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 | 4.1890 | 4.1890 | 0 |
| TFLOP/s | 114.794 | 114.750 | -0.044 |
| Tok/s | 17303.0 | 17296.4 | -6.576 |
| Avg s/step | 4.384 | 6.011 | +1.627 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-16 13:30:02.374084: 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 13:30:02.490113 124912045997184 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 13:31:01.504303 124912045997184 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 13:31:01.504551 124912045997184 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 13:31:01.504603 124912045997184 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 13:31:01.504631 124912045997184 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0416 13:31:01.504655 124912045997184 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 13:31:01.504810 124912045997184 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 13:31:01.603951 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:31:02.009369 124912045997184 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 13:31:02.009768 124912045997184 checkpointing.py:677] Setting up checkpoint logger... I0416 13:31:02.009816 124912045997184 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 13:31:02.009905 124912045997184 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 13:31:02.010775 124912045997184 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 0x719ac33f8ad0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 13:31:04.326310 124912045997184 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0416 13:31:04.326545 124912045997184 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 0x719447139a60>}, handler_registry=None I0416 13:31:04.326932 124912045997184 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x719447139a60>` 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 13:31:04.326983 124912045997184 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x719447139fa0>` 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 13:31:04.327015 124912045997184 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 0x719447139a60>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x719447139a60>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x719447139fa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x719447139fa0>}). I0416 13:31:04.327371 124912045997184 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 13:31:04.327429 124912045997184 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x71944712d300> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 13:31:04.456120 124912045997184 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints I0416 13:31:04.456359 124912045997184 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=5), 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_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x719447138950> I0416 13:31:04.456449 124912045997184 checkpointing.py:301] Checkpoint manager created! W0416 13:31:04.473310 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:31:04.670411 124912045997184 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 13:31:04.670518 124912045997184 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:31:04.766555 124912045997184 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:31:04.766637 124912045997184 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:31:04.781027 124912045997184 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:31:04.781083 124912045997184 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:31:04.801850 124912045997184 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:31:04.801906 124912045997184 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:31:04.816262 124912045997184 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:31:04.816311 124912045997184 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:31:04.830801 124912045997184 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:31:04.830849 124912045997184 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:31:04.852786 124912045997184 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 13:31:04.852842 124912045997184 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:31:04.871294 124912045997184 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 13:31:04.871345 124912045997184 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 13:31:05.053275 124912045997184 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0416 13:31:05.053369 124912045997184 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0416 13:31:05.103740 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0416 13:31:05.483194 124912045997184 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 13:31:05.483300 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 13:31:05.483340 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 13:31:05.483383 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 13:31:05.483422 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 13:31:05.483455 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 13:31:05.483494 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0416 13:31:05.483535 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0416 13:31:05.483565 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0416 13:31:05.483591 124912045997184 maxtext_utils.py:1620] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0416 13:31:05.483626 124912045997184 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 13:31:05.483659 124912045997184 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 13:31:05.906857 124912045997184 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:31:05.906943 124912045997184 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 13:31:05.920848 124912045997184 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:31:05.920898 124912045997184 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 13:31:06.597054 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:31:06.712711 124912045997184 max_utils.py:791] Total memory size: 3.6 GB, Output size: 1.5 GB, Temp size: 2.