main (51c7f2b8e) vs feat/migrate-nnx-utils (dc4d58891)| Metric | main 51c7f2b8e | feat/migrate-nnx-utils dc4d58891 | Diff (feat/migrate-nnx-utils − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.8880 | 8.8880 | 0 |
| TFLOP/s | 64.849 | 65.327 | +0.478 |
| Tok/s | 9774.7 | 9846.7 | +71.999 |
| Avg s/step | 1.240 | 1.251 | +0.011 |
| Memory % | 5.06 | 5.06 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-15 21:35:42.461330: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0415 21:35:42.577404 136723942116480 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 21:36:32.796548 136723942116480 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 21:36:32.796657 136723942116480 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 21:36:32.796690 136723942116480 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0415 21:36:32.796716 136723942116480 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 21:36:32.796804 136723942116480 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 21:36:32.890174 3252062 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:36:33.278016 136723942116480 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 2, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0415 21:36:33.278342 136723942116480 checkpointing.py:677] Setting up checkpoint logger... I0415 21:36:33.278384 136723942116480 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 21:36:33.278425 136723942116480 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 21:36:33.279114 136723942116480 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 0x7c58efbf8b30>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 21:36:35.707122 136723942116480 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 21:36:35.707297 136723942116480 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 0x7c52741cba70>}, handler_registry=None I0415 21:36:35.707557 136723942116480 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c52741cba70>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0415 21:36:35.707600 136723942116480 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c5272f3a180>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0415 21:36:35.707638 136723942116480 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 0x7c52741cba70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c52741cba70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c5272f3a180>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c5272f3a180>}). I0415 21:36:35.707941 136723942116480 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 21:36:35.707994 136723942116480 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7c5272f31260> timeout: 600 secs and primary_host=0 for async checkpoint writes I0415 21:36:35.841021 136723942116480 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints I0415 21:36:35.841270 136723942116480 checkpoint_manager.py:921] [process=0][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_hns=False, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False), root_directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c5272f39340> I0415 21:36:35.841349 136723942116480 checkpointing.py:301] Checkpoint manager created! I0415 21:36:35.922290 136723942116480 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0415 21:36:35.925250 136723942116480 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. I0415 21:36:35.978551 136723942116480 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 I0415 21:36:36.009250 136723942116480 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:36:36.009315 136723942116480 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:36:36.978237 136723942116480 nnx_wrappers.py:437] Unknown Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 21:36:36.978341 136723942116480 nnx_wrappers.py:437] Unknown Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0415 21:36:37.079700 136723942116480 attentions.py:1088] attentions/inputs_q Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:36:37.079778 136723942116480 attentions.py:1088] attentions/inputs_q Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0415 21:36:37.094830 136723942116480 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:36:37.094884 136723942116480 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0415 21:36:37.117020 136723942116480 attentions.py:1154] attentions/query Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:36:37.117090 136723942116480 attentions.py:1154] attentions/query Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0415 21:36:37.132097 136723942116480 attentions.py:1155] attentions/key Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:36:37.132148 136723942116480 attentions.py:1155] attentions/key Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0415 21:36:37.147191 136723942116480 attentions.py:1156] attentions/value Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:36:37.147240 136723942116480 attentions.py:1156] attentions/value Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0415 21:36:37.169405 136723942116480 attentions.py:1197] attentions/out Logical: bfloat16[2,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 21:36:37.169462 136723942116480 attentions.py:1197] attentions/out Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0415 21:36:37.188530 136723942116480 linears.py:525] linears/x Logical: bfloat16[2,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 21:36:37.188582 136723942116480 linears.py:525] linears/x Physical: bfloat16[2,2048,7168]....................................... ('fsdp', None, 'tensor'). I0415 21:36:37.375308 136723942116480 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 21:36:37.375400 136723942116480 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0415 21:36:37.426229 3252062 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 I0415 21:36:37.815764 136723942116480 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: ('tensor',) I0415 21:36:37.815875 136723942116480 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') I0415 