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 | 8.1750 | 8.1750 | 0 |
| TFLOP/s | 128.748 | 128.516 | -0.232 |
| Tok/s | 19406.3 | 19371.3 | -34.967 |
| Avg s/step | 4.678 | 4.922 | +0.244 |
| Memory % | 5.09 | 5.12 | +0.03 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-17 13:05:49.390933: 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 13:05:49.506703 125892250631296 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0417 13:06:35.051976 125892250631296 max_utils.py:800] System Information: Jax Version: 0.9.2 I0417 13:06:35.052129 125892250631296 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0417 13:06:35.052169 125892250631296 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 13:06:35.052199 125892250631296 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system I0417 13:06:35.052222 125892250631296 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0417 13:06:35.052312 125892250631296 train.py:671] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0417 13:06:35.146122 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:06:35.546767 125892250631296 maxtext_utils.py:1548] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 13:06:35.547130 125892250631296 checkpointing.py:677] Setting up checkpoint logger... I0417 13:06:35.547180 125892250631296 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0417 13:06:35.547250 125892250631296 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0417 13:06:35.547866 125892250631296 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 0x727efbfcf5f0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0417 13:06:38.024255 125892250631296 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0417 13:06:38.024465 125892250631296 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 0x72787fd323f0>}, handler_registry=None I0417 13:06:38.024800 125892250631296 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72787fd323f0>` 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 13:06:38.024846 125892250631296 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72787fd318e0>` 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 13:06:38.024875 125892250631296 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 0x72787fd323f0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72787fd323f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72787fd318e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72787fd318e0>}). I0417 13:06:38.025182 125892250631296 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0417 13:06:38.025238 125892250631296 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x72787fd2e200> timeout: 600 secs and primary_host=0 for async checkpoint writes I0417 13:06:38.139842 125892250631296 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints I0417 13:06:38.140116 125892250631296 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_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x72788013cc80> I0417 13:06:38.140205 125892250631296 checkpointing.py:301] Checkpoint manager created! I0417 13:06:39.345261 125892250631296 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0417 13:06:39.345359 125892250631296 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 13:06:39.428207 125892250631296 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 13:06:39.428278 125892250631296 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 13:06:39.442554 125892250631296 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 13:06:39.442607 125892250631296 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 13:06:39.463835 125892250631296 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 13:06:39.463906 125892250631296 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 13:06:39.478335 125892250631296 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 13:06:39.478391 125892250631296 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 13:06:39.492675 125892250631296 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 13:06:39.492725 125892250631296 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 13:06:39.514610 125892250631296 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0417 13:06:39.514671 125892250631296 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 13:06:39.532941 125892250631296 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0417 13:06:39.532994 125892250631296 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0417 13:06:39.750607 125892250631296 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0417 13:06:39.750706 125892250631296 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0417 13:06:39.800871 428872 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 13:06:40.182866 125892250631296 maxtext_utils.py:1651] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0417 13:06:40.182970 125892250631296 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 13:06:40.183012 125892250631296 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 13:06:40.183071 125892250631296 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 13:06:40.183115 125892250631296 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 13:06:40.183148 125892250631296 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 13:06:40.183188 125892250631296 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 13:06:40.183227 125892250631296 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 13:06:40.183255 125892250631296 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 13:06:40.183281 125892250631296 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 13:06:40.183315 125892250631296 maxtext_utils.py:1651] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0417 13:06:40.183349 125892250631296 maxtext_utils.py:1651] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0417 13:06:40.201374 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0417 13:06:40.201426 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0417 13:06:40.214501 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0417 13:06:40.214549 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0417 13:06:40.240512 