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0416 13:31:06.713189 124912045997184 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 13:31:06.713331 124912045997184 metric_logger.py:289] number parameters: 1.104 billion W0416 13:31:07.547710 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:31:07.642861 124912045997184 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0416 13:31:07.759880 124912045997184 checkpointing.py:776] Waited 0.11700606346130371 seconds for step 0 to finish before starting checkpointing. I0416 13:31:07.760344 124912045997184 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:31:07.760498 124912045997184 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 13:31:07.760870 124912045997184 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0. I0416 13:31:07.851869 124912045997184 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 13:31:07.936666 124912045997184 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:31:07.936765 124912045997184 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 13:31:07.942513 124804589946432 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0 W0416 13:31:08.151294 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.164079 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.172732 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.192859 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.202723 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.208096 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.212272 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.216455 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:31:08.401865 124912045997184 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.465620s I0416 13:31:08.402207 124912045997184 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 22.476 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 549 milliseconds) (per-host) I0416 13:31:08.402259 124912045997184 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.549145s (batch_requests_ready=0.079465s, total_serialization_initiated=0.469443s, others=0.000238s) I0416 13:31:08.402318 124912045997184 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.549977s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.549959) I0416 13:31:08.403324 124804527031872 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:31:08.403414 124912045997184 async_checkpointer.py:561] Finished blocking save. Time taken: 0.642878s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0. I0416 13:31:08.403584 124912045997184 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:31:08.403725 124804600432192 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:31:08.403800 124912045997184 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776346267.7603312, 'wait_for_prev_duration_secs': 4.410743713378906e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776346267.7605186, 'checkpointer_blocking_duration_secs': 0.6429955959320068, 'get_old_steps_start_time': 1776346268.4035306, 'get_old_steps_duration_secs': 2.0265579223632812e-05, 'checkpoint_manager_blocking_start_time': 1776346267.7602484, 'checkpoint_manager_blocking_duration_secs': 0.6435229778289795} I0416 13:31:08.403881 124912045997184 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0416 13:31:08.403917 124912045997184 max_utils.py:750] Memstats: After params initialized: I0416 13:31:08.403996 124912045997184 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0416 13:31:08.404023 124912045997184 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 13:31:08.404055 124912045997184 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 13:31:08.404074 124912045997184 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 13:31:08.404092 124912045997184 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 13:31:08.404110 124912045997184 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 13:31:08.404127 124912045997184 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 13:31:08.404143 124912045997184 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 13:31:08.409835 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.415814 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:31:08.420647 3874019 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:31:08.619035 124912045997184 metric_logger.py:185] completed step: 0, seconds: 0.929, TFLOP/s/device: 14.624, Tokens/s/device: 2204.353, total_weights: 16384, loss: 10.874 I0416 13:31:08.619917 124912045997184 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/tensorboard/' I0416 13:31:08.623028 124804579460672 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0/items I0416 13:31:08.735088 124912045997184 metric_logger.py:185] completed step: 1, seconds: 0.975, TFLOP/s/device: 13.942, Tokens/s/device: 2101.541, total_weights: 16384, loss: 10.874 I0416 13:31:08.767869 124804508157504 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776346268538503045, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776346268538503045, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0416 13:31:08.853759 124912045997184 metric_logger.py:185] completed step: 2, seconds: 0.006, TFLOP/s/device: 2463.668, Tokens/s/device: 371350.861, total_weights: 16384, loss: 10.276 I0416 13:31:08.972105 124912045997184 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.557, Tokens/s/device: 17568.692, total_weights: 16384, loss: 9.282 I0416 13:31:08.976727 124912045997184 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0416 13:31:09.095814 3876191 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 13:31:09.095874 3876191 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 13:31:09.209507 124912045997184 checkpointing.py:776] Waited 0.23275017738342285 seconds for step 5 to finish before starting checkpointing. I0416 13:31:09.209889 124912045997184 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:31:09.832719 124804548003392 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_15_ocdbt_true/checkpoints/0/items/array_metadatas/process_0 I0416 13:31:38.514549 124804537517632 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 29.422279s (commit=29.023055s, array_metadata_write=0.399225s) I0416 13:31:38.515668 124804527031872 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 412.125 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 30 seconds) (per-host) I0416 13:31:38.515770 