21:36:37.815919 136723942116480 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') I0415 21:36:37.815963 136723942116480 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') I0415 21:36:37.816003 136723942116480 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) I0415 21:36:37.816031 136723942116480 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) I0415 21:36:37.816090 136723942116480 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) I0415 21:36:37.816131 136723942116480 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') I0415 21:36:37.816159 136723942116480 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) I0415 21:36:37.816184 136723942116480 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) I0415 21:36:37.816220 136723942116480 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0415 21:36:37.816253 136723942116480 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0415 21:36:38.266980 136723942116480 train.py:155] train/xent Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:36:38.267074 136723942116480 train.py:155] train/xent Physical: float32[2,2048]............................................. ('fsdp', None). I0415 21:36:38.281403 136723942116480 train.py:162] train/z_loss Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:36:38.281454 136723942116480 train.py:162] train/z_loss Physical: float32[2,2048]............................................. ('fsdp', None). W0415 21:36:38.539004 3252062 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:36:38.618527 136723942116480 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. I0415 21:36:38.619020 136723942116480 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 21:36:38.619179 136723942116480 metric_logger.py:289] number parameters: 1.104 billion W0415 21:36:41.690449 3252062 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:36:41.783440 136723942116480 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0415 21:36:41.838332 136723942116480 checkpointing.py:776] Waited 0.054889678955078125 seconds for step 0 to finish before starting checkpointing. I0415 21:36:41.838722 136723942116480 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:36:41.838867 136723942116480 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0415 21:36:41.839232 136723942116480 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0. I0415 21:36:41.927323 136723942116480 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0415 21:36:42.020994 136723942116480 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:36:42.021095 136723942116480 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0415 21:36:42.026805 136610792343104 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0 I0415 21:36:42.649179 136723942116480 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.628629s I0415 21:36:42.649507 136723942116480 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 17.111 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 721 milliseconds) (per-host) I0415 21:36:42.649556 136723942116480 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.721314s (batch_requests_ready=0.088900s, total_serialization_initiated=0.632182s, others=0.000232s) I0415 21:36:42.649614 136723942116480 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.721845s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=0.721826) I0415 21:36:42.650310 136610708457024 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:36:42.650402 136723942116480 async_checkpointer.py:561] Finished blocking save. Time taken: 0.811495s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0. I0415 21:36:42.650566 136723942116480 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:36:42.650726 136610697971264 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:36:42.650802 136723942116480 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776289001.8387094, 'wait_for_prev_duration_secs': 4.0531158447265625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776289001.838888, 'checkpointer_blocking_duration_secs': 0.8116090297698975, 'get_old_steps_start_time': 1776289002.6505125, 'get_old_steps_duration_secs': 1.9788742065429688e-05, 'checkpoint_manager_blocking_start_time': 1776289001.8386266, 'checkpoint_manager_blocking_duration_secs': 0.8121464252471924} I0415 21:36:42.650877 136723942116480 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0415 21:36:42.650914 136723942116480 max_utils.py:750] Memstats: After params initialized: I0415 21:36:42.650994 136723942116480 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_0(process=0,(0,0,0,0)) I0415 21:36:42.651021 136723942116480 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_1(process=0,(1,0,0,0)) I0415 21:36:42.651054 136723942116480 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_2(process=0,(0,1,0,0)) I0415 21:36:42.651073 136723942116480 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_3(process=0,(1,1,0,0)) I0415 21:36:42.651089 136723942116480 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_4(process=0,(0,2,0,0)) I0415 21:36:42.651106 136723942116480 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_5(process=0,(1,2,0,0)) I0415 21:36:42.651123 136723942116480 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_6(process=0,(0,3,0,0)) I0415 21:36:42.651139 136723942116480 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_7(process=0,(1,3,0,0)) W0415 21:36:42.656929 3252062 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:36:42.662665 3252062 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:36:42.667383 3252062 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:36:42.777347 136610781857344 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0/items I0415 21:36:42.992145 136723942116480 metric_logger.py:185] completed step: 0, seconds: 3.164, TFLOP/s/device: 1.074, Tokens/s/device: 161.831, total_weights: 3650, loss: 10.896 I0415 21:36:42.993277 136723942116480 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/tensorboard/' I0415 21:36:43.041414 136723942116480 metric_logger.py:185] completed step: 1, seconds: 1.206, TFLOP/s/device: 2.817, Tokens/s/device: 424.661, total_weights: 2676, loss: 10.833 I0415 21:36:43.093742 136723942116480 metric_logger.py:185] completed step: 2, seconds: 0.017, TFLOP/s/device: 201.925, Tokens/s/device: 30436.333, total_weights: 3783, loss: 10.136 I0415 21:36:43.120266 136610802828864 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289002677440451, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776289002677440451, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0415 21:36:43.146115 136723942116480 metric_logger.py:185] completed step: 3, seconds: 0.048, TFLOP/s/device: 70.600, Tokens/s/device: 10641.615, total_weights: 3680, loss: 9.403 I0415 21:36:43.380601 3254713 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 21:36:43.380650 3254713 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 21:36:44.141205 136610729428544 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_0 I0415 21:36:52.820562 136723942116480 metric_logger.py:185] completed step: 4, seconds: 0.053, TFLOP/s/device: 64.345, Tokens/s/device: 9698.807, total_weights: 3736, loss: 9.172 I0415 21:36:52.831395 136723942116480 metric_logger.py:185] completed step: 5, seconds: 0.052, TFLOP/s/device: 64.829, Tokens/s/device: 9771.738, total_weights: 4096, loss: 9.075 I0415 21:36:52.883026 136723942116480 metric_logger.py:185] completed step: 6, seconds: 9.671, TFLOP/s/device: 0.351, Tokens/s/device: 52.940, total_weights: 2931, loss: 8.899 I0415 21:36:52.935407 136723942116480 metric_logger.py:185] completed step: 7, seconds: 0.008, TFLOP/s/device: 400.375, Tokens/s/device: 60348.892, total_weights: 3526, loss: 8.818 I0415 21:36:52.987775 136723942116480 metric_logger.py:185] completed step: 8, seconds: 0.052, TFLOP/s/device: 65.224, Tokens/s/device: 9831.218, total_weights: 3012, loss: 8.750 I0415 21:36:53.039868 136723942116480 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0415 21:36:53.040639 136723942116480 checkpointing.py:776] Waited 0.0007901191711425781 seconds for step 9 to finish before starting checkpointing. I0415 21:36:53.040900 136723942116480 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:37:10.027270 136610718942784 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 26.650609s (commit=26.191942s, array_metadata_write=0.458668s) I0415 21:37:10.028199 136610708457024 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 449.709 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 28 seconds) (per-host) I0415 21:37:10.028251 136610708457024 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 27.377806s. I0415 21:37:10.244028 136610708457024 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289002677440451, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0415 21:37:10.425587 136610708457024 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 21:37:10.625324 136610802828864 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776289002677440451, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0415 21:37:10.866847 136610708457024 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:37:10.867455 136610708457024 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.582495s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0/items I0415 21:37:10.868114 136610708457024 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0/items I0415 21:37:11.090713 136610708457024 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0 I0415 21:37:11.729829 136610708457024 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0`. I0415 21:37:11.730477 136610708457024 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 29.891577s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/0 I0415 21:37:11.730546 136610708457024 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 29.080101s. I0415 21:37:11.730661 136610697971264 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:37:11.730711 136610697971264 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:37:11.730760 136610697971264 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:37:11.730797 136610697971264 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0415 21:37:11.730880 136723942116480 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. W0415 21:37:11.730940 136723942116480 checkpoint_manager.py:1441] Waiting for previous save to complete took 18.690056 seconds. If this number is high, consider checkpointing less frequently. I0415 21:37:11.731466 136723942116480 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0415 21:37:11.731709 136723942116480 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9. I0415 21:37:11.904249 136723942116480 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 21:37:11.904338 136723942116480 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0415 21:37:11.922319 136610697971264 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9 I0415 21:37:12.583598 136610582627904 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9/items I0415 21:37:25.884905 136723942116480 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 13.981051s I0415 21:37:25.885522 136723942116480 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 898.414 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 14 seconds) (per-host) I0415 21:37:25.885582 136723942116480 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 14.065752s (batch_requests_ready=0.081023s, total_serialization_initiated=13.984277s, others=0.000453s) I0415 21:37:25.885654 136723942116480 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 14.066247s (all_items=0.000015s, per_item={'items': '0.00001550'}, temp_paths=14.066231) I0415 21:37:25.890900 136610687485504 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:37:25.891105 136723942116480 async_checkpointer.py:561] Finished blocking save. Time taken: 14.159596s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9. I0415 21:37:25.891294 136723942116480 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:37:25.891446 136610708457024 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:37:25.891534 136723942116480 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776289013.0408635, 'wait_for_prev_duration_secs': 18.69005584716797, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776289031.73149, 'checkpointer_blocking_duration_secs': 14.159701824188232, 'get_old_steps_start_time': 1776289045.8912182, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776289013.0408177, 'checkpoint_manager_blocking_duration_secs': 32.85068678855896} I0415 21:37:25.891628 136723942116480 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0415 21:37:25.891662 136723942116480 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 21:37:26.645800 136610729428544 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_0 I0415 21:37:57.028135 136610582627904 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 31.138074s (commit=30.676322s, array_metadata_write=0.461752s) I0415 21:37:57.029182 136610687485504 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 279.517 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 45 seconds) (per-host) I0415 21:37:57.029281 136610687485504 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 31.138216s. I0415 21:37:57.450059 136610687485504 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 21:37:57.884508 136610687485504 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 21:37:57.884680 136610687485504 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.577463s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9/items I0415 21:37:57.885323 136610687485504 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9/items I0415 21:37:58.137000 136610687485504 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9 I0415 21:37:58.845532 136610687485504 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9`. I0415 21:37:58.846156 136610687485504 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 47.114655s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_09_pdb_lt_1/checkpoints/9 I0415 21:37:58.846227 136610687485504 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 32.955164s. I0415 21:37:58.846328 136610708457024 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:37:58.846376 136610708457024 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 21:37:58.846430 136610708457024 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:37:58.846467 136610708457024 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0415 21:37:58.846550 136723942116480 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0415 21:37:58.846617 136723942116480 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:37:58.847394 136723942116480 metric_logger.py:185] completed step: 9, seconds: 0.052, TFLOP/s/device: 64.849, Tokens/s/device: 9774.723, 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
2026-04-15 23:35:11.870113: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0415 23:35:11.987529 136961595919488 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 23:36:10.043650 136961595919488 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 23:36:10.043757 136961595919488 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 23:36:10.043791 136961595919488 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0415 23:36:10.043815 136961595919488 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 23:36:10.043902 136961595919488 train.py:715] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 23:36:10.135643 3379399 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:36:10.521097 136961595919488 maxtext_utils.py:1520] Num_devices: 8, shape (1, 1, 1, 2, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0415 23:36:10.521381 136961595919488 checkpointing.py:677] Setting up checkpoint logger... I0415 23:36:10.521427 136961595919488 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 23:36:10.521467 136961595919488 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 23:36:10.521980 136961595919488 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 0x7c90445fbc50>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 23:36:12.927744 136961595919488 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 23:36:12.927925 136961595919488 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 0x7c89c8332180>}, handler_registry=None I0415 23:36:12.928207 136961595919488 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c89c8332180>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0415 23:36:12.928248 136961595919488 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c89c8332690>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0415 23:36:12.928277 136961595919488 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 0x7c89c8332180>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c89c8332180>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c89c8332690>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c89c8332690>}). I0415 23:36:12.928823 136961595919488 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 23:36:12.928878 136961595919488 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7c89c83354e0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0415 23:36:13.080713 136961595919488 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints I0415 23:36:13.080927 136961595919488 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_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c89c8331910> I0415 23:36:13.081007 136961595919488 checkpointing.py:301] Checkpoint manager created! I0415 23:36:13.164489 136961595919488 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0415 23:36:13.167519 136961595919488 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. I0415 23:36:13.222118 136961595919488 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 I0415 23:36:13.254020 136961595919488 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 23:36:13.254101 136961595919488 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 23:36:14.239880 136961595919488 nnx_wrappers.py:437] Unknown Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 23:36:14.239982 136961595919488 nnx_wrappers.py:437] Unknown Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0415 23:36:14.342842 136961595919488 attentions.py:1088] attentions/inputs_q Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 23:36:14.342922 136961595919488 attentions.py:1088] attentions/inputs_q Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0415 23:36:14.358077 136961595919488 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[2,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 23:36:14.358125 136961595919488 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[2,2048,2048]....................................... ('fsdp', None, 'tensor'). I0415 23:36:14.380438 136961595919488 attentions.py:1154] attentions/query Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 23:36:14.380494 136961595919488 attentions.py:1154] attentions/query Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0415 23:36:14.395571 136961595919488 attentions.py:1155] attentions/key Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 23:36:14.395626 136961595919488 attentions.py:1155] attentions/key Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0415 23:36:14.410706 136961595919488 attentions.py:1156] attentions/value Logical: bfloat16[2,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 23:36:14.410755 136961595919488 attentions.py:1156] attentions/value Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0415 23:36:14.433431 136961595919488 attentions.py:1197] attentions/out Logical: bfloat16[2,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 23:36:14.433491 136961595919488 attentions.py:1197] attentions/out Physical: bfloat16[2,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0415 23:36:14.453091 136961595919488 linears.py:525] linears/x Logical: bfloat16[2,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 23:36:14.453141 136961595919488 linears.py:525] linears/x Physical: bfloat16[2,2048,7168]....................................... ('fsdp', None, 'tensor'). I0415 23:36:14.645850 136961595919488 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 23:36:14.645944 136961595919488 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0415 23:36:14.698159 3379399 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 I0415 23:36:15.096786 136961595919488 maxtext_utils.py:1623] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: ('tensor',) I0415 23:36:15.096893 136961595919488 maxtext_utils.py:1623] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, 'tensor') I0415 23:36:15.096935 136961595919488 maxtext_utils.py:1623] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, 'tensor') I0415 23:36:15.096980 136961595919488 maxtext_utils.py:1623] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0415 23:36:15.097018 136961595919488 maxtext_utils.py:1623] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0415 23:36:15.097167 136961595919488 maxtext_utils.py:1623] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0415 23:36:15.097219 136961595919488 maxtext_utils.py:1623] 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) I0415 23:36:15.097260 136961595919488 maxtext_utils.py:1623] 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') I0415 23:36:15.097289 136961595919488 maxtext_utils.py:1623] 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) I0415 23:36:15.097315 136961595919488 maxtext_utils.py:1623] 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) I0415 23:36:15.097350 136961595919488 maxtext_utils.py:1623] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed', 'vocab') Physical: ('fsdp', 'tensor') I0415 23:36:15.097384 136961595919488 maxtext_utils.py:1623] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed') Physical: ('tensor', 'fsdp') I0415 23:36:15.540149 136961595919488 train.py:155] train/xent Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 23:36:15.540232 136961595919488 train.py:155] train/xent Physical: float32[2,2048]............................................. ('fsdp', None). I0415 23:36:15.554813 136961595919488 train.py:162] train/z_loss Logical: float32[2,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 23:36:15.554864 136961595919488 train.py:162] train/z_loss Physical: float32[2,2048]............................................. ('fsdp', None). W0415 23:36:15.842172 3379399 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:36:15.925674 136961595919488 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. I0415 23:36:15.926165 136961595919488 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 23:36:15.926304 136961595919488 metric_logger.py:289] number parameters: 1.104 billion W0415 23:36:18.992200 3379399 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:36:19.086896 136961595919488 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0415 23:36:19.141844 136961595919488 checkpointing.py:776] Waited 0.05494189262390137 seconds for step 0 to finish before starting checkpointing. I0415 23:36:19.142248 136961595919488 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 23:36:19.142397 136961595919488 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0415 23:36:19.142779 136961595919488 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0. I0415 23:36:19.228010 136961595919488 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0415 23:36:19.316588 136848370304576 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0 I0415 23:36:19.321939 136961595919488 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 23:36:19.322026 136961595919488 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0415 23:36:20.061784 136961595919488 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.740276s I0415 23:36:20.062152 136961595919488 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 14.817 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 832 milliseconds) (per-host) I0415 23:36:20.062201 136961595919488 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.832978s (batch_requests_ready=0.088416s, total_serialization_initiated=0.744277s, others=0.000285s) I0415 23:36:20.062259 136961595919488 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.833791s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=0.833771) I0415 23:36:20.062945 136848290612800 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 23:36:20.063038 136961595919488 async_checkpointer.py:561] Finished blocking save. Time taken: 0.920602s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0. I0415 23:36:20.063229 136961595919488 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0415 23:36:20.063423 136961595919488 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776296179.1422358, 'wait_for_prev_duration_secs': 4.100799560546875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776296179.1424184, 'checkpointer_blocking_duration_secs': 0.9207406044006348, 'get_old_steps_start_time': 1776296180.063175, 'get_old_steps_duration_secs': 1.9311904907226562e-05, 'checkpoint_manager_blocking_start_time': 1776296179.142157, 'checkpoint_manager_blocking_duration_secs': 0.9212455749511719} I0415 23:36:20.063491 136961595919488 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0415 23:36:20.063529 136961595919488 max_utils.py:750] Memstats: After params initialized: I0415 23:36:20.063586 136961595919488 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_0(process=0,(0,0,0,0)) I0415 23:36:20.063618 136961595919488 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_1(process=0,(1,0,0,0)) I0415 23:36:20.063638 136961595919488 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_2(process=0,(0,1,0,0)) I0415 23:36:20.063655 136961595919488 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_3(process=0,(1,1,0,0)) I0415 23:36:20.063673 136961595919488 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_4(process=0,(0,2,0,0)) I0415 23:36:20.063692 136961595919488 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_5(process=0,(1,2,0,0)) I0415 23:36:20.063708 136961595919488 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_6(process=0,(0,3,0,0)) I0415 23:36:20.063724 136961595919488 max_utils.py:756] Using (GB) 1.58 / 31.25 (5.056000%) on TPU_7(process=0,(1,3,0,0)) W0415 23:36:20.069400 3379399 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:36:20.075453 3379399 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 23:36:20.080793 3379399 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 23:36:20.096342 136848359818816 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0/items I0415 23:36:20.266007 136848311584320 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776296180006856105, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776296180006856105, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0415 23:36:20.266261 136848280127040 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 23:36:20.444417 136961595919488 metric_logger.py:185] completed step: 0, seconds: 3.160, TFLOP/s/device: 1.075, Tokens/s/device: 162.020, total_weights: 3650, loss: 10.896 I0415 23:36:20.445419 136961595919488 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/tensorboard/' I0415 23:36:20.495222 136961595919488 metric_logger.py:185] completed step: 1, seconds: 1.356, TFLOP/s/device: 2.505, Tokens/s/device: 377.653, total_weights: 2676, loss: 10.833 I0415 23:36:20.547471 136961595919488 metric_logger.py:185] completed step: 2, seconds: 0.012, TFLOP/s/device: 271.917, Tokens/s/device: 40986.231, total_weights: 3783, loss: 10.136 I0415 23:36:20.599821 136961595919488 metric_logger.py:185] completed step: 3, seconds: 0.052, TFLOP/s/device: 65.660, Tokens/s/device: 9896.971, total_weights: 3680, loss: 9.403 I0415 23:36:20.785012 3382143 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 23:36:20.785090 3382143 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 23:36:21.587981 136848349333056 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_0 I0415 23:36:30.225569 136961595919488 metric_logger.py:185] completed step: 4, seconds: 0.052, TFLOP/s/device: 65.360, Tokens/s/device: 9851.838, total_weights: 3736, loss: 9.172 I0415 23:36:30.237065 136961595919488 metric_logger.py:185] completed step: 5, seconds: 0.052, TFLOP/s/device: 64.928, Tokens/s/device: 9786.681, total_weights: 4096, loss: 9.075 I0415 23:36:30.288743 136961595919488 metric_logger.py:185] completed step: 6, seconds: 9.625, TFLOP/s/device: 0.353, Tokens/s/device: 53.195, total_weights: 2931, loss: 8.899 I0415 23:36:30.340968 136961595919488 metric_logger.py:185] completed step: 7, seconds: 0.009, TFLOP/s/device: 390.749, Tokens/s/device: 58897.964, total_weights: 3526, loss: 8.818 I0415 23:36:30.393983 136961595919488 metric_logger.py:185] completed step: 8, seconds: 0.052, TFLOP/s/device: 64.911, Tokens/s/device: 9784.063, total_weights: 3012, loss: 8.750 I0415 23:36:30.445411 136961595919488 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0415 23:36:30.446152 136961595919488 checkpointing.py:776] Waited 0.0007486343383789062 seconds for step 9 to finish before starting checkpointing. I0415 23:36:30.446403 136961595919488 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 23:36:50.612989 136848301098560 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 29.825836s (commit=29.347470s, array_metadata_write=0.478367s) I0415 23:36:50.613959 136848290612800 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 402.642 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 31 seconds) (per-host) I0415 23:36:50.614070 136848290612800 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 30.550972s. I0415 23:36:50.827166 136848290612800 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776296180006856105, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0415 23:36:51.016629 136848290612800 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 23:36:51.226256 136848311584320 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776296180006856105, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0415 23:36:51.490336 136848290612800 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 23:36:51.490929 136848290612800 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.620378s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0/items I0415 23:36:51.491584 136848290612800 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0/items I0415 23:36:51.733316 136848290612800 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0 I0415 23:36:52.427495 136848290612800 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0`. I0415 23:36:52.428128 136848290612800 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 33.285697s. directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/0 I0415 23:36:52.428206 136848290612800 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 32.365110s. I0415 23:36:52.428397 136848280127040 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 23:36:52.428506 136848280127040 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 23:36:52.428563 136848280127040 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0415 23:36:52.428609 136848280127040 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0415 23:36:52.428752 136961595919488 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. W0415 23:36:52.428884 136961595919488 checkpoint_manager.py:1441] Waiting for previous save to complete took 21.982480 seconds. If this number is high, consider checkpointing less frequently. I0415 23:36:52.429900 136961595919488 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0415 23:36:52.430207 136961595919488 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9. I0415 23:36:52.586244 136848280127040 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9 I0415 23:36:52.598823 136961595919488 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0415 23:36:52.598914 136961595919488 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0415 23:36:53.308109 136847506277952 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9/items I0415 23:37:05.434248 136961595919488 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 12.835830s I0415 23:37:05.438425 136961595919488 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 977.187 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 12 seconds) (per-host) I0415 23:37:05.438484 136961595919488 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 12.931893s (batch_requests_ready=0.088967s, total_serialization_initiated=12.838852s, others=0.004073s) I0415 23:37:05.438553 136961595919488 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 12.932373s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=12.932356) I0415 23:37:05.439934 136848280127040 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 23:37:05.440014 136961595919488 async_checkpointer.py:561] Finished blocking save. Time taken: 13.010070s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9. I0415 23:37:05.440226 136961595919488 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0415 23:37:05.440399 136848290612800 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 23:37:05.440495 136961595919488 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776296190.44637, 'wait_for_prev_duration_secs': 21.9824800491333, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776296212.4299233, 'checkpointer_blocking_duration_secs': 13.01021432876587, 'get_old_steps_start_time': 1776296225.440153, 'get_old_steps_duration_secs': 3.409385681152344e-05, 'checkpoint_manager_blocking_start_time': 1776296190.4463308, 'checkpoint_manager_blocking_duration_secs': 34.994139671325684} I0415 23:37:05.440622 136961595919488 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0415 23:37:05.440656 136961595919488 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0415 23:37:06.212463 136848380790336 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_0 I0415 23:37:44.861939 136847506277952 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 39.422822s (commit=38.997643s, array_metadata_write=0.425179s) I0415 23:37:44.862947 136848280127040 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 241.361 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 52 seconds) (per-host) I0415 23:37:44.863000 136848280127040 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.422929s. I0415 23:37:45.277925 136848280127040 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0415 23:37:45.727428 136848280127040 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0415 23:37:45.727599 136848280127040 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.600296s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9/items I0415 23:37:45.728270 136848280127040 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9/items I0415 23:37:45.977106 136848280127040 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9 I0415 23:37:46.665423 136848280127040 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9`. I0415 23:37:46.666039 136848280127040 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 54.236102s. directory=gs://wanglance-maxtext/linen_ckpt_feat_migrate_nnx_utils_20260415_230805/linen_feat_migrate_nnx_utils_20260415_230805_09_pdb_lt_1/checkpoints/9 I0415 23:37:46.666129 136848280127040 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 41.226059s. I0415 23:37:46.666325 136848290612800 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 23:37:46.666429 136848290612800 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0415 23:37:46.666496 136848290612800 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0415 23:37:46.666547 136848290612800 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0415 23:37:46.666709 136961595919488 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0415 23:37:46.666843 136961595919488 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 23:37:46.667493 136961595919488 metric_logger.py:185] completed step: 9, seconds: 0.052, TFLOP/s/device: 65.327, Tokens/s/device: 9846.722, 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