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0417 13:06:40.240566 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0417 13:06:40.253577 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0417 13:06:40.253625 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0417 13:06:40.279528 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0417 13:06:40.279580 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0417 13:06:40.292730 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0417 13:06:40.292779 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0417 13:06:40.331857 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0417 13:06:40.331914 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0417 13:06:40.344942 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0417 13:06:40.344990 125892250631296 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0417 13:06:40.927672 125892250631296 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0417 13:06:40.927751 125892250631296 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0417 13:06:40.941413 125892250631296 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0417 13:06:40.941462 125892250631296 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0417 13:06:48.899555 125892250631296 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. I0417 13:06:48.900145 125892250631296 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0417 13:06:48.900341 125892250631296 metric_logger.py:301] number parameters: 1.104 billion W0417 13:06:50.274946 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:06:50.384566 125892250631296 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0417 13:06:50.795612 125892250631296 checkpointing.py:776] Waited 0.41103100776672363 seconds for step 0 to finish before starting checkpointing. I0417 13:06:50.796106 125892250631296 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 13:06:50.796254 125892250631296 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0417 13:06:50.796616 125892250631296 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0. I0417 13:06:50.881441 125892250631296 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0417 13:06:50.968827 125784624727616 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0 I0417 13:06:50.985560 125892250631296 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0417 13:06:50.985655 125892250631296 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 W0417 13:06:51.231304 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.249200 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.271727 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.281200 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.287414 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.292862 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.299779 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.304973 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:06:51.638196 125892250631296 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.653034s I0417 13:06:51.638639 125892250631296 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 16.325 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 755 milliseconds) (per-host) I0417 13:06:51.638695 125892250631296 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.756040s (batch_requests_ready=0.099108s, total_serialization_initiated=0.656580s, others=0.000352s) I0417 13:06:51.638755 125892250631296 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.756852s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.756834) I0417 13:06:51.639585 125784561813056 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0417 13:06:51.639682 125892250631296 async_checkpointer.py:561] Finished blocking save. Time taken: 0.843381s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0. I0417 13:06:51.639853 125892250631296 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0417 13:06:51.640021 125784635213376 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 13:06:51.640137 125892250631296 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776431210.7960935, 'wait_for_prev_duration_secs': 4.1484832763671875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776431210.7962756, 'checkpointer_blocking_duration_secs': 0.8435051441192627, 'get_old_steps_start_time': 1776431211.6397994, 'get_old_steps_duration_secs': 2.0503997802734375e-05, 'checkpoint_manager_blocking_start_time': 1776431210.7959783, 'checkpoint_manager_blocking_duration_secs': 0.8441298007965088} I0417 13:06:51.640262 125892250631296 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0417 13:06:51.640302 125892250631296 max_utils.py:750] Memstats: After params initialized: I0417 13:06:51.640375 125892250631296 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0417 13:06:51.640401 125892250631296 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0417 13:06:51.640420 125892250631296 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0417 13:06:51.640438 125892250631296 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0417 13:06:51.640455 125892250631296 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0417 13:06:51.640471 125892250631296 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0417 13:06:51.640488 125892250631296 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0417 13:06:51.640504 125892250631296 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0417 13:06:51.646374 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.651904 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:06:51.656740 428872 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:06:51.972814 125784542938688 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776431211701713201, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776431211701713201, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0417 13:06:51.973650 125784614241856 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0/items I0417 13:06:51.980335 125892250631296 metric_logger.py:196] completed step: 0, seconds: 1.483, TFLOP/s/device: 36.653, Tokens/s/device: 5524.724, total_weights: 65536, loss: 10.871, lm_loss: 10.871, perplexity: 52637.648 I0417 13:06:51.981264 125892250631296 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/tensorboard/' I0417 13:06:52.389835 125892250631296 metric_logger.py:196] completed step: 1, seconds: 1.593, TFLOP/s/device: 34.110, Tokens/s/device: 5141.359, total_weights: 65536, loss: 10.871, lm_loss: 10.871, perplexity: 52637.648 I0417 13:06:52.470475 431951 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0417 13:06:52.470544 431951 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0417 13:06:52.818822 125892250631296 metric_logger.py:196] completed step: 2, seconds: 0.007, TFLOP/s/device: 7312.771, Tokens/s/device: 1102260.495, total_weights: 65536, loss: 10.261, lm_loss: 10.261, perplexity: 28586.895 I0417 13:06:53.242262 125892250631296 metric_logger.py:196] completed step: 3, seconds: 0.410, TFLOP/s/device: 132.469, Tokens/s/device: 19967.095, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16845.295 I0417 13:06:53.273375 125784582784576 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_06_grad_accum/checkpoints/0/items/array_metadatas/process_0 I0417 13:07:20.966687 125784572298816 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 28.493491s (commit=28.095564s, array_metadata_write=0.397927s) I0417 13:07:20.967909 125784561813056 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 420.033 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 30 seconds) (per-host) I0417 13:07:20.968027 125784561813056 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 29.328300s. I0417 13:07:21.223696 125784561813056 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776431211701713201, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0417 13:07:21.420591 125784561813056 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 13:07:21.629398 125784542938688 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776431211701713201, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0417 13:07:21.842678 125784561813056 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0417 13:07:21.842854 125784561813056 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.582400s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0/items I0417 13:07:21.843970 125784561813056 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0/items I0417 13:07:22.084426 125784561813056 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0 I0417 13:07:22.764418 125784561813056 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_06_grad_accum/checkpoints/0`. I0417 13:07:22.765079 125784561813056 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 31.968793s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/0 I0417 13:07:22.765161 125784561813056 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 31.125435s. I0417 13:07:22.765334 125784635213376 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 13:07:22.765442 125784635213376 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 13:07:22.765515 125784635213376 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0417 13:07:22.765563 125784635213376 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0417 13:07:31.635401 125892250631296 metric_logger.py:196] completed step: 4, seconds: 0.429, TFLOP/s/device: 126.614, Tokens/s/device: 19084.672, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10654.189 I0417 13:07:31.643260 125892250631296 metric_logger.py:196] completed step: 5, seconds: 0.423, TFLOP/s/device: 128.400, Tokens/s/device: 19353.848, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7241.146 I0417 13:07:32.059361 125892250631296 metric_logger.py:196] completed step: 6, seconds: 38.394, TFLOP/s/device: 1.416, Tokens/s/device: 213.366, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5368.870 I0417 13:07:32.482414 125892250631296 metric_logger.py:196] completed step: 7, seconds: 0.005, TFLOP/s/device: 11672.791, Tokens/s/device: 1759450.172, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4361.125 I0417 13:07:32.905975 125892250631296 metric_logger.py:196] completed step: 8, seconds: 0.418, TFLOP/s/device: 129.945, Tokens/s/device: 19586.700, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3832.059 I0417 13:07:33.333686 125892250631296 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0417 13:07:33.334509 125892250631296 checkpointing.py:776] Waited 0.0008425712585449219 seconds for step 9 to finish before starting checkpointing. I0417 13:07:33.334774 125892250631296 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 13:07:33.334931 125892250631296 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0417 13:07:33.335217 125892250631296 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/9. I0417 13:07:33.518236 125892250631296 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0417 13:07:33.518330 125892250631296 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 13:07:33.522763 125784635213376 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/9 I0417 13:07:34.207842 125784406623808 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/9/items I0417 13:07:41.103534 125892250631296 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 7.585704s I0417 13:07:41.107770 125892250631296 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.607 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 7 seconds) (per-host) I0417 13:07:41.107893 125892250631296 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 7.677572s (batch_requests_ready=0.084610s, total_serialization_initiated=7.588780s, others=0.004182s) I0417 13:07:41.107991 125892250631296 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 7.678095s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=7.678078) I0417 13:07:41.109450 125784582784576 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0417 13:07:41.109547 125892250631296 async_checkpointer.py:561] Finished blocking save. Time taken: 7.774575s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/9. I0417 13:07:41.109743 125892250631296 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0417 13:07:41.109993 125784561813056 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 13:07:41.110120 125892250631296 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776431253.3347373, 'wait_for_prev_duration_secs': 6.222724914550781e-05, 'time_between_consecutive_saves_sec': 10.569158792495728, 'checkpointer_blocking_start_time': 1776431253.3349524, 'checkpointer_blocking_duration_secs': 7.774705171585083, 'get_old_steps_start_time': 1776431261.1096764, 'get_old_steps_duration_secs': 2.384185791015625e-05, 'checkpoint_manager_blocking_start_time': 1776431253.3347034, 'checkpoint_manager_blocking_duration_secs': 7.775391101837158} I0417 13:07:41.110253 125892250631296 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0417 13:07:41.110288 125892250631296 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 13:07:41.827071 125784624727616 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_06_grad_accum/checkpoints/9/items/array_metadatas/process_0 I0417 13:08:14.102394 125784572298816 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.993877s (commit=32.535439s, array_metadata_write=0.458438s) I0417 13:08:14.103420 125784582784576 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 310.692 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 40 seconds) (per-host) I0417 13:08:14.103536 125784582784576 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.993939s. I0417 13:08:14.522731 125784582784576 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 13:08:14.978091 125784582784576 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0417 13:08:14.978802 125784582784576 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.601053s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/9/items I0417 13:08:14.979437 125784582784576 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/9/items I0417 13:08:15.201379 125784582784576 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/9 I0417 13:08:15.860599 125784582784576 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_06_grad_accum/checkpoints/9`. I0417 13:08:15.861243 125784582784576 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 42.526273s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_06_grad_accum/checkpoints/9 I0417 13:08:15.861317 125784582784576 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.751724s. I0417 13:08:15.861493 125784561813056 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 13:08:15.861605 125784561813056 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 13:08:15.861691 125784561813056 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0417 13:08:15.861740 125784561813056 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0417 13:08:15.861923 125892250631296 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 13:08:15.862340 125892250631296 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 13:08:15.863193 125892250631296 metric_logger.py:196] completed step: 9, seconds: 0.422, TFLOP/s/device: 128.748, Tokens/s/device: 19406.252, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3551.987 Per train step: Total TFLOPs: 54.35 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 14:37:27.385933: 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 14:37:27.505069 126122499230848 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0420 14:38:15.833318 126122499230848 max_utils.py:800] System Information: Jax Version: 0.9.2 I0420 14:38:15.833432 126122499230848 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0420 14:38:15.833466 126122499230848 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 14:38:15.833490 126122499230848 train_utils.py:365] WARNING: 'dataset_path' might be pointing your local file system I0420 14:38:15.833510 126122499230848 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0420 14:38:15.833609 126122499230848 train.py:806] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. I0420 14:38:16.345761 126122499230848 maxtext_utils.py:1718] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 14:38:16.345938 126122499230848 maxtext_utils.py:1718] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 14:38:16.346149 126122499230848 checkpointing.py:688] Setting up checkpoint logger... I0420 14:38:16.346191 126122499230848 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0420 14:38:16.346233 126122499230848 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0420 14:38:16.346723 126122499230848 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 0x72b497d00c20>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0420 14:38:18.747622 126122499230848 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0420 14:38:18.747828 126122499230848 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 0x72ae1bf3f440>}, handler_registry=None I0420 14:38:18.748113 126122499230848 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72ae1bf3f440>` 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 14:38:18.748155 126122499230848 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72aff6f449b0>` 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 14:38:18.748183 126122499230848 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 0x72ae1bf3f440>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x72ae1bf3f440>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72aff6f449b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x72aff6f449b0>}). I0420 14:38:18.748475 126122499230848 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0420 14:38:18.748524 126122499230848 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x72ae1bf4eac0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0420 14:38:18.880818 126122499230848 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_06_grad_accum/checkpoints I0420 14:38:18.881079 126122499230848 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_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x72ae1bf60350> I0420 14:38:18.881160 126122499230848 checkpointing.py:302] Checkpoint manager created! I0420 14:38:20.072232 126122499230848 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0420 14:38:20.072326 126122499230848 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0420 14:38:20.156129 126122499230848 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 14:38:20.156204 126122499230848 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0420 14:38:20.170319 126122499230848 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 14:38:20.170368 126122499230848 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0420 14:38:20.190744 126122499230848 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 14:38:20.190801 126122499230848 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0420 14:38:20.205016 126122499230848 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 14:38:20.205083 126122499230848 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0420 14:38:20.219183 126122499230848 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 14:38:20.219230 126122499230848 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0420 14:38:20.240797 126122499230848 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0420 14:38:20.240850 126122499230848 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0420 14:38:20.258901 126122499230848 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0420 14:38:20.258954 126122499230848 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0420 14:38:20.449028 126122499230848 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0420 14:38:20.449131 126122499230848 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0420 14:38:20.499681 1051146 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 14:38:20.880793 126122499230848 maxtext_utils.py:1821] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0420 14:38:20.880896 126122499230848 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 14:38:20.880938 126122499230848 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 14:38:20.880981 126122499230848 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 14:38:20.881021 126122499230848 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 14:38:20.881065 126122499230848 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 14:38:20.881107 126122499230848 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 14:38:20.881146 126122499230848 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 14:38:20.881174 126122499230848 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 14:38:20.881199 126122499230848 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 14:38:20.881234 126122499230848 maxtext_utils.py:1821] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0420 14:38:20.881268 126122499230848 maxtext_utils.py:1821] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0420 14:38:20.900223 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0420 14:38:20.900278 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0420 14:38:20.913388 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0420 14:38:20.913434 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0420 14:38:20.939302 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0420 14:38:20.939357 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0420 14:38:20.952331 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0420 14:38:20.952377 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0420 14:38:20.978347 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0420 14:38:20.978400 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0420 14:38:20.991600 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0420 14:38:20.991645 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0420 14:38:21.030493 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0420 14:38:21.030553 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0420 14:38:21.043566 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0420 14:38:21.043616 126122499230848 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0420 14:38:21.625782 126122499230848 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0420 14:38:21.625866 126122499230848 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0420 14:38:21.639667 126122499230848 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0420 14:38:21.639717 126122499230848 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0420 14:38:29.894710 126122499230848 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. I0420 14:38:29.895275 126122499230848 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0420 14:38:29.895611 126122499230848 metric_logger.py:301] number parameters: 1.104 billion W0420 14:38:31.311465 1051146 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0420 14:38:31.435689 126122499230848 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0420 14:38:31.857642 126122499230848 checkpointing.py:798] Waited 0.4219346046447754 seconds for step 0 to finish before starting checkpointing. I0420 14:38:31.858148 126122499230848 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 14:38:31.858296 126122499230848 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0420 14:38:31.858666 126122499230848 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_06_grad_accum/checkpoints/0. I0420 14:38:31.945369 126122499230848 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0420 14:38:32.037656 126014900405824 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_06_grad_accum/checkpoints/0 I0420 14:38:32.047758 126122499230848 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0420 14:38:32.047848 126122499230848 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 14:38:32.823633 126014889920064 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_06_grad_accum/checkpoints/0/items I0420 14:38:32.964822 126014841685568 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776695912706374393, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776695912706374393, "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_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0420 14:38:32.967428 126122499230848 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.920078s I0420 14:38:32.968090 126122499230848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 12.084 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0420 14:38:32.968244 126122499230848 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.021590s (batch_requests_ready=0.096899s, total_serialization_initiated=0.924096s, others=0.000595s) I0420 14:38:32.968373 126122499230848 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.022531s (all_items=0.000020s, per_item={'items': '0.00002003'}, temp_paths=1.022511) I0420 14:38:32.969580 126014820714048 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0420 14:38:32.969684 126122499230848 async_checkpointer.py:561] Finished blocking save. Time taken: 1.111348s. 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_06_grad_accum/checkpoints/0. I0420 14:38:32.969895 126122499230848 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0420 14:38:32.970140 126014910891584 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0420 14:38:32.970255 126122499230848 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_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776695911.8581355, 'wait_for_prev_duration_secs': 4.172325134277344e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776695911.858317, 'checkpointer_blocking_duration_secs': 1.1114866733551025, 'get_old_steps_start_time': 1776695912.9698265, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776695911.8580189, 'checkpoint_manager_blocking_duration_secs': 1.1122069358825684} I0420 14:38:32.970379 126122499230848 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0420 14:38:32.970426 126122499230848 max_utils.py:750] Memstats: After params initialized: I0420 14:38:32.970533 126122499230848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0420 14:38:32.970561 126122499230848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_1(process=0,(1,0,0,0)) I0420 14:38:32.970587 126122499230848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_2(process=0,(0,1,0,0)) I0420 14:38:32.970607 126122499230848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_3(process=0,(1,1,0,0)) I0420 14:38:32.970625 126122499230848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_4(process=0,(0,2,0,0)) I0420 14:38:32.970643 126122499230848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_5(process=0,(1,2,0,0)) I0420 14:38:32.970660 126122499230848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_6(process=0,(0,3,0,0)) I0420 14:38:32.970677 126122499230848 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_7(process=0,(1,3,0,0)) I0420 14:38:33.404964 1053603 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0420 14:38:33.405031 1053603 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0420 14:38:33.432225 126122499230848 metric_logger.py:196] completed step: 0, seconds: 1.539, TFLOP/s/device: 35.304, Tokens/s/device: 5321.457, total_weights: 65536, loss: 10.871, lm_loss: 10.871, perplexity: 52637.648 I0420 14:38:33.433291 126122499230848 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_06_grad_accum/tensorboard/' I0420 14:38:33.833002 126122499230848 metric_logger.py:196] completed step: 1, seconds: 1.979, TFLOP/s/device: 27.468, Tokens/s/device: 4140.345, total_weights: 65536, loss: 10.871, lm_loss: 10.871, perplexity: 52637.648 I0420 14:38:34.147077 126014900405824 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_06_grad_accum/checkpoints/0/items/array_metadatas/process_0 I0420 14:38:34.256732 126122499230848 metric_logger.py:196] completed step: 2, seconds: 0.023, TFLOP/s/device: 2322.487, Tokens/s/device: 350070.510, total_weights: 65536, loss: 10.261, lm_loss: 10.261, perplexity: 28586.895 I0420 14:38:34.678585 126122499230848 metric_logger.py:196] completed step: 3, seconds: 0.401, TFLOP/s/device: 135.461, Tokens/s/device: 20418.082, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16845.295 I0420 14:39:12.819920 126014831199808 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 39.421639s (commit=38.956136s, array_metadata_write=0.465503s) I0420 14:39:12.821189 126014820714048 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 309.161 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 40 seconds) (per-host) I0420 14:39:12.821302 126014820714048 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.851571s. I0420 14:39:13.068611 126014820714048 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776695912706374393, '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_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0420 14:39:13.254631 126014820714048 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 14:39:13.457221 126014841685568 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776695912706374393, '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_06_grad_accum/checkpoints/0/_CHECKPOINT_METADATA I0420 14:39:13.718547 126014820714048 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0420 14:39:13.719207 126014820714048 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.609373s. 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_06_grad_accum/checkpoints/0/items I0420 14:39:13.719899 126014820714048 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_06_grad_accum/checkpoints/0/items I0420 14:39:13.956246 126014820714048 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_06_grad_accum/checkpoints/0 I0420 14:39:14.621115 126014820714048 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_06_grad_accum/checkpoints/0`. I0420 14:39:14.621830 126014820714048 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 42.763502s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_06_grad_accum/checkpoints/0 I0420 14:39:14.621904 126014820714048 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 41.652175s. I0420 14:39:14.622120 126014910891584 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 14:39:14.622238 126014910891584 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0420 14:39:14.622318 126014910891584 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0420 14:39:14.622369 126014910891584 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0420 14:39:14.869235 126122499230848 metric_logger.py:196] completed step: 4, seconds: 0.424, TFLOP/s/device: 128.159, Tokens/s/device: 19317.565, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10654.189 I0420 14:39:14.877570 126122499230848 metric_logger.py:196] completed step: 5, seconds: 0.422, TFLOP/s/device: 128.922, Tokens/s/device: 19432.538, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7241.146 I0420 14:39:15.304876 126122499230848 metric_logger.py:196] completed step: 6, seconds: 40.192, TFLOP/s/device: 1.352, Tokens/s/device: 203.823, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5368.870 I0420 14:39:15.727982 126122499230848 metric_logger.py:196] completed step: 7, seconds: 0.005, TFLOP/s/device: 10310.855, Tokens/s/device: 1554164.295, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4361.125 I0420 14:39:16.150401 126122499230848 metric_logger.py:196] completed step: 8, seconds: 0.428, TFLOP/s/device: 126.913, Tokens/s/device: 19129.773, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3832.059 I0420 14:39:16.576132 126122499230848 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0420 14:39:16.577006 126122499230848 checkpointing.py:798] Waited 0.0009050369262695312 seconds for step 9 to finish before starting checkpointing. I0420 14:39:16.577287 126122499230848 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 14:39:16.577453 126122499230848 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0420 14:39:16.577754 126122499230848 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_06_grad_accum/checkpoints/9. I0420 14:39:16.744937 126014910891584 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_06_grad_accum/checkpoints/9 I0420 14:39:16.748858 126122499230848 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0420 14:39:16.748956 126122499230848 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 14:39:17.476798 126014709564992 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_06_grad_accum/checkpoints/9/items I0420 14:39:19.310005 126122499230848 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 2.561664s I0420 14:39:19.314364 126122499230848 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.662 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 2 seconds) (per-host) I0420 14:39:19.314428 126122499230848 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 2.646938s (batch_requests_ready=0.077637s, total_serialization_initiated=2.565084s, others=0.004217s) I0420 14:39:19.315063 126122499230848 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.648047s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=2.648031) I0420 14:39:19.316179 126014910891584 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0420 14:39:19.316288 126122499230848 async_checkpointer.py:561] Finished blocking save. Time taken: 2.738787s. 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_06_grad_accum/checkpoints/9. I0420 14:39:19.316535 126122499230848 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0420 14:39:19.316725 126014820714048 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0420 14:39:19.316835 126122499230848 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_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776695956.577248, 'wait_for_prev_duration_secs': 6.628036499023438e-05, 'time_between_consecutive_saves_sec': 1.9548625946044922, 'checkpointer_blocking_start_time': 1776695956.5774744, 'checkpointer_blocking_duration_secs': 2.7389557361602783, 'get_old_steps_start_time': 1776695959.316456, 'get_old_steps_duration_secs': 3.528594970703125e-05, 'checkpoint_manager_blocking_start_time': 1776695956.5772111, 'checkpoint_manager_blocking_duration_secs': 2.739596128463745} I0420 14:39:19.316970 126122499230848 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0420 14:39:19.317005 126122499230848 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 14:39:20.042745 126014900405824 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_06_grad_accum/checkpoints/9/items/array_metadatas/process_0 I0420 14:39:56.813860 126014831199808 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.499245s (commit=37.025294s, array_metadata_write=0.473951s) I0420 14:39:56.814945 126014910891584 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 314.760 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 40 seconds) (per-host) I0420 14:39:56.815076 126014910891584 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.498722s. I0420 14:39:57.250367 126014910891584 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 14:39:57.654361 126014910891584 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0420 14:39:57.654553 126014910891584 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.551027s. 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_06_grad_accum/checkpoints/9/items I0420 14:39:57.655235 126014910891584 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_06_grad_accum/checkpoints/9/items I0420 14:39:57.896661 126014910891584 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_06_grad_accum/checkpoints/9 I0420 14:39:58.589334 126014910891584 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_06_grad_accum/checkpoints/9`. I0420 14:39:58.590019 126014910891584 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 42.012527s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260420_142345/linen_feat_nnx_trainstate_and_training_loop_20260420_142345_06_grad_accum/checkpoints/9 I0420 14:39:58.590112 126014910891584 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 39.273765s. I0420 14:39:58.590270 126014820714048 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 14:39:58.590381 126014820714048 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0420 14:39:58.590475 126014820714048 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0420 14:39:58.590524 126014820714048 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0420 14:39:58.590662 126122499230848 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 14:39:58.590793 126122499230848 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 14:39:58.591614 126122499230848 metric_logger.py:196] completed step: 9, seconds: 0.423, TFLOP/s/device: 128.516, Tokens/s/device: 19371.285, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3551.987 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops