main (0ba93e21a) vs feat/nnx-post-train-fixes (d8cde296b)| Metric | main 0ba93e21a | feat/nnx-post-train-fixes d8cde296b | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.8880 | 8.8880 | 0 |
| TFLOP/s | 64.855 | 65.156 | +0.301 |
| Tok/s | 9775.7 | 9821.0 | +45.379 |
| Avg s/step | 1.400 | 1.361 | -0.039 |
| Memory % | 5.06 | 5.06 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-16 13:02:52.181665: 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 13:02:52.299225 137380359089280 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 13:03:46.237938 137380359089280 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 13:03:46.238135 137380359089280 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 13:03:46.238172 137380359089280 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 13:03:46.238198 137380359089280 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0416 13:03:46.238279 137380359089280 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 13:03:46.329646 3847140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:03:46.724759 137380359089280 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 2, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0416 13:03:46.725034 137380359089280 checkpointing.py:677] Setting up checkpoint logger... I0416 13:03:46.725095 137380359089280 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 13:03:46.725141 137380359089280 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 13:03:46.725586 137380359089280 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 0x7cf1c49cf8c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 13:03:49.093338 137380359089280 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0416 13:03:49.093519 137380359089280 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 0x7ceb49f28f20>}, handler_registry=None I0416 13:03:49.093788 137380359089280 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ceb49f28f20>` 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 13:03:49.093827 137380359089280 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ceb487363f0>` 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 13:03:49.093860 137380359089280 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 0x7ceb49f28f20>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ceb49f28f20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ceb487363f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ceb487363f0>}). I0416 13:03:49.094423 137380359089280 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 13:03:49.094476 137380359089280 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7ceb487311c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 13:03:49.223177 137380359089280 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints I0416 13:03:49.223396 137380359089280 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_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ceb487341d0> I0416 13:03:49.223476 137380359089280 checkpointing.py:301] Checkpoint manager created! I0416 13:03:49.304945 137380359089280 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 13:03:49.307995 137380359089280 reader.py:262] Creating a tf.data.Dataset reading 8 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0. I0416 13:03:49.361029 137380359089280 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split train, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 13:03:49.392091 137380359089280 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 13:03:49.392159 137380359089280 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 13:03:50.369888 137380359089280 nnx_wrappers.py:437] Unknown Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 13:03:50.369988 137380359089280 nnx_wrappers.py:437] Unknown Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0416 13:03:50.471856 137380359089280 attentions.py:1088] attentions/inputs_q Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:03:50.471932 137380359089280 attentions.py:1088] attentions/inputs_q Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0416 13:03:50.486908 137380359089280 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:03:50.486960 137380359089280 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0416 13:03:50.509135 137380359089280 attentions.py:1154] attentions/query Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:03:50.509189 137380359089280 attentions.py:1154] attentions/query Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0416 13:03:50.524201 137380359089280 attentions.py:1155] attentions/key Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:03:50.524249 137380359089280 attentions.py:1155] attentions/key Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0416 13:03:50.539183 137380359089280 attentions.py:1156] attentions/value Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:03:50.539231 137380359089280 attentions.py:1156] attentions/value Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0416 13:03:50.561380 137380359089280 attentions.py:1197] attentions/out Logical: bfloat16[2,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 13:03:50.561439 137380359089280 attentions.py:1197] attentions/out Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0416 13:03:50.580449 137380359089280 linears.py:525] linears/x Logical: bfloat16[2,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 13:03:50.580503 137380359089280 linears.py:525] linears/x Physical: bfloat16[2,2048,7168]....................................... ('fsdp', None, 'tensor'). I0416 13:03:50.767811 137380359089280 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0416 13:03:50.767904 137380359089280 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0416 13:03:50.818662 3847140 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: 2 tensor: 4 I0416 13:03:51.215177 137380359089280 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: ('tensor',) I0416 13:03:51.215285 137380359089280 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, 'tensor') I0416 13:03:51.215327 137380359089280 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, 'tensor') I0416 13:03:51.215369 137380359089280 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0416 13:03:51.215407 137380359089280 maxtext_utils.py:1620] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0416 13:03:51.215435 137380359089280 maxtext_utils.py:1620] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0416 13:03:51.215472 137380359089280 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, 'tensor', None) I0416 13:03:51.215511 137380359089280 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: ('tensor', None, None, 'fsdp') I0416 13:03:51.215539 137380359089280 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, 'tensor', None) I0416 13:03:51.215565 137380359089280 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, 'tensor', None) I0416 13:03:51.215601 137380359089280 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0416 13:03:51.215635 137380359089280 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0416 13:03:51.647786 137380359089280 train.py:155] train/xent Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:03:51.647867 137380359089280 train.py:155] train/xent Physical: float32[2,2048]............................................. ('fsdp', None). I0416 13:03:51.662128 137380359089280 train.py:162] train/z_loss Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:03:51.662181 137380359089280 train.py:162] train/z_loss Physical: float32[2,2048]............................................. ('fsdp', None). W0416 13:03:51.919522 3847140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:03:51.998568 137380359089280 max_utils.py:791] Total memory size: 2.5 GB, Output size: 1.5 GB, Temp size: 0.9 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0416 13:03:51.999058 137380359089280 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 13:03:51.999198 137380359089280 metric_logger.py:289] number parameters: 1.104 billion W0416 13:03:55.073303 3847140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:03:55.168344 137380359089280 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0416 13:03:55.223359 137380359089280 checkpointing.py:776] Waited 0.05500650405883789 seconds for step 0 to finish before starting checkpointing. I0416 13:03:55.223785 137380359089280 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:03:55.223931 137380359089280 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 13:03:55.224309 137380359089280 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0. I0416 13:03:55.317451 137380359089280 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 13:03:55.403723 137267219793472 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0 I0416 13:03:55.413340 137380359089280 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:03:55.413434 137380359089280 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 13:03:56.196797 137267209307712 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0/items I0416 13:03:56.354315 137267169461824 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776344636100653907, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776344636100653907, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0416 13:03:56.724166 3849769 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 13:03:56.724220 3849769 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 13:03:56.815544 137380359089280 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.402626s I0416 13:03:56.819849 137380359089280 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 8.222 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 13:03:56.819910 137380359089280 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.501122s (batch_requests_ready=0.090023s, total_serialization_initiated=1.406894s, others=0.004204s) I0416 13:03:56.819985 137380359089280 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.502056s (all_items=0.000020s, per_item={'items': '0.00001979'}, temp_paths=1.502036) I0416 13:03:56.821116 137267085575744 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:03:56.821197 137380359089280 async_checkpointer.py:561] Finished blocking save. Time taken: 1.597227s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0. I0416 13:03:56.821377 137380359089280 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:03:56.821612 137267230279232 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:03:56.821718 137380359089280 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776344635.2237725, 'wait_for_prev_duration_secs': 4.100799560546875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776344635.2239518, 'checkpointer_blocking_duration_secs': 1.5973501205444336, 'get_old_steps_start_time': 1776344636.8213224, 'get_old_steps_duration_secs': 2.0503997802734375e-05, 'checkpoint_manager_blocking_start_time': 1776344635.223691, 'checkpoint_manager_blocking_duration_secs': 1.5980024337768555} I0416 13:03:56.821825 137380359089280 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0416 13:03:56.821863 137380359089280 max_utils.py:750] Memstats: After params initialized: I0416 13:03:56.821907 137380359089280 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_0(process=0,(0,0,0,0)) I0416 13:03:56.821938 137380359089280 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_1(process=0,(1,0,0,0)) I0416 13:03:56.821960 137380359089280 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_2(process=0,(0,1,0,0)) I0416 13:03:56.821978 137380359089280 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_3(process=0,(1,1,0,0)) I0416 13:03:56.821995 137380359089280 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_4(process=0,(0,2,0,0)) I0416 13:03:56.822011 137380359089280 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_5(process=0,(1,2,0,0)) I0416 13:03:56.822027 137380359089280 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_6(process=0,(0,3,0,0)) I0416 13:03:56.822057 137380359089280 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_7(process=0,(1,3,0,0)) W0416 13:03:56.827481 3847140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:03:56.834389 3847140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:03:56.838983 3847140 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:03:57.196553 137380359089280 metric_logger.py:185] completed step: 0, seconds: 3.169, TFLOP/s/device: 1.072, Tokens/s/device: 161.582, total_weights: 3650, loss: 10.896 I0416 13:03:57.197472 137380359089280 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/tensorboard/' I0416 13:03:57.247238 137380359089280 metric_logger.py:185] completed step: 1, seconds: 2.026, TFLOP/s/device: 1.676, Tokens/s/device: 252.664, total_weights: 2676, loss: 10.833 I0416 13:03:57.299698 137380359089280 metric_logger.py:185] completed step: 2, seconds: 0.012, TFLOP/s/device: 271.960, Tokens/s/device: 40992.794, total_weights: 3783, loss: 10.136 I0416 13:03:57.351864 137380359089280 metric_logger.py:185] completed step: 3, seconds: 0.051, TFLOP/s/device: 66.450, Tokens/s/device: 10016.041, total_weights: 3680, loss: 9.403 I0416 13:03:57.627885 137267117033024 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_0 I0416 13:04:07.649659 137380359089280 metric_logger.py:185] completed step: 4, seconds: 0.054, TFLOP/s/device: 62.891, Tokens/s/device: 9479.550, total_weights: 3736, loss: 9.172 I0416 13:04:07.660640 137380359089280 metric_logger.py:185] completed step: 5, seconds: 0.051, TFLOP/s/device: 66.864, Tokens/s/device: 10078.542, total_weights: 4096, loss: 9.075 I0416 13:04:07.712321 137380359089280 metric_logger.py:185] completed step: 6, seconds: 10.297, TFLOP/s/device: 0.330, Tokens/s/device: 49.725, total_weights: 2931, loss: 8.899 I0416 13:04:07.764594 137380359089280 metric_logger.py:185] completed step: 7, seconds: 0.008, TFLOP/s/device: 409.152, Tokens/s/device: 61671.886, total_weights: 3526, loss: 8.818 I0416 13:04:07.816915 137380359089280 metric_logger.py:185] completed step: 8, seconds: 0.052, TFLOP/s/device: 64.910, Tokens/s/device: 9783.876, total_weights: 3012, loss: 8.750 I0416 13:04:07.869087 137380359089280 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0416 13:04:07.869891 137380359089280 checkpointing.py:776] Waited 0.0008203983306884766 seconds for step 9 to finish before starting checkpointing. I0416 13:04:07.870159 137380359089280 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:04:27.487843 137267106547264 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 30.667375s (commit=30.217569s, array_metadata_write=0.449806s) I0416 13:04:27.488803 137267085575744 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 392.813 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 32 seconds) (per-host) I0416 13:04:27.488907 137267085575744 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 30.667668s. I0416 13:04:27.718297 137267085575744 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776344636100653907, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0416 13:04:27.909159 137267085575744 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 13:04:28.097908 137267169461824 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776344636100653907, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0416 13:04:28.319633 137267085575744 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 13:04:28.320314 137267085575744 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.559222s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0/items I0416 13:04:28.320890 137267085575744 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0/items I0416 13:04:28.559153 137267085575744 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0 I0416 13:04:29.215444 137267085575744 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0`. I0416 13:04:29.216089 137267085575744 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 33.992125s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/0 I0416 13:04:29.216161 137267085575744 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 32.394924s. I0416 13:04:29.216335 137267230279232 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:04:29.216443 137267230279232 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:04:29.216501 137267230279232 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:04:29.216546 137267230279232 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 13:04:29.216691 137380359089280 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0416 13:04:29.216817 137380359089280 checkpoint_manager.py:1441] Waiting for previous save to complete took 21.346663 seconds. If this number is high, consider checkpointing less frequently. I0416 13:04:29.217552 137380359089280 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 13:04:29.217846 137380359089280 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9. I0416 13:04:29.395266 137380359089280 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:04:29.395357 137380359089280 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 13:04:29.396851 137267230279232 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9 I0416 13:04:30.093434 137267075089984 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9/items I0416 13:04:41.569352 137380359089280 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 12.174505s I0416 13:04:41.573550 137380359089280 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.006 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 12 seconds) (per-host) I0416 13:04:41.573623 137380359089280 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 12.266462s (batch_requests_ready=0.084852s, total_serialization_initiated=12.177510s, others=0.004100s) I0416 13:04:41.573705 137380359089280 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 12.266982s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=12.266963) I0416 13:04:41.575911 137267138004544 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:04:41.576011 137380359089280 async_checkpointer.py:561] Finished blocking save. Time taken: 12.358412s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9. I0416 13:04:41.576221 137380359089280 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:04:41.576421 137267117033024 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:04:41.576493 137380359089280 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776344647.8701255, 'wait_for_prev_duration_secs': 21.34666347503662, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776344669.217578, 'checkpointer_blocking_duration_secs': 12.358553171157837, 'get_old_steps_start_time': 1776344681.5761492, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776344647.8700855, 'checkpoint_manager_blocking_duration_secs': 33.70638155937195} I0416 13:04:41.576610 137380359089280 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0416 13:04:41.576642 137380359089280 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:04:42.344505 137267209307712 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_0 I0416 13:05:18.783917 137267106547264 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.209745s (commit=36.750809s, array_metadata_write=0.458937s) I0416 13:05:18.785023 137267138004544 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 255.403 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 49 seconds) (per-host) I0416 13:05:18.785147 137267138004544 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.209075s. I0416 13:05:19.219103 137267138004544 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 13:05:19.633182 137267138004544 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 13:05:19.633859 137267138004544 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.562642s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9/items I0416 13:05:19.634515 137267138004544 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9/items I0416 13:05:19.882588 137267138004544 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9 I0416 13:05:20.569414 137267138004544 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9`. I0416 13:05:20.570067 137267138004544 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 51.352458s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_09_pdb_lt_1/checkpoints/9 I0416 13:05:20.570157 137267138004544 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 38.994092s. I0416 13:05:20.570268 137267117033024 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:05:20.570319 137267117033024 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:05:20.570382 137267117033024 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:05:20.570428 137267117033024 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 13:05:20.570588 137380359089280 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 13:05:20.570719 137380359089280 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:05:20.571304 137380359089280 metric_logger.py:185] completed step: 9, seconds: 0.052, TFLOP/s/device: 64.855, Tokens/s/device: 9775.656, total_weights: 3004, loss: 8.888 Per train step: Total TFLOPs: 3.40 split as 93.93% learnable weight flops and 6.07% attention flops
2026-04-16 18:43:02.543612: 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 18:43:02.660134 136332058107008 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 18:44:00.190378 136332058107008 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 18:44:00.190482 136332058107008 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 18:44:00.190518 136332058107008 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 18:44:00.190544 136332058107008 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 18:44:00.190630 136332058107008 train.py:811] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 18:44:00.284291 3997044 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:44:00.678778 136332058107008 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 2, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0416 18:44:00.678950 136332058107008 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 2, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0416 18:44:00.679188 136332058107008 checkpointing.py:688] Setting up checkpoint logger... I0416 18:44:00.679230 136332058107008 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 18:44:00.679265 136332058107008 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 18:44:00.679913 136332058107008 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 0x7bfdb19fd070>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 18:44:03.123079 136332058107008 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 18:44:03.123340 136332058107008 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 0x7bf735f3a150>}, handler_registry=None I0416 18:44:03.123632 136332058107008 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bf735f3a150>` 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 18:44:03.123672 136332058107008 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bf735f3a720>` 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 18:44:03.123701 136332058107008 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 0x7bf735f3a150>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bf735f3a150>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bf735f3a720>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bf735f3a720>}). I0416 18:44:03.124027 136332058107008 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 18:44:03.124097 136332058107008 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7bf735f31ee0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 18:44:03.254639 136332058107008 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints I0416 18:44:03.254859 136332058107008 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_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7bf735f3a5a0> I0416 18:44:03.254938 136332058107008 checkpointing.py:302] Checkpoint manager created! I0416 18:44:03.341766 136332058107008 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 18:44:03.345236 136332058107008 reader.py:262] Creating a tf.data.Dataset reading 8 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0. I0416 18:44:03.398458 136332058107008 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split train, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 18:44:03.429618 136332058107008 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 18:44:03.429686 136332058107008 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 18:44:04.404829 136332058107008 nnx_wrappers.py:437] Unknown Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 18:44:04.404927 136332058107008 nnx_wrappers.py:437] Unknown Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0416 18:44:04.507251 136332058107008 attentions.py:1088] attentions/inputs_q Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 18:44:04.507327 136332058107008 attentions.py:1088] attentions/inputs_q Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0416 18:44:04.522360 136332058107008 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 18:44:04.522409 136332058107008 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0416 18:44:04.543618 136332058107008 attentions.py:1154] attentions/query Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 18:44:04.543672 136332058107008 attentions.py:1154] attentions/query Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0416 18:44:04.558862 136332058107008 attentions.py:1155] attentions/key Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 18:44:04.558912 136332058107008 attentions.py:1155] attentions/key Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0416 18:44:04.573867 136332058107008 attentions.py:1156] attentions/value Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 18:44:04.573915 136332058107008 attentions.py:1156] attentions/value Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0416 18:44:04.597491 136332058107008 attentions.py:1197] attentions/out Logical: bfloat16[2,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 18:44:04.597548 136332058107008 attentions.py:1197] attentions/out Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0416 18:44:04.616582 136332058107008 linears.py:525] linears/x Logical: bfloat16[2,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 18:44:04.616643 136332058107008 linears.py:525] linears/x Physical: bfloat16[2,2048,7168]....................................... ('fsdp', None, 'tensor'). I0416 18:44:04.808384 136332058107008 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 18:44:04.808473 136332058107008 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0416 18:44:04.859535 3997044 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: 2 tensor: 4 I0416 18:44:05.250823 136332058107008 maxtext_utils.py:1796] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: ('tensor',) I0416 18:44:05.250928 136332058107008 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, 'tensor') I0416 18:44:05.250970 136332058107008 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, 'tensor') I0416 18:44:05.251013 136332058107008 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0416 18:44:05.251067 136332058107008 maxtext_utils.py:1796] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0416 18:44:05.251098 136332058107008 maxtext_utils.py:1796] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0416 18:44:05.251144 136332058107008 maxtext_utils.py:1796] 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, 'tensor', None) I0416 18:44:05.251183 136332058107008 maxtext_utils.py:1796] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: ('tensor', None, None, 'fsdp') I0416 18:44:05.251212 136332058107008 maxtext_utils.py:1796] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, 'tensor', None) I0416 18:44:05.251239 136332058107008 maxtext_utils.py:1796] 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, 'tensor', None) I0416 18:44:05.251273 136332058107008 maxtext_utils.py:1796] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0416 18:44:05.251307 136332058107008 maxtext_utils.py:1796] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0416 18:44:05.683998 136332058107008 train.py:157] train/xent Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 18:44:05.684088 136332058107008 train.py:157] train/xent Physical: float32[2,2048]............................................. ('fsdp', None). I0416 18:44:05.698789 136332058107008 train.py:164] train/z_loss Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 18:44:05.698841 136332058107008 train.py:164] train/z_loss Physical: float32[2,2048]............................................. ('fsdp', None). W0416 18:44:05.979427 3997044 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:44:06.060323 136332058107008 max_utils.py:791] Total memory size: 2.5 GB, Output size: 1.5 GB, Temp size: 0.9 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0416 18:44:06.060795 136332058107008 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 18:44:06.060936 136332058107008 metric_logger.py:289] number parameters: 1.104 billion W0416 18:44:09.239546 3997044 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:44:09.336125 136332058107008 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 18:44:09.391359 136332058107008 checkpointing.py:798] Waited 0.05523395538330078 seconds for step 0 to finish before starting checkpointing. I0416 18:44:09.391740 136332058107008 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 18:44:09.391888 136332058107008 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 18:44:09.392258 136332058107008 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0. I0416 18:44:09.483454 136332058107008 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 18:44:09.566662 136218794788416 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0 I0416 18:44:09.587057 136332058107008 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 18:44:09.587117 136332058107008 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 18:44:10.316492 136218784302656 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0/items I0416 18:44:10.481437 136218736068160 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776365050236214413, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776365050236214413, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0416 18:44:10.543549 136332058107008 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.956865s I0416 18:44:10.543956 136332058107008 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 11.652 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 18:44:10.544008 136332058107008 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.059218s (batch_requests_ready=0.098659s, total_serialization_initiated=0.960247s, others=0.000313s) I0416 18:44:10.544075 136332058107008 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.060172s (all_items=0.000024s, per_item={'items': '0.00002360'}, temp_paths=1.060148) I0416 18:44:10.544825 136218715096640 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 18:44:10.544919 136332058107008 async_checkpointer.py:561] Finished blocking save. Time taken: 1.152993s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0. I0416 18:44:10.545108 136332058107008 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 18:44:10.545355 136218805274176 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 18:44:10.545462 136332058107008 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776365049.3917274, 'wait_for_prev_duration_secs': 4.0531158447265625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776365049.3919091, 'checkpointer_blocking_duration_secs': 1.1531190872192383, 'get_old_steps_start_time': 1776365050.5450566, 'get_old_steps_duration_secs': 1.9550323486328125e-05, 'checkpoint_manager_blocking_start_time': 1776365049.3916464, 'checkpoint_manager_blocking_duration_secs': 1.1537883281707764} I0416 18:44:10.545567 136332058107008 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 18:44:10.545605 136332058107008 max_utils.py:750] Memstats: After params initialized: I0416 18:44:10.545666 136332058107008 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_0(process=0,(0,0,0,0)) I0416 18:44:10.545690 136332058107008 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_1(process=0,(1,0,0,0)) I0416 18:44:10.545711 136332058107008 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_2(process=0,(0,1,0,0)) I0416 18:44:10.545729 136332058107008 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_3(process=0,(1,1,0,0)) I0416 18:44:10.545746 136332058107008 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_4(process=0,(0,2,0,0)) I0416 18:44:10.545763 136332058107008 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_5(process=0,(1,2,0,0)) I0416 18:44:10.545779 136332058107008 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_6(process=0,(0,3,0,0)) I0416 18:44:10.545796 136332058107008 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_7(process=0,(1,3,0,0)) W0416 18:44:10.551414 3997044 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:44:10.556872 3997044 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:44:10.561565 3997044 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:44:10.884670 3999685 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 18:44:10.884719 3999685 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 18:44:10.912627 136332058107008 metric_logger.py:185] completed step: 0, seconds: 3.275, TFLOP/s/device: 1.037, Tokens/s/device: 156.349, total_weights: 3650, loss: 10.896 I0416 18:44:10.913497 136332058107008 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/tensorboard/' I0416 18:44:10.960813 136332058107008 metric_logger.py:185] completed step: 1, seconds: 1.572, TFLOP/s/device: 2.160, Tokens/s/device: 325.649, total_weights: 2676, loss: 10.833 I0416 18:44:11.013301 136332058107008 metric_logger.py:185] completed step: 2, seconds: 0.015, TFLOP/s/device: 229.466, Tokens/s/device: 34587.584, total_weights: 3783, loss: 10.136 I0416 18:44:11.065514 136332058107008 metric_logger.py:185] completed step: 3, seconds: 0.048, TFLOP/s/device: 70.041, Tokens/s/device: 10557.354, total_weights: 3680, loss: 9.403 I0416 18:44:11.645498 136218773816896 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_0 I0416 18:44:21.459486 136332058107008 metric_logger.py:185] completed step: 4, seconds: 0.053, TFLOP/s/device: 64.350, Tokens/s/device: 9699.542, total_weights: 3736, loss: 9.172 I0416 18:44:21.469690 136332058107008 metric_logger.py:185] completed step: 5, seconds: 0.052, TFLOP/s/device: 65.422, Tokens/s/device: 9861.135, total_weights: 4096, loss: 9.075 I0416 18:44:21.521530 136332058107008 metric_logger.py:185] completed step: 6, seconds: 10.393, TFLOP/s/device: 0.327, Tokens/s/device: 49.266, total_weights: 2931, loss: 8.899 I0416 18:44:21.573804 136332058107008 metric_logger.py:185] completed step: 7, seconds: 0.008, TFLOP/s/device: 423.750, Tokens/s/device: 63872.255, total_weights: 3526, loss: 8.818 I0416 18:44:21.626106 136332058107008 metric_logger.py:185] completed step: 8, seconds: 0.052, TFLOP/s/device: 65.089, Tokens/s/device: 9810.872, total_weights: 3012, loss: 8.750 I0416 18:44:21.678318 136332058107008 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0416 18:44:21.678997 136332058107008 checkpointing.py:798] Waited 0.0006995201110839844 seconds for step 9 to finish before starting checkpointing. I0416 18:44:21.679257 136332058107008 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 18:44:42.947856 136218794788416 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.069894s (commit=31.646214s, array_metadata_write=0.423680s) I0416 18:44:42.948904 136218715096640 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 377.623 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 33 seconds) (per-host) I0416 18:44:42.949013 136218715096640 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.404050s. I0416 18:44:43.172515 136218715096640 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776365050236214413, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0416 18:44:43.344687 136218715096640 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 18:44:43.556695 136218736068160 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776365050236214413, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0416 18:44:43.784199 136218715096640 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 18:44:43.784802 136218715096640 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.580463s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0/items I0416 18:44:43.785459 136218715096640 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0/items I0416 18:44:44.033249 136218715096640 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0 I0416 18:44:44.710835 136218715096640 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0`. I0416 18:44:44.711479 136218715096640 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 35.319558s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/0 I0416 18:44:44.711553 136218715096640 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.166591s. I0416 18:44:44.711734 136218805274176 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 18:44:44.711842 136218805274176 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 18:44:44.711902 136218805274176 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 18:44:44.711949 136218805274176 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 18:44:44.712086 136332058107008 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0416 18:44:44.712201 136332058107008 checkpoint_manager.py:1441] Waiting for previous save to complete took 23.032946 seconds. If this number is high, consider checkpointing less frequently. I0416 18:44:44.712896 136332058107008 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 18:44:44.713185 136332058107008 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9. I0416 18:44:44.883402 136218805274176 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9 I0416 18:44:44.906826 136332058107008 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 18:44:44.906915 136332058107008 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 18:44:45.562036 136218603947584 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9/items I0416 18:44:54.046617 136332058107008 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.140181s I0416 18:44:54.051227 136332058107008 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.335 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0416 18:44:54.051288 136332058107008 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.243113s (batch_requests_ready=0.095478s, total_serialization_initiated=9.143130s, others=0.004505s) I0416 18:44:54.051359 136332058107008 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.243632s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=9.243614) I0416 18:44:54.052826 136218794788416 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 18:44:54.052944 136332058107008 async_checkpointer.py:561] Finished blocking save. Time taken: 9.340003s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9. I0416 18:44:54.053208 136332058107008 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 18:44:54.053367 136218715096640 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 18:44:54.053439 136332058107008 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776365061.6792257, 'wait_for_prev_duration_secs': 23.03294610977173, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776365084.7129233, 'checkpointer_blocking_duration_secs': 9.340198278427124, 'get_old_steps_start_time': 1776365094.0531366, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776365061.679188, 'checkpoint_manager_blocking_duration_secs': 32.374226093292236} I0416 18:44:54.053555 136332058107008 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0416 18:44:54.053589 136332058107008 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 18:44:54.814871 136218773816896 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_0 I0416 18:45:29.932538 136218805274176 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 35.880603s (commit=35.379782s, array_metadata_write=0.500821s) I0416 18:45:29.933547 136218794788416 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 280.038 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 45 seconds) (per-host) I0416 18:45:29.933654 136218794788416 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.880665s. I0416 18:45:30.364685 136218794788416 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 18:45:30.795023 136218794788416 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 18:45:30.795687 136218794788416 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.575538s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9/items I0416 18:45:30.796329 136218794788416 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9/items I0416 18:45:31.048069 136218794788416 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9 I0416 18:45:31.727058 136218794788416 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9`. I0416 18:45:31.727689 136218794788416 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 47.014744s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_09_pdb_lt_1/checkpoints/9 I0416 18:45:31.727766 136218794788416 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 37.674779s. I0416 18:45:31.727868 136218715096640 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 18:45:31.727916 136218715096640 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 18:45:31.727962 136218715096640 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 18:45:31.728004 136218715096640 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 18:45:31.728159 136332058107008 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 18:45:31.728277 136332058107008 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 18:45:31.728888 136332058107008 metric_logger.py:185] completed step: 9, seconds: 0.052, TFLOP/s/device: 65.156, Tokens/s/device: 9821.035, total_weights: 3004, loss: 8.888 Per train step: Total TFLOPs: 3.40 split as 93.93% learnable weight flops and 6.07% attention flops