124804527031872 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 30.112311s. I0416 13:31:38.728619 124804527031872 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776346268538503045, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0416 13:31:38.913970 124804527031872 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 13:31:39.112495 124804508157504 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776346268538503045, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0416 13:31:39.331569 124804527031872 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 13:31:39.332115 124804527031872 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.559491s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0/items I0416 13:31:39.332767 124804527031872 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0/items I0416 13:31:39.558413 124804527031872 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0 I0416 13:31:40.212918 124804527031872 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_15_ocdbt_true/checkpoints/0`. I0416 13:31:40.213527 124804527031872 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 32.452997s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/0 I0416 13:31:40.213611 124804527031872 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 31.810155s. I0416 13:31:40.213725 124804600432192 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:31:40.213776 124804600432192 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:31:40.213835 124804600432192 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:31:40.213883 124804600432192 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 13:31:40.214203 124912045997184 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 13:31:40.214330 124912045997184 checkpoint_manager.py:1441] Waiting for previous save to complete took 31.004437 seconds. If this number is high, consider checkpointing less frequently. I0416 13:31:40.215203 124912045997184 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 5 I0416 13:31:40.215492 124912045997184 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/5. I0416 13:31:40.379513 124804600432192 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/5 I0416 13:31:40.390201 124912045997184 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:31:40.390295 124912045997184 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 13:31:41.092994 124804487185984 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/5/items I0416 13:31:46.842718 124912045997184 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 6.452914s I0416 13:31:46.843254 124912045997184 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.885 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 6 seconds) (per-host) I0416 13:31:46.843310 124912045997184 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 6.547649s (batch_requests_ready=0.090715s, total_serialization_initiated=6.456504s, others=0.000430s) I0416 13:31:46.843376 124912045997184 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 6.548161s (all_items=0.000016s, per_item={'items': '0.00001645'}, temp_paths=6.548145) I0416 13:31:46.848331 124804600432192 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:31:46.848425 124912045997184 async_checkpointer.py:561] Finished blocking save. Time taken: 6.633176s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/5. I0416 13:31:46.848621 124912045997184 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:31:46.848801 124804466214464 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:31:46.848896 124912045997184 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776346269.2098665, 'wait_for_prev_duration_secs': 31.00443744659424, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776346300.215228, 'checkpointer_blocking_duration_secs': 6.633303165435791, 'get_old_steps_start_time': 1776346306.8485518, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1776346269.2098022, 'checkpoint_manager_blocking_duration_secs': 37.639068603515625} I0416 13:31:46.849022 124912045997184 checkpointing.py:408] Started an asynchronous checkpoint save for step 5 I0416 13:31:46.849494 124912045997184 metric_logger.py:185] completed step: 4, seconds: 0.119, TFLOP/s/device: 114.261, Tokens/s/device: 17222.675, total_weights: 16384, loss: 8.327 I0416 13:31:46.856796 124912045997184 metric_logger.py:185] completed step: 5, seconds: 0.118, TFLOP/s/device: 114.955, Tokens/s/device: 17327.298, total_weights: 16384, loss: 7.367 I0416 13:31:46.972743 124912045997184 metric_logger.py:185] completed step: 6, seconds: 37.878, TFLOP/s/device: 0.359, Tokens/s/device: 54.068, total_weights: 16384, loss: 6.459 I0416 13:31:47.091159 124912045997184 metric_logger.py:185] completed step: 7, seconds: 0.006, TFLOP/s/device: 2425.407, Tokens/s/device: 365583.720, total_weights: 16384, loss: 5.622 I0416 13:31:47.209087 124912045997184 metric_logger.py:185] completed step: 8, seconds: 0.117, TFLOP/s/device: 116.386, Tokens/s/device: 17542.958, total_weights: 16384, loss: 4.864 I0416 13:31:47.327635 124912045997184 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0416 13:31:47.328361 124912045997184 checkpointing.py:776] Waited 0.0007586479187011719 seconds for step 9 to finish before starting checkpointing. I0416 13:31:47.328598 124912045997184 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:31:47.574655 124804548003392 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_15_ocdbt_true/checkpoints/5/items/array_metadatas/process_0 I0416 13:32:16.896650 124804487185984 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 30.049021s (commit=29.642190s, array_metadata_write=0.406831s) I0416 13:32:16.897785 124804600432192 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 345.248 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 36 seconds) (per-host) I0416 13:32:16.897846 124804600432192 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 30.049376s. I0416 13:32:17.312066 124804600432192 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 13:32:17.708699 124804600432192 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 13:32:17.708863 124804600432192 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.552932s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/5/items I0416 13:32:17.709511 124804600432192 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/5/items I0416 13:32:17.951792 124804600432192 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/5 I0416 13:32:18.606630 124804600432192 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_15_ocdbt_true/checkpoints/5`. I0416 13:32:18.607260 124804600432192 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 38.392017s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/5 I0416 13:32:18.607338 124804600432192 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 31.758870s. I0416 13:32:18.607563 124804466214464 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:32:18.607672 124804466214464 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:32:18.607737 124804466214464 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:32:18.607783 124804466214464 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0416 13:32:18.608567 124912045997184 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5. W0416 13:32:18.609345 124912045997184 checkpoint_manager.py:1441] Waiting for previous save to complete took 31.280749 seconds. If this number is high, consider checkpointing less frequently. I0416 13:32:18.609889 124912045997184 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 13:32:18.610186 124912045997184 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/9. I0416 13:32:18.783736 124804466214464 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/9 I0416 13:32:18.793775 124912045997184 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:32:18.793866 124912045997184 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 13:32:19.478129 124804476700224 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/9/items I0416 13:32:27.623598 124912045997184 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 8.830223s I0416 13:32:27.627781 124912045997184 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.383 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 8 seconds) (per-host) I0416 13:32:27.627845 124912045997184 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 8.925339s (batch_requests_ready=0.087650s, total_serialization_initiated=8.833607s, others=0.004082s) I0416 13:32:27.627924 124912045997184 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 8.925847s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=8.925833) I0416 13:32:27.629423 124804466214464 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:32:27.629521 124912045997184 async_checkpointer.py:561] Finished blocking save. Time taken: 9.019585s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/9. I0416 13:32:27.629745 124912045997184 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:32:27.629953 124804600432192 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:32:27.630079 124912045997184 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776346307.3285747, 'wait_for_prev_duration_secs': 31.280749082565308, 'time_between_consecutive_saves_sec': 7.114675998687744, 'checkpointer_blocking_start_time': 1776346338.6099129, 'checkpointer_blocking_duration_secs': 9.019715785980225, 'get_old_steps_start_time': 1776346347.6296535, 'get_old_steps_duration_secs': 4.291534423828125e-05, 'checkpoint_manager_blocking_start_time': 1776346307.328544, 'checkpoint_manager_blocking_duration_secs': 40.301509618759155} I0416 13:32:27.630204 124912045997184 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0416 13:32:27.630242 124912045997184 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:32:28.362474 124804548003392 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_15_ocdbt_true/checkpoints/9/items/array_metadatas/process_0 I0416 13:33:03.533934 124804476700224 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 35.905431s (commit=35.465348s, array_metadata_write=0.440084s) I0416 13:33:03.534966 124804466214464 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 281.867 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 44 seconds) (per-host) I0416 13:33:03.535089 124804466214464 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.905523s. I0416 13:33:03.961299 124804466214464 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 13:33:04.414094 124804466214464 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 13:33:04.414270 124804466214464 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.595280s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/9/items I0416 13:33:04.414892 124804466214464 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/9/items I0416 13:33:04.640064 124804466214464 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/9 I0416 13:33:05.282602 124804466214464 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_15_ocdbt_true/checkpoints/9`. I0416 13:33:05.283224 124804466214464 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 46.673298s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_15_ocdbt_true/checkpoints/9 I0416 13:33:05.283292 124804466214464 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 37.653728s. I0416 13:33:05.283468 124804600432192 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:33:05.283581 124804600432192 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:33:05.283639 124804600432192 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:33:05.283689 124804600432192 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 13:33:05.283848 124912045997184 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 13:33:05.283987 124912045997184 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:33:05.284640 124912045997184 metric_logger.py:185] completed step: 9, seconds: 0.118, TFLOP/s/device: 114.794, Tokens/s/device: 17302.997, total_weights: 16384, loss: 4.189 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
2026-04-16 19:09:57.384249: 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 19:09:57.501169 137725793389696 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 19:10:57.479930 137725793389696 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 19:10:57.480032 137725793389696 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 19:10:57.480085 137725793389696 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 19:10:57.480110 137725793389696 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 19:10:57.480134 137725793389696 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 19:10:57.480212 137725793389696 train.py:811] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 19:10:57.572632 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:10:57.972321 137725793389696 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 19:10:57.972486 137725793389696 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 19:10:57.972713 137725793389696 checkpointing.py:688] Setting up checkpoint logger... I0416 19:10:57.972753 137725793389696 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 19:10:57.972791 137725793389696 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 19:10:57.973473 137725793389696 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 0x7d4232df8c80>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 19:11:00.389139 137725793389696 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 19:11:00.389326 137725793389696 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 0x7d3bb653dc40>}, handler_registry=None I0416 19:11:00.389570 137725793389696 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d3bb653dc40>` 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 19:11:00.389620 137725793389696 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d3bb653e210>` 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 19:11:00.389651 137725793389696 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 0x7d3bb653dc40>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d3bb653dc40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d3bb653e210>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d3bb653e210>}). I0416 19:11:00.389949 137725793389696 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 19:11:00.390000 137725793389696 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7d3bb6531ee0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 19:11:00.520070 137725793389696 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_15_ocdbt_true/checkpoints I0416 19:11:00.520294 137725793389696 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=5), 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_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d3bb653c260> I0416 19:11:00.520369 137725793389696 checkpointing.py:302] Checkpoint manager created! W0416 19:11:00.537479 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:11:00.732686 137725793389696 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 19:11:00.732844 137725793389696 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:11:00.824281 137725793389696 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 19:11:00.824359 137725793389696 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:11:00.838532 137725793389696 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 19:11:00.838580 137725793389696 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:11:00.859696 137725793389696 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 19:11:00.859750 137725793389696 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:11:00.874033 137725793389696 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 19:11:00.874099 137725793389696 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:11:00.888386 137725793389696 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 19:11:00.888433 137725793389696 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:11:00.909993 137725793389696 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 19:11:00.910062 137725793389696 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:11:00.928243 137725793389696 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 19:11:00.928296 137725793389696 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 19:11:01.134186 137725793389696 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 19:11:01.134284 137725793389696 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0416 19:11:01.184847 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0416 19:11:01.578958 137725793389696 maxtext_utils.py:1796] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 19:11:01.579085 137725793389696 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, None) I0416 19:11:01.579128 137725793389696 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, None) I0416 19:11:01.579176 137725793389696 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 19:11:01.579218 137725793389696 maxtext_utils.py:1796] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 19:11:01.579246 137725793389696 maxtext_utils.py:1796] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 19:11:01.579286 137725793389696 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, None, None) I0416 19:11:01.579324 137725793389696 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: (None, None, None, 'fsdp') I0416 19:11:01.579353 137725793389696 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, None, None) I0416 19:11:01.579378 137725793389696 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, None, None) I0416 19:11:01.579412 137725793389696 maxtext_utils.py:1796] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 19:11:01.579447 137725793389696 maxtext_utils.py:1796] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 19:11:02.013790 137725793389696 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 19:11:02.013870 137725793389696 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 19:11:02.027818 137725793389696 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 19:11:02.027869 137725793389696 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 19:11:02.666097 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:11:02.790887 137725793389696 max_utils.py:791] Total memory size: 3.6 GB, Output size: 1.5 GB, Temp size: 2.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0416 19:11:02.791363 137725793389696 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 19:11:02.791502 137725793389696 metric_logger.py:289] number parameters: 1.104 billion W0416 19:11:03.692731 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:11:03.792126 137725793389696 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 19:11:03.909214 137725793389696 checkpointing.py:798] Waited 0.11712813377380371 seconds for step 0 to finish before starting checkpointing. I0416 19:11:03.909742 137725793389696 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 19:11:03.909901 137725793389696 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 19:11:03.910367 137725793389696 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_15_ocdbt_true/checkpoints/0. I0416 19:11:03.997108 137725793389696 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 19:11:04.081236 137618295621184 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_15_ocdbt_true/checkpoints/0 I0416 19:11:04.089232 137725793389696 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:11:04.089322 137725793389696 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 W0416 19:11:04.301620 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:04.312295 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:04.319196 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:04.333309 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:04.348467 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:04.354642 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:04.360934 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:04.364925 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:11:04.805508 137618285135424 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_15_ocdbt_true/checkpoints/0/items I0416 19:11:04.950336 137618253678144 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776366664709570595, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776366664709570595, "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_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0416 19:11:05.319761 4024562 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 19:11:05.319811 4024562 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 19:11:06.195250 137725793389696 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 2.106387s I0416 19:11:06.200180 137725793389696 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.605 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 2 seconds) (per-host) I0416 19:11:06.200254 137725793389696 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 2.201937s (batch_requests_ready=0.086589s, total_serialization_initiated=2.110563s, others=0.004785s) I0416 19:11:06.200352 137725793389696 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.202789s (all_items=0.000021s, per_item={'items': '0.00002098'}, temp_paths=2.202768) I0416 19:11:06.201985 137618190763584 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:11:06.202085 137725793389696 async_checkpointer.py:561] Finished blocking save. Time taken: 2.292143s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/0. I0416 19:11:06.202303 137725793389696 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:11:06.202527 137618306106944 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:11:06.202633 137725793389696 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_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776366663.9097285, 'wait_for_prev_duration_secs': 4.2438507080078125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776366663.9099228, 'checkpointer_blocking_duration_secs': 2.2922816276550293, 'get_old_steps_start_time': 1776366666.2022364, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1776366663.909631, 'checkpoint_manager_blocking_duration_secs': 2.2929723262786865} I0416 19:11:06.202766 137725793389696 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 19:11:06.202825 137725793389696 max_utils.py:750] Memstats: After params initialized: I0416 19:11:06.202881 137725793389696 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0416 19:11:06.202906 137725793389696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 19:11:06.202926 137725793389696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 19:11:06.202944 137725793389696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 19:11:06.202961 137725793389696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 19:11:06.202978 137725793389696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 19:11:06.202994 137725793389696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 19:11:06.203010 137725793389696 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 19:11:06.209017 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:06.215015 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:11:06.219430 4022459 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:11:06.407229 137725793389696 metric_logger.py:185] completed step: 0, seconds: 1.000, TFLOP/s/device: 13.586, Tokens/s/device: 2047.826, total_weights: 16384, loss: 10.874 I0416 19:11:06.408150 137725793389696 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_15_ocdbt_true/tensorboard/' I0416 19:11:06.523262 137725793389696 metric_logger.py:185] completed step: 1, seconds: 2.614, TFLOP/s/device: 5.199, Tokens/s/device: 783.613, total_weights: 16384, loss: 10.874 I0416 19:11:06.641897 137725793389696 metric_logger.py:185] completed step: 2, seconds: 0.006, TFLOP/s/device: 2451.665, Tokens/s/device: 369541.682, total_weights: 16384, loss: 10.276 I0416 19:11:06.760076 137725793389696 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.475, Tokens/s/device: 17556.342, total_weights: 16384, loss: 9.282 I0416 19:11:06.764592 137725793389696 checkpointing.py:794] Waiting for step 5 to finish before checkpoint... I0416 19:11:06.937224 137618264163904 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_15_ocdbt_true/checkpoints/0/items/array_metadatas/process_0 I0416 19:11:06.996366 137725793389696 checkpointing.py:798] Waited 0.23174500465393066 seconds for step 5 to finish before starting checkpointing. I0416 19:11:06.996750 137725793389696 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:11:44.301543 137618211735104 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 38.100774s (commit=37.628020s, array_metadata_write=0.472755s) I0416 19:11:44.302490 137618190763584 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 313.535 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 40 seconds) (per-host) I0416 19:11:44.302545 137618190763584 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.100412s. I0416 19:11:44.554275 137618190763584 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776366664709570595, '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_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0416 19:11:44.744959 137618190763584 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 19:11:44.946750 137618253678144 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776366664709570595, '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_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0416 19:11:45.177793 137618190763584 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 19:11:45.178457 137618190763584 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.576671s. 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_15_ocdbt_true/checkpoints/0/items I0416 19:11:45.179108 137618190763584 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/0/items I0416 19:11:45.436529 137618190763584 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/0 I0416 19:11:46.154687 137618190763584 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_15_ocdbt_true/checkpoints/0`. I0416 19:11:46.155345 137618190763584 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 42.245409s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/0 I0416 19:11:46.155420 137618190763584 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 39.953289s. I0416 19:11:46.155612 137618306106944 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:11:46.155720 137618306106944 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:11:46.155781 137618306106944 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:11:46.155826 137618306106944 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 19:11:46.155977 137725793389696 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 19:11:46.156114 137725793389696 checkpoint_manager.py:1441] Waiting for previous save to complete took 39.159363 seconds. If this number is high, consider checkpointing less frequently. I0416 19:11:46.156916 137725793389696 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 5 I0416 19:11:46.157212 137725793389696 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_15_ocdbt_true/checkpoints/5. I0416 19:11:46.334233 137725793389696 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:11:46.334373 137618306106944 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_15_ocdbt_true/checkpoints/5 I0416 19:11:46.334448 137725793389696 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 19:11:47.053485 137618180277824 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_15_ocdbt_true/checkpoints/5/items I0416 19:11:57.635629 137725793389696 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 11.302126s I0416 19:11:57.639757 137725793389696 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.083 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 11 seconds) (per-host) I0416 19:11:57.639828 137725793389696 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 11.397292s (batch_requests_ready=0.088027s, total_serialization_initiated=11.305231s, others=0.004034s) I0416 19:11:57.639907 137725793389696 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 11.397799s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=11.397782) I0416 19:11:57.641400 137618413061696 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:11:57.641497 137725793389696 async_checkpointer.py:561] Finished blocking save. Time taken: 11.484537s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/5. I0416 19:11:57.641701 137725793389696 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:11:57.641869 137618190763584 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:11:57.641969 137725793389696 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776366666.9967232, 'wait_for_prev_duration_secs': 39.15936303138733, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776366706.156941, 'checkpointer_blocking_duration_secs': 11.484680652618408, 'get_old_steps_start_time': 1776366717.641637, 'get_old_steps_duration_secs': 2.09808349609375e-05, 'checkpoint_manager_blocking_start_time': 1776366666.9966595, 'checkpoint_manager_blocking_duration_secs': 50.64528250694275} I0416 19:11:57.642103 137725793389696 checkpointing.py:409] Started an asynchronous checkpoint save for step 5 I0416 19:11:57.642555 137725793389696 metric_logger.py:185] completed step: 4, seconds: 0.119, TFLOP/s/device: 114.652, Tokens/s/device: 17281.534, total_weights: 16384, loss: 8.327 I0416 19:11:57.649099 137725793389696 metric_logger.py:185] completed step: 5, seconds: 0.118, TFLOP/s/device: 114.971, Tokens/s/device: 17329.644, total_weights: 16384, loss: 7.367 I0416 19:11:57.765289 137725793389696 metric_logger.py:185] completed step: 6, seconds: 50.883, TFLOP/s/device: 0.267, Tokens/s/device: 40.249, total_weights: 16384, loss: 6.459 I0416 19:11:57.883868 137725793389696 metric_logger.py:185] completed step: 7, seconds: 0.005, TFLOP/s/device: 2602.898, Tokens/s/device: 392337.165, total_weights: 16384, loss: 5.622 I0416 19:11:58.002586 137725793389696 metric_logger.py:185] completed step: 8, seconds: 0.117, TFLOP/s/device: 115.797, Tokens/s/device: 17454.149, total_weights: 16384, loss: 4.864 I0416 19:11:58.120389 137725793389696 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0416 19:11:58.121162 137725793389696 checkpointing.py:798] Waited 0.0007939338684082031 seconds for step 9 to finish before starting checkpointing. I0416 19:11:58.121405 137725793389696 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:11:58.453375 137618423547456 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_15_ocdbt_true/checkpoints/5/items/array_metadatas/process_0 I0416 19:12:32.131221 137618211735104 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 34.490846s (commit=33.994949s, array_metadata_write=0.495897s) I0416 19:12:32.132221 137618413061696 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 275.373 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 45 seconds) (per-host) I0416 19:12:32.132327 137618413061696 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 34.490771s. I0416 19:12:32.574672 137618413061696 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 19:12:33.046683 137618413061696 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 19:12:33.046851 137618413061696 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.624030s. 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_15_ocdbt_true/checkpoints/5/items I0416 19:12:33.047467 137618413061696 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/5/items I0416 19:12:33.307209 137618413061696 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/5 I0416 19:12:34.035811 137618413061696 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_15_ocdbt_true/checkpoints/5`. I0416 19:12:34.036437 137618413061696 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 47.879484s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/5 I0416 19:12:34.036513 137618413061696 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 36.394959s. I0416 19:12:34.036686 137618190763584 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:12:34.036788 137618190763584 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:12:34.036853 137618190763584 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:12:34.036903 137618190763584 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0416 19:12:34.037034 137725793389696 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5. W0416 19:12:34.037148 137725793389696 checkpoint_manager.py:1441] Waiting for previous save to complete took 35.915753 seconds. If this number is high, consider checkpointing less frequently. I0416 19:12:34.037341 137725793389696 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 19:12:34.037613 137725793389696 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_15_ocdbt_true/checkpoints/9. I0416 19:12:34.217218 137725793389696 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:12:34.217307 137725793389696 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 19:12:34.226294 137618413061696 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_15_ocdbt_true/checkpoints/9 I0416 19:12:34.931095 137618402575936 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_15_ocdbt_true/checkpoints/9/items I0416 19:12:45.420331 137725793389696 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 11.203481s I0416 19:12:45.424369 137725793389696 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.092 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 11 seconds) (per-host) I0416 19:12:45.424430 137725793389696 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 11.297661s (batch_requests_ready=0.087075s, total_serialization_initiated=11.206651s, others=0.003934s) I0416 19:12:45.424498 137725793389696 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 11.298137s (all_items=0.000011s, per_item={'items': '0.00001073'}, temp_paths=11.298126) I0416 19:12:45.426114 137618413061696 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:12:45.426200 137725793389696 async_checkpointer.py:561] Finished blocking save. Time taken: 11.388818s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/9. I0416 19:12:45.426364 137725793389696 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:12:45.426520 137618190763584 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:12:45.426626 137725793389696 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_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776366718.1213808, 'wait_for_prev_duration_secs': 35.91575264930725, 'time_between_consecutive_saves_sec': 11.965540409088135, 'checkpointer_blocking_start_time': 1776366754.0373645, 'checkpointer_blocking_duration_secs': 11.388941049575806, 'get_old_steps_start_time': 1776366765.426317, 'get_old_steps_duration_secs': 1.5974044799804688e-05, 'checkpoint_manager_blocking_start_time': 1776366718.121348, 'checkpoint_manager_blocking_duration_secs': 47.305254220962524} I0416 19:12:45.426738 137725793389696 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0416 19:12:45.426769 137725793389696 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:12:46.204342 137617676961344 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_15_ocdbt_true/checkpoints/9/items/array_metadatas/process_0 I0416 19:13:23.951453 137618306106944 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 38.526590s (commit=38.011091s, array_metadata_write=0.515499s) I0416 19:13:23.952626 137618413061696 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 253.619 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 49 seconds) (per-host) I0416 19:13:23.952732 137618413061696 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.526486s. I0416 19:13:24.361334 137618413061696 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 19:13:24.813541 137618413061696 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 19:13:24.814303 137618413061696 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.600622s. 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_15_ocdbt_true/checkpoints/9/items I0416 19:13:24.814985 137618413061696 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/9/items I0416 19:13:25.069691 137618413061696 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/9 I0416 19:13:25.765969 137618413061696 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_15_ocdbt_true/checkpoints/9`. I0416 19:13:25.766608 137618413061696 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 51.729232s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_15_ocdbt_true/checkpoints/9 I0416 19:13:25.766673 137618413061696 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 40.340428s. I0416 19:13:25.766835 137618190763584 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:13:25.766943 137618190763584 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:13:25.767003 137618190763584 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:13:25.767053 137618190763584 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 19:13:25.767507 137725793389696 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 19:13:25.768175 137725793389696 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 19:13:25.769493 137725793389696 metric_logger.py:185] completed step: 9, seconds: 0.118, TFLOP/s/device: 114.750, Tokens/s/device: 17296.421, total_weights: 16384, loss: 4.189 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops