:43.083710 137006241836160 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776288282.057212, 'wait_for_prev_duration_secs': 4.458427429199219e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776288282.057437, 'checkpointer_blocking_duration_secs': 1.0259056091308594, 'get_old_steps_start_time': 1776288283.083363, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776288282.0570784, 'checkpoint_manager_blocking_duration_secs': 1.0266053676605225} I0415 21:24:43.083825 137006241836160 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0415 21:24:43.083866 137006241836160 max_utils.py:750] Memstats: After params initialized: I0415 21:24:43.083938 137006241836160 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0415 21:24:43.083964 137006241836160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0415 21:24:43.083983 137006241836160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0415 21:24:43.084001 137006241836160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0415 21:24:43.084017 137006241836160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0415 21:24:43.084033 137006241836160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0415 21:24:43.084068 137006241836160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0415 21:24:43.084085 137006241836160 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0415 21:24:43.089702 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:43.094902 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:24:43.099600 3239479 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:24:43.400802 137006241836160 metric_logger.py:185] completed step: 0, seconds: 1.457, TFLOP/s/device: 37.291, Tokens/s/device: 5620.930, total_weights: 65536, loss: 10.871 I0415 21:24:43.401780 137006241836160 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/tensorboard/' I0415 21:24:43.426319 3241713 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 21:24:43.426374 3241713 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 21:24:43.816971 137006241836160 metric_logger.py:185] completed step: 1, seconds: 1.775, TFLOP/s/device: 30.614, Tokens/s/device: 4614.533, total_weights: 65536, loss: 10.871 I0415 21:24:44.190361 136898748089920 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0/items/array_metadatas/process_0 I0415 21:24:44.240828 137006241836160 metric_logger.py:185] completed step: 2, seconds: 0.006, TFLOP/s/device: 8495.938, Tokens/s/device: 1280600.281, total_weights: 65536, loss: 10.261 I0415 21:24:44.659225 137006241836160 metric_logger.py:185] completed step: 3, seconds: 0.417, TFLOP/s/device: 130.315, Tokens/s/device: 19642.446, total_weights: 65536, loss: 9.732 I0415 21:25:15.978759 136898695661120 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.556793s (commit=32.071650s, array_metadata_write=0.485143s) I0415 21:25:15.980066 136898685175360 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 373.491 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 33 seconds) (per-host) I0415 21:25:15.980181 136898685175360 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.896896s. I0415 21:25:16.241750 136898685175360 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776288282851810969, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0415 21:25:16.430104 136898685175360 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 21:25:16.644997 136898706146880 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776288282851810969, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0415 21:25:16.886962 136898685175360 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:25:16.887533 136898685175360 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.598157s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0/items I0415 21:25:16.888183 136898685175360 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0/items I0415 21:25:17.135339 136898685175360 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0 I0415 21:25:17.851951 136898685175360 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0`. I0415 21:25:17.852603 136898685175360 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 35.795144s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/0 I0415 21:25:17.852688 136898685175360 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.769409s. I0415 21:25:17.852900 136898758575680 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:25:17.853024 136898758575680 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:25:17.853137 136898758575680 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:25:17.853193 136898758575680 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0415 21:25:21.471101 137006241836160 metric_logger.py:185] completed step: 4, seconds: 0.424, TFLOP/s/device: 128.214, Tokens/s/device: 19325.860, total_weights: 65536, loss: 9.274 I0415 21:25:21.478781 137006241836160 metric_logger.py:185] completed step: 5, seconds: 0.422, TFLOP/s/device: 128.935, Tokens/s/device: 19434.520, total_weights: 65536, loss: 8.888 I0415 21:25:21.900774 137006241836160 metric_logger.py:185] completed step: 6, seconds: 36.810, TFLOP/s/device: 1.476, Tokens/s/device: 222.549, total_weights: 65536, loss: 8.588 I0415 21:25:22.324299 137006241836160 metric_logger.py:185] completed step: 7, seconds: 0.005, TFLOP/s/device: 11327.327, Tokens/s/device: 1707378.074, total_weights: 65536, loss: 8.380 I0415 21:25:22.758679 137006241836160 metric_logger.py:185] completed step: 8, seconds: 0.423, TFLOP/s/device: 128.531, Tokens/s/device: 19373.575, total_weights: 65536, loss: 8.251 I0415 21:25:23.192172 137006241836160 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0415 21:25:23.192992 137006241836160 checkpointing.py:776] Waited 0.0008478164672851562 seconds for step 9 to finish before starting checkpointing. I0415 21:25:23.193277 137006241836160 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:25:23.193453 137006241836160 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0415 21:25:23.193759 137006241836160 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9. I0415 21:25:23.364004 137006241836160 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:25:23.364110 137006241836160 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 I0415 21:25:23.414486 136898758575680 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9 I0415 21:25:24.136485 136898574026304 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9/items I0415 21:25:34.818437 137006241836160 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 11.454832s I0415 21:25:34.822981 137006241836160 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.069 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 11 seconds) (per-host) I0415 21:25:34.823062 137006241836160 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 11.547532s (batch_requests_ready=0.085098s, total_serialization_initiated=11.458003s, others=0.004431s) I0415 21:25:34.823165 137006241836160 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 11.548115s (all_items=0.000024s, per_item={'items': '0.00002408'}, temp_paths=11.548090) I0415 21:25:34.824377 136898695661120 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:25:34.824506 137006241836160 async_checkpointer.py:561] Finished blocking save. Time taken: 11.631009s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9. I0415 21:25:34.824701 137006241836160 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:25:34.824866 136898685175360 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:25:34.824986 137006241836160 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776288323.1932337, 'wait_for_prev_duration_secs': 7.033348083496094e-05, 'time_between_consecutive_saves_sec': 5.340025186538696, 'checkpointer_blocking_start_time': 1776288323.1934745, 'checkpointer_blocking_duration_secs': 11.631113290786743, 'get_old_steps_start_time': 1776288334.8246064, 'get_old_steps_duration_secs': 4.9114227294921875e-05, 'checkpoint_manager_blocking_start_time': 1776288323.1931927, 'checkpoint_manager_blocking_duration_secs': 11.631767272949219} I0415 21:25:34.825137 137006241836160 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0415 21:25:34.825175 137006241836160 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:25:35.611492 136897017939520 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9/items/array_metadatas/process_0 I0415 21:26:07.064487 136898748089920 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.240884s (commit=31.825371s, array_metadata_write=0.415513s) I0415 21:26:07.065494 136898695661120 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 288.577 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 43 seconds) (per-host) I0415 21:26:07.065593 136898695661120 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.241123s. I0415 21:26:07.516564 136898695661120 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 21:26:07.945553 136898695661120 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:26:07.945728 136898695661120 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.586740s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9/items I0415 21:26:07.946379 136898695661120 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9/items I0415 21:26:08.174887 136898695661120 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9 I0415 21:26:08.876879 136898695661120 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9`. I0415 21:26:08.877508 136898695661120 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 45.684017s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_06_grad_accum/checkpoints/9 I0415 21:26:08.877576 136898695661120 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.053108s. I0415 21:26:08.877761 136898685175360 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:26:08.877869 136898685175360 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:26:08.877925 136898685175360 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:26:08.877973 136898685175360 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0415 21:26:08.878228 137006241836160 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. I0415 21:26:08.878356 137006241836160 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:26:08.878961 137006241836160 metric_logger.py:185] completed step: 9, seconds: 0.423, TFLOP/s/device: 128.416, Tokens/s/device: 19356.272, total_weights: 65536, loss: 8.175 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops
a2a803184
· feat_nnx_trainstate_and_training_loop_20260416_004836
· full log
2026-04-16 01:01:32.182735: 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) I0416 01:01:32.301238 130828600429696 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 01:02:03.385430 130828600429696 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 01:02:03.385549 130828600429696 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 01:02:03.385584 130828600429696 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 I0416 01:02:03.385608 130828600429696 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 01:02:03.385633 130828600429696 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 01:02:03.385713 130828600429696 train.py:812] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 01:02:03.478717 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 01:02:03.892414 130828600429696 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 01:02:03.892587 130828600429696 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 01:02:03.892758 130828600429696 checkpointing.py:688] Setting up checkpoint logger... I0416 01:02:03.892796 130828600429696 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 01:02:03.892849 130828600429696 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 01:02:03.893313 130828600429696 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 0x76fc515f4e60>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 01:02:06.270721 130828600429696 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 01:02:06.270910 130828600429696 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 0x76f5d5b26a20>}, handler_registry=None I0416 01:02:06.271195 130828600429696 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x76f5d5b26a20>` 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`. I0416 01:02:06.271237 130828600429696 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x76f5d5331b20>` 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`. I0416 01:02:06.271268 130828600429696 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 0x76f5d5b26a20>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x76f5d5b26a20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x76f5d5331b20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x76f5d5331b20>}). I0416 01:02:06.271634 130828600429696 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 01:02:06.271689 130828600429696 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x76f5d533e0c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 01:02:06.402870 130828600429696 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints I0416 01:02:06.403151 130828600429696 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_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x76f5d5b90e60> I0416 01:02:06.403231 130828600429696 checkpointing.py:302] Checkpoint manager created! I0416 01:02:07.272595 130828600429696 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 01:02:07.272708 130828600429696 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 01:02:07.368308 130828600429696 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 01:02:07.368394 130828600429696 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 01:02:07.382491 130828600429696 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 01:02:07.382539 130828600429696 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 01:02:07.403569 130828600429696 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 01:02:07.403630 130828600429696 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 01:02:07.421268 130828600429696 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 01:02:07.421386 130828600429696 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 01:02:07.435704 130828600429696 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 01:02:07.435766 130828600429696 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 01:02:07.458426 130828600429696 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 01:02:07.458485 130828600429696 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 01:02:07.476443 130828600429696 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 01:02:07.476495 130828600429696 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 01:02:07.666751 130828600429696 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 01:02:07.666874 130828600429696 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0416 01:02:07.718853 3459413 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 I0416 01:02:08.100881 130828600429696 maxtext_utils.py:1790] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 01:02:08.100992 130828600429696 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 01:02:08.101032 130828600429696 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 01:02:08.101096 130828600429696 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 01:02:08.101138 130828600429696 maxtext_utils.py:1790] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 01:02:08.101166 130828600429696 maxtext_utils.py:1790] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 01:02:08.101203 130828600429696 maxtext_utils.py:1790] 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) I0416 01:02:08.101242 130828600429696 maxtext_utils.py:1790] 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') I0416 01:02:08.101270 130828600429696 maxtext_utils.py:1790] 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) I0416 01:02:08.101295 130828600429696 maxtext_utils.py:1790] 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) I0416 01:02:08.101329 130828600429696 maxtext_utils.py:1790] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 01:02:08.101362 130828600429696 maxtext_utils.py:1790] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 01:02:08.120186 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0416 01:02:08.120242 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0416 01:02:08.133129 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0416 01:02:08.133176 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0416 01:02:08.158817 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0416 01:02:08.158875 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0416 01:02:08.171733 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0416 01:02:08.171785 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0416 01:02:08.197483 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0416 01:02:08.197541 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0416 01:02:08.210402 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0416 01:02:08.210452 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0416 01:02:08.248877 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0416 01:02:08.248943 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0416 01:02:08.261826 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0416 01:02:08.261874 130828600429696 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0416 01:02:08.840417 130828600429696 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 01:02:08.840505 130828600429696 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 01:02:08.854084 130828600429696 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 01:02:08.854135 130828600429696 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 01:02:09.706510 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 01:02:09.843126 130828600429696 max_utils.py:791] Total memory size: 4.5 GB, Output size: 1.5 GB, Temp size: 2.9 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0416 01:02:09.843686 130828600429696 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 01:02:09.843845 130828600429696 metric_logger.py:289] number parameters: 1.104 billion W0416 01:02:11.234026 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 01:02:11.341170 130828600429696 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 01:02:11.763175 130828600429696 checkpointing.py:798] Waited 0.4219844341278076 seconds for step 0 to finish before starting checkpointing. I0416 01:02:11.763762 130828600429696 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 01:02:11.763937 130828600429696 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 01:02:11.764384 130828600429696 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0. I0416 01:02:11.852539 130828600429696 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 01:02:11.948489 130828600429696 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 01:02:11.948593 130828600429696 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 I0416 01:02:11.954573 130721056294464 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0 W0416 01:02:12.188493 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.203540 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.217203 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.226536 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.239007 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.245869 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.250525 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.254750 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 01:02:12.680986 130721045808704 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0/items I0416 01:02:12.685102 130828600429696 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.737467s I0416 01:02:12.685528 130828600429696 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 14.842 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 831 milliseconds) (per-host) I0416 01:02:12.685582 130828600429696 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.831589s (batch_requests_ready=0.089841s, total_serialization_initiated=0.741436s, others=0.000312s) I0416 01:02:12.685641 130828600429696 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.832517s (all_items=0.000030s, per_item={'items': '0.00002980'}, temp_paths=0.832488) I0416 01:02:12.686531 130720984991296 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 01:02:12.686661 130828600429696 async_checkpointer.py:561] Finished blocking save. Time taken: 0.922683s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0. I0416 01:02:12.686805 130828600429696 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 01:02:12.687038 130828600429696 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776301331.763746, 'wait_for_prev_duration_secs': 4.458427429199219e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776301331.7639577, 'checkpointer_blocking_duration_secs': 0.9227612018585205, 'get_old_steps_start_time': 1776301332.6867387, 'get_old_steps_duration_secs': 3.075599670410156e-05, 'checkpoint_manager_blocking_start_time': 1776301331.7636135, 'checkpoint_manager_blocking_duration_secs': 0.9234001636505127} I0416 01:02:12.687122 130828600429696 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 01:02:12.687170 130828600429696 max_utils.py:750] Memstats: After params initialized: I0416 01:02:12.687254 130828600429696 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0416 01:02:12.687281 130828600429696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 01:02:12.687301 130828600429696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 01:02:12.687318 130828600429696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 01:02:12.687336 130828600429696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 01:02:12.687354 130828600429696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 01:02:12.687371 130828600429696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 01:02:12.687388 130828600429696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 01:02:12.693505 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.699814 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 01:02:12.704697 3459413 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 01:02:12.821803 130721005962816 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776301332588353725, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776301332588353725, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0416 01:02:12.822022 130721066780224 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 01:02:13.013107 130828600429696 metric_logger.py:185] completed step: 0, seconds: 1.497, TFLOP/s/device: 36.309, Tokens/s/device: 5472.940, total_weights: 65536, loss: 10.871 I0416 01:02:13.014255 130828600429696 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/tensorboard/' I0416 01:02:13.185389 3461615 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 01:02:13.185467 3461615 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 01:02:13.428482 130828600429696 metric_logger.py:185] completed step: 1, seconds: 1.669, TFLOP/s/device: 32.559, Tokens/s/device: 4907.584, total_weights: 65536, loss: 10.871 I0416 01:02:13.852522 130828600429696 metric_logger.py:185] completed step: 2, seconds: 0.009, TFLOP/s/device: 6302.739, Tokens/s/device: 950017.395, total_weights: 65536, loss: 10.261 I0416 01:02:13.920493 130721035322944 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_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0/items/array_metadatas/process_0 I0416 01:02:14.270798 130828600429696 metric_logger.py:185] completed step: 3, seconds: 0.415, TFLOP/s/device: 130.839, Tokens/s/device: 19721.416, total_weights: 65536, loss: 9.732 I0416 01:02:46.727838 130721056294464 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 33.547343s (commit=33.085207s, array_metadata_write=0.462136s) I0416 01:02:46.729213 130720984991296 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 362.343 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 34 seconds) (per-host) I0416 01:02:46.729341 130720984991296 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 34.042716s. I0416 01:02:46.949870 130720984991296 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776301332588353725, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0416 01:02:47.121569 130720984991296 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 01:02:47.347794 130721005962816 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776301332588353725, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0416 01:02:47.516253 130720984991296 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 01:02:47.516821 130720984991296 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.522951s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0/items I0416 01:02:47.517564 130720984991296 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0/items I0416 01:02:47.758420 130720984991296 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0 I0416 01:02:48.429579 130720984991296 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_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0`. I0416 01:02:48.430352 130720984991296 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 36.666381s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/0 I0416 01:02:48.430433 130720984991296 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 35.743810s. I0416 01:02:48.430623 130721066780224 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 01:02:48.430738 130721066780224 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 01:02:48.430819 130721066780224 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 01:02:48.430870 130721066780224 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 01:02:50.165533 130828600429696 metric_logger.py:185] completed step: 4, seconds: 0.424, TFLOP/s/device: 128.158, Tokens/s/device: 19317.429, total_weights: 65536, loss: 9.274 I0416 01:02:50.173359 130828600429696 metric_logger.py:185] completed step: 5, seconds: 0.420, TFLOP/s/device: 129.547, Tokens/s/device: 19526.753, total_weights: 65536, loss: 8.888 I0416 01:02:50.595392 130828600429696 metric_logger.py:185] completed step: 6, seconds: 35.894, TFLOP/s/device: 1.514, Tokens/s/device: 228.225, total_weights: 65536, loss: 8.588 I0416 01:02:51.029439 130828600429696 metric_logger.py:185] completed step: 7, seconds: 0.005, TFLOP/s/device: 11521.839, Tokens/s/device: 1736697.053, total_weights: 65536, loss: 8.380 I0416 01:02:51.442526 130828600429696 metric_logger.py:185] completed step: 8, seconds: 0.423, TFLOP/s/device: 128.515, Tokens/s/device: 19371.239, total_weights: 65536, loss: 8.251 I0416 01:02:51.865409 130828600429696 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0416 01:02:51.866367 130828600429696 checkpointing.py:798] Waited 0.0009937286376953125 seconds for step 9 to finish before starting checkpointing. I0416 01:02:51.866631 130828600429696 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 01:02:51.866789 130828600429696 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 01:02:51.867112 130828600429696 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9. I0416 01:02:52.036248 130721066780224 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9 I0416 01:02:52.045227 130828600429696 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 01:02:52.045334 130828600429696 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 I0416 01:02:52.801123 130721045808704 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9/items I0416 01:02:54.674702 130828600429696 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 2.629932s I0416 01:02:54.679207 130828600429696 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.521 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 2 seconds) (per-host) I0416 01:02:54.679269 130828600429696 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 2.729880s (batch_requests_ready=0.092366s, total_serialization_initiated=2.633151s, others=0.004363s) I0416 01:02:54.679372 130828600429696 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.730486s (all_items=0.000020s, per_item={'items': '0.00002027'}, temp_paths=2.730466) I0416 01:02:54.681076 130720945145408 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 01:02:54.681161 130828600429696 async_checkpointer.py:561] Finished blocking save. Time taken: 2.814334s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9. I0416 01:02:54.681369 130828600429696 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 01:02:54.681595 130720984991296 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 01:02:54.681727 130828600429696 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776301371.8665931, 'wait_for_prev_duration_secs': 6.29425048828125e-05, 'time_between_consecutive_saves_sec': 3.4357054233551025, 'checkpointer_blocking_start_time': 1776301371.8668091, 'checkpointer_blocking_duration_secs': 2.814458131790161, 'get_old_steps_start_time': 1776301374.6812892, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776301371.8665593, 'checkpoint_manager_blocking_duration_secs': 2.815138101577759} I0416 01:02:54.681900 130828600429696 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0416 01:02:54.681938 130828600429696 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 01:02:55.428246 130721035322944 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_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9/items/array_metadatas/process_0 I0416 01:03:36.158416 130720966116928 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.478266s (commit=41.020679s, array_metadata_write=0.457587s) I0416 01:03:36.159554 130720945145408 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 285.835 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 44 seconds) (per-host) I0416 01:03:36.159616 130720945145408 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.478406s. I0416 01:03:36.577616 130720945145408 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 01:03:37.016260 130720945145408 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 01:03:37.016450 130720945145408 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.574952s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9/items I0416 01:03:37.017210 130720945145408 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9/items I0416 01:03:37.264147 130720945145408 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9 I0416 01:03:37.949128 130720945145408 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_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9`. I0416 01:03:37.949870 130720945145408 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 46.083046s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_004836/linen_feat_nnx_trainstate_and_training_loop_20260416_004836_06_grad_accum/checkpoints/9 I0416 01:03:37.949948 130720945145408 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 43.268741s. I0416 01:03:37.950199 130720984991296 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 01:03:37.950323 130720984991296 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 01:03:37.950402 130720984991296 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 01:03:37.950455 130720984991296 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 01:03:37.950712 130828600429696 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. I0416 01:03:37.950846 130828600429696 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 01:03:37.951574 130828600429696 metric_logger.py:185] completed step: 9, seconds: 0.434, TFLOP/s/device: 125.098, Tokens/s/device: 18856.068, total_weights: 65536, loss: 8.175 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops