main (586e69205) vs feat/nnx-post-train-fixes (7f06c99ac)| Metric | main 586e69205 | feat/nnx-post-train-fixes 7f06c99ac | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 7.1810 | 7.1810 | 0 |
| TFLOP/s | 91.922 | 92.220 | +0.298 |
| Tok/s | 13855.4 | 13900.4 | +44.952 |
| Avg s/step | 4.638 | 4.636 | -0.002 |
| Memory % | 1.38 | 1.38 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Thu Apr 23 08:29:51 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-23 08:30:16.335368: 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) I0423 08:30:16.547871 132865507964736 max_utils.py:273] Attempting to initialize the jax distributed system... I0423 08:30:25.588027 132865507964736 distributed.py:149] Starting JAX distributed service on [::]:8482 I0423 08:30:25.590349 132865507964736 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-pvvo2-slice-job-0-0.mt-15-ocdbt-false-sv-pvvo2:8482 I0423 08:30:26.727733 132865507964736 max_utils.py:284] Jax distributed system initialized! I0423 08:30:31.828818 132865507964736 max_utils.py:800] System Information: Jax Version: 0.9.2 I0423 08:30:31.828939 132865507964736 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0423 08:30:31.828984 132865507964736 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0423 08:30:31.829021 132865507964736 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 08:30:32.521319 132865507964736 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 08:30:32.521606 132865507964736 checkpointing.py:677] Setting up checkpoint logger... I0423 08:30:32.521661 132865507964736 checkpointing.py:233] Creating checkpoint manager with ocdbt=False and zarr3=True I0423 08:30:32.521706 132865507964736 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0423 08:30:32.522058 132865507964736 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x78d6907190a0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 08:30:35.831199 132865507964736 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0423 08:30:35.831470 132865507964736 checkpoint_manager.py:708] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78d565818380>}, handler_registry=None I0423 08:30:35.831881 132865507964736 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78d565818380>` 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`. I0423 08:30:35.831971 132865507964736 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78c2903fcef0>` 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`. I0423 08:30:35.832026 132865507964736 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 0x78d565818380>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78d565818380>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78c2903fcef0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78c2903fcef0>}). I0423 08:30:35.832386 132865507964736 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0423 08:30:35.832476 132865507964736 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x78c290159a80> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0423 08:30:36.545762 132865507964736 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints I0423 08:30:37.004840 132865507964736 checkpoint_manager.py:929] [process=3][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_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, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x78c29061bbf0> I0423 08:30:37.005044 132865507964736 checkpointing.py:301] Checkpoint manager created! I0423 08:30:37.933543 132865507964736 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 08:30:37.933659 132865507964736 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:30:38.313962 132865507964736 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0423 08:30:38.314055 132865507964736 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:30:38.330562 132865507964736 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0423 08:30:38.330623 132865507964736 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:30:38.354600 132865507964736 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:30:38.354672 132865507964736 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:30:38.371298 132865507964736 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:30:38.371358 132865507964736 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:30:38.387986 132865507964736 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:30:38.388052 132865507964736 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:30:38.412833 132865507964736 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0423 08:30:38.412906 132865507964736 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:30:38.434405 132865507964736 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 08:30:38.434477 132865507964736 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0423 08:30:38.646897 132865507964736 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0423 08:30:38.647021 132865507964736 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0423 08:30:40.075982 132865507964736 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0423 08:30:40.076107 132865507964736 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 08:30:40.076160 132865507964736 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 08:30:40.076216 132865507964736 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0423 08:30:40.076268 132865507964736 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 08:30:40.076307 132865507964736 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 08:30:40.076359 132865507964736 maxtext_utils.py:1707] 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) I0423 08:30:40.076410 132865507964736 maxtext_utils.py:1707] 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') I0423 08:30:40.076455 132865507964736 maxtext_utils.py:1707] 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) I0423 08:30:40.076492 132865507964736 maxtext_utils.py:1707] 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) I0423 08:30:40.076543 132865507964736 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 08:30:40.076593 132865507964736 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 08:30:40.575580 132865507964736 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 08:30:40.575682 132865507964736 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0423 08:30:40.591452 132865507964736 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 08:30:40.591529 132865507964736 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0423 08:30:51.527806 132865507964736 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0423 08:30:51.528592 132865507964736 metric_logger.py:301] number parameters: 1.104 billion I0423 08:31:03.240464 132865507964736 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0423 08:31:03.393254 132865507964736 checkpointing.py:776] Waited 0.1527721881866455 seconds for step 0 to finish before starting checkpointing. I0423 08:31:03.395957 132865507964736 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:31:03.397541 132865507964736 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0 I0423 08:31:03.398907 132865507964736 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/0. I0423 08:31:03.760399 132865507964736 signaling_client.py:364] Using JaxDistributedSignalingClient I0423 08:31:03.763003 132865507964736 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:31:03.763061 132865507964736 replica_slices.py:424] 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 I0423 08:31:04.043425 132865507964736 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.283094s I0423 08:31:04.043599 132865507964736 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 5.345 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28858208656311035 s) (per-host) I0423 08:31:04.043650 132865507964736 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.288644s (batch_requests_ready=0.002203s, total_serialization_initiated=0.286371s, others=0.000071s) I0423 08:31:04.043745 132865507964736 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.292760s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.292743) I0423 08:31:04.044601 132865507964736 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.65 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/0. I0423 08:31:04.044960 132739588159232 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:51:04.044909 I0423 08:31:04.046903 132865507964736 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0423 08:31:04.047193 132738514417408 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0423 08:31:04.047321 132865507964736 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776933063.3959374, 'wait_for_prev_duration_secs': 6.318092346191406e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776933063.3975787, 'checkpointer_blocking_duration_secs': 0.6475358009338379, 'get_old_steps_start_time': 1776933064.0451403, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776933063.393722, 'checkpoint_manager_blocking_duration_secs': 0.6535618305206299} I0423 08:31:04.047479 132865507964736 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0423 08:31:04.047529 132865507964736 max_utils.py:750] Memstats: After params initialized: I0423 08:31:04.047575 132865507964736 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0)) I0423 08:31:04.047607 132865507964736 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0)) I0423 08:31:04.047635 132865507964736 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0)) I0423 08:31:04.047661 132865507964736 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0)) I0423 08:31:04.363073 132865507964736 metric_logger.py:196] completed step: 0, seconds: 11.712, TFLOP/s/device: 1.160, Tokens/s/device: 174.867, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 08:31:04.544290 132865507964736 metric_logger.py:196] completed step: 1, seconds: 1.121, TFLOP/s/device: 12.120, Tokens/s/device: 1826.811, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 08:31:04.986155 132865507964736 metric_logger.py:196] completed step: 2, seconds: 0.034, TFLOP/s/device: 398.520, Tokens/s/device: 60069.220, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0423 08:31:05.133652 132865507964736 metric_logger.py:196] completed step: 3, seconds: 0.442, TFLOP/s/device: 30.735, Tokens/s/device: 4632.761, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0423 08:31:05.138303 132865507964736 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0423 08:31:05.427618 132865507964736 checkpointing.py:776] Waited 0.2892787456512451 seconds for step 5 to finish before starting checkpointing. I0423 08:31:05.430583 132865507964736 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 08:31:06.989534 2737 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 08:31:09.518484 132739529664256 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_3 I0423 08:31:14.028451 132739588159232 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 153.756 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.273394584655762 s) (per-host) I0423 08:31:14.028579 132739588159232 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 9.983504s. I0423 08:31:21.656656 132739588159232 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 17.611564s. I0423 08:31:21.656955 132738514417408 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 08:31:21.657072 132738514417408 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0423 08:31:21.657124 132738514417408 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:31:21.658803 132738514417408 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0423 08:31:21.659005 132865507964736 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0423 08:31:21.659138 132865507964736 checkpoint_manager.py:1452] Waiting for previous save to complete took 16.228553 seconds. If this number is high, consider checkpointing less frequently. I0423 08:31:21.661258 132865507964736 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 5 I0423 08:31:21.663746 132865507964736 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/5. I0423 08:31:21.950366 132865507964736 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:31:21.950463 132865507964736 replica_slices.py:424] 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 I0423 08:31:21.985687 132865507964736 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036465s I0423 08:31:21.985862 132865507964736 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 38.630 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.0399320125579834 s) (per-host) I0423 08:31:21.985916 132865507964736 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.039995s (batch_requests_ready=0.001768s, total_serialization_initiated=0.038159s, others=0.000069s) I0423 08:31:21.986024 132865507964736 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.045158s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.045143) I0423 08:31:21.986682 132865507964736 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/5. I0423 08:31:21.987030 132739571373824 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:51:21.986989 I0423 08:31:21.994497 132865507964736 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0423 08:31:21.994729 132738514417408 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0423 08:31:21.994859 132865507964736 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776933065.4305527, 'wait_for_prev_duration_secs': 16.228553295135498, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776933081.6613023, 'checkpointer_blocking_duration_secs': 0.32587265968322754, 'get_old_steps_start_time': 1776933081.9872038, 'get_old_steps_duration_secs': 3.719329833984375e-05, 'checkpoint_manager_blocking_start_time': 1776933065.42832, 'checkpoint_manager_blocking_duration_secs': 16.566500425338745} I0423 08:31:21.994992 132865507964736 checkpointing.py:408] Started an asynchronous checkpoint save for step 5 I0423 08:31:21.995907 132865507964736 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.704, Tokens/s/device: 13370.503, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0423 08:31:22.002679 132865507964736 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.231, Tokens/s/device: 13902.088, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0423 08:31:27.794533 132739529664256 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_3 I0423 08:31:33.357706 132739571373824 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 138.419 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.411736011505127 s) (per-host) I0423 08:31:33.357833 132739571373824 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.370687s. I0423 08:31:39.982445 132739571373824 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 17.995282s. I0423 08:31:39.982698 132738514417408 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 08:31:39.982812 132738514417408 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0423 08:31:39.982861 132738514417408 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:31:39.985458 132738514417408 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0423 08:31:44.832757 132865507964736 metric_logger.py:196] completed step: 6, seconds: 16.863, TFLOP/s/device: 0.806, Tokens/s/device: 121.451, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0423 08:31:44.980133 132865507964736 metric_logger.py:196] completed step: 7, seconds: 22.684, TFLOP/s/device: 0.599, Tokens/s/device: 90.284, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0423 08:31:45.127297 132865507964736 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.118, Tokens/s/device: 13432.855, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0423 08:31:45.274196 132865507964736 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0423 08:31:45.274881 132865507964736 checkpointing.py:776] Waited 0.0007050037384033203 seconds for step 9 to finish before starting checkpointing. I0423 08:31:45.276972 132865507964736 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:31:45.278685 132865507964736 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 9 I0423 08:31:45.279913 132865507964736 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/9. I0423 08:31:45.592002 132865507964736 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:31:45.592092 132865507964736 replica_slices.py:424] 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 I0423 08:31:45.628070 132865507964736 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037009s I0423 08:31:45.628220 132865507964736 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 38.114 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04047250747680664 s) (per-host) I0423 08:31:45.628267 132865507964736 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040530s (batch_requests_ready=0.001643s, total_serialization_initiated=0.038825s, others=0.000062s) I0423 08:31:45.628351 132865507964736 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044666s (all_items=0.000011s, per_item={'items': '0.00001097'}, temp_paths=0.044655) I0423 08:31:45.629071 132865507964736 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/9. I0423 08:31:45.629338 132738514417408 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:51:45.629314 I0423 08:31:45.633088 132865507964736 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0423 08:31:45.633348 132737943975680 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0423 08:31:45.633506 132865507964736 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776933105.2769406, 'wait_for_prev_duration_secs': 7.128715515136719e-05, 'time_between_consecutive_saves_sec': 5.2914347648620605, 'checkpointer_blocking_start_time': 1776933105.2787247, 'checkpointer_blocking_duration_secs': 0.35071372985839844, 'get_old_steps_start_time': 1776933105.629457, 'get_old_steps_duration_secs': 2.5510787963867188e-05, 'checkpoint_manager_blocking_start_time': 1776933105.2751257, 'checkpoint_manager_blocking_duration_secs': 0.3583500385284424} I0423 08:31:45.633614 132865507964736 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0423 08:31:45.633661 132865507964736 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 08:31:50.682037 132739529664256 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_3 I0423 08:31:56.145634 132738514417408 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 149.613 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.557860374450684 s) (per-host) I0423 08:31:56.145747 132738514417408 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.516331s. I0423 08:32:02.120198 132738514417408 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 16.490766s. I0423 08:32:02.120468 132737943975680 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 08:32:02.120585 132737943975680 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0423 08:32:02.120626 132737943975680 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:32:02.122262 132737943975680 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0423 08:32:02.122420 132865507964736 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0423 08:32:02.122595 132865507964736 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:32:02.123493 132865507964736 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.922, Tokens/s/device: 13855.438, total_weights: 65536, loss: 7.181, lm_loss: 7.181, perplexity: 1314.495 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Thu Apr 23 08:32:11 UTC 2026 EXIT_CODE=0
XPK Start: Thu Apr 23 13:56:09 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-23 13:56:34.706911: 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) I0423 13:56:34.918933 132816453080896 max_utils.py:273] Attempting to initialize the jax distributed system... I0423 13:56:43.959080 132816453080896 distributed.py:149] Starting JAX distributed service on [::]:8482 I0423 13:56:43.961531 132816453080896 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-auqvv-slice-job-0-0.mt-15-ocdbt-false-sv-auqvv:8482 I0423 13:56:44.975193 132816453080896 max_utils.py:284] Jax distributed system initialized! I0423 13:56:51.072948 132816453080896 max_utils.py:800] System Information: Jax Version: 0.9.2 I0423 13:56:51.073052 132816453080896 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0423 13:56:51.073092 132816453080896 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0423 13:56:51.073127 132816453080896 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 13:56:51.775915 132816453080896 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 13:56:51.776531 132816453080896 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 13:56:51.776721 132816453080896 checkpointing.py:688] Setting up checkpoint logger... I0423 13:56:51.776775 132816453080896 checkpointing.py:234] Creating checkpoint manager with ocdbt=False and zarr3=True I0423 13:56:51.776818 132816453080896 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0423 13:56:51.777174 132816453080896 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x78cb2411cda0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 13:56:54.702532 132816453080896 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0423 13:56:54.702775 132816453080896 checkpoint_manager.py:708] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78b6e8389d00>}, handler_registry=None I0423 13:56:54.703026 132816453080896 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78b6e8389d00>` 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`. I0423 13:56:54.703077 132816453080896 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b6e838c6b0>` 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`. I0423 13:56:54.703114 132816453080896 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 0x78b6e8389d00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78b6e8389d00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b6e838c6b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b6e838c6b0>}). I0423 13:56:54.703444 132816453080896 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0423 13:56:54.703514 132816453080896 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x78b6e81c9e40> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0423 13:56:56.031993 132816453080896 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints I0423 13:56:56.041507 132816453080896 checkpoint_manager.py:929] [process=3][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_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, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x78b6e838b140> I0423 13:56:56.041629 132816453080896 checkpointing.py:302] Checkpoint manager created! I0423 13:56:56.971252 132816453080896 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 13:56:56.971361 132816453080896 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 13:56:57.353807 132816453080896 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 13:56:57.353903 132816453080896 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 13:56:57.370675 132816453080896 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 13:56:57.370736 132816453080896 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 13:56:57.395138 132816453080896 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 13:56:57.395206 132816453080896 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 13:56:57.412013 132816453080896 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 13:56:57.412085 132816453080896 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 13:56:57.428977 132816453080896 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 13:56:57.429040 132816453080896 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 13:56:57.454619 132816453080896 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0423 13:56:57.454688 132816453080896 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 13:56:57.476163 132816453080896 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 13:56:57.476228 132816453080896 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0423 13:56:57.696526 132816453080896 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0423 13:56:57.696633 132816453080896 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0423 13:56:59.131804 132816453080896 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0423 13:56:59.131947 132816453080896 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 13:56:59.132005 132816453080896 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 13:56:59.132064 132816453080896 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0423 13:56:59.132118 132816453080896 maxtext_utils.py:1880] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 13:56:59.132163 132816453080896 maxtext_utils.py:1880] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 13:56:59.132216 132816453080896 maxtext_utils.py:1880] 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) I0423 13:56:59.132269 132816453080896 maxtext_utils.py:1880] 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') I0423 13:56:59.132308 132816453080896 maxtext_utils.py:1880] 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) I0423 13:56:59.132345 132816453080896 maxtext_utils.py:1880] 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) I0423 13:56:59.132392 132816453080896 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 13:56:59.132445 132816453080896 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 13:56:59.626501 132816453080896 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 13:56:59.626594 132816453080896 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0423 13:56:59.642208 132816453080896 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 13:56:59.642268 132816453080896 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0423 13:57:10.599140 132816453080896 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0423 13:57:10.600039 132816453080896 metric_logger.py:301] number parameters: 1.104 billion I0423 13:57:22.433167 132816453080896 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0423 13:57:22.586005 132816453080896 checkpointing.py:798] Waited 0.15282082557678223 seconds for step 0 to finish before starting checkpointing. I0423 13:57:22.588468 132816453080896 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 13:57:22.592043 132816453080896 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0 I0423 13:57:22.593428 132816453080896 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/0. I0423 13:57:23.706483 132816453080896 signaling_client.py:364] Using JaxDistributedSignalingClient I0423 13:57:23.707674 132816453080896 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 13:57:23.707733 132816453080896 replica_slices.py:424] 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 I0423 13:57:23.980741 132816453080896 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274338s I0423 13:57:23.980913 132816453080896 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 2.197 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.7021822929382324 s) (per-host) I0423 13:57:23.980998 132816453080896 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.702273s (batch_requests_ready=0.424549s, total_serialization_initiated=0.277627s, others=0.000098s) I0423 13:57:23.981122 132816453080896 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.706576s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.706558) I0423 13:57:23.981958 132816453080896 event_tracking.py:125] [process=3] [async] Finished blocking save in 1.39 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/0. I0423 13:57:23.982239 132687931086592 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:17:23.982208 I0423 13:57:23.993727 132816453080896 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0423 13:57:23.993986 132687386076928 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0423 13:57:23.994106 132816453080896 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776952642.58845, 'wait_for_prev_duration_secs': 6.842613220214844e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776952642.5920827, 'checkpointer_blocking_duration_secs': 1.3902993202209473, 'get_old_steps_start_time': 1776952643.9824066, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776952642.5865319, 'checkpoint_manager_blocking_duration_secs': 1.407536506652832} I0423 13:57:23.994209 132816453080896 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0423 13:57:23.994259 132816453080896 max_utils.py:750] Memstats: After params initialized: I0423 13:57:23.994311 132816453080896 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0)) I0423 13:57:23.994352 132816453080896 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0)) I0423 13:57:23.994382 132816453080896 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0)) I0423 13:57:23.994407 132816453080896 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0)) I0423 13:57:24.311684 132816453080896 metric_logger.py:196] completed step: 0, seconds: 11.833, TFLOP/s/device: 1.148, Tokens/s/device: 173.075, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 13:57:24.498743 132816453080896 metric_logger.py:196] completed step: 1, seconds: 1.877, TFLOP/s/device: 7.239, Tokens/s/device: 1091.138, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 13:57:24.901271 132816453080896 metric_logger.py:196] completed step: 2, seconds: 0.040, TFLOP/s/device: 341.025, Tokens/s/device: 51403.042, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0423 13:57:25.048826 132816453080896 metric_logger.py:196] completed step: 3, seconds: 0.403, TFLOP/s/device: 33.748, Tokens/s/device: 5086.897, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0423 13:57:25.053379 132816453080896 checkpointing.py:794] Waiting for step 5 to finish before checkpoint... I0423 13:57:25.342626 132816453080896 checkpointing.py:798] Waited 0.28920865058898926 seconds for step 5 to finish before starting checkpointing. I0423 13:57:25.345720 132816453080896 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 13:57:26.676331 2834 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 13:57:28.697882 132687394469632 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_3 I0423 13:57:33.693812 132687931086592 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 151.665 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.415042638778687 s) (per-host) I0423 13:57:33.693956 132687931086592 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 9.711607s. I0423 13:57:40.289868 132687931086592 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 16.307503s. I0423 13:57:40.290114 132687386076928 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 13:57:40.290172 132687386076928 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0423 13:57:40.290219 132687386076928 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0423 13:57:40.292845 132687386076928 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0423 13:57:40.293070 132816453080896 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0423 13:57:40.293205 132816453080896 checkpoint_manager.py:1452] Waiting for previous save to complete took 14.947483 seconds. If this number is high, consider checkpointing less frequently. I0423 13:57:40.295195 132816453080896 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 5 I0423 13:57:40.297733 132816453080896 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/5. I0423 13:57:41.004124 132816453080896 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 13:57:41.004215 132816453080896 replica_slices.py:424] 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 I0423 13:57:41.037636 132816453080896 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034684s I0423 13:57:41.037803 132816453080896 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 40.472 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03811502456665039 s) (per-host) I0423 13:57:41.037852 132816453080896 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038175s (batch_requests_ready=0.001781s, total_serialization_initiated=0.036323s, others=0.000071s) I0423 13:57:41.037961 132816453080896 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043393s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.043378) I0423 13:57:41.038611 132816453080896 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/5. I0423 13:57:41.038969 132687386076928 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:17:41.038921 I0423 13:57:41.048634 132816453080896 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0423 13:57:41.048998 132687436179200 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0423 13:57:41.049158 132816453080896 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776952645.3456914, 'wait_for_prev_duration_secs': 14.94748306274414, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776952660.2952325, 'checkpointer_blocking_duration_secs': 0.7438380718231201, 'get_old_steps_start_time': 1776952661.0390923, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1776952645.3432226, 'checkpoint_manager_blocking_duration_secs': 15.705901384353638} I0423 13:57:41.049268 132816453080896 checkpointing.py:409] Started an asynchronous checkpoint save for step 5 I0423 13:57:41.050038 132816453080896 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.592, Tokens/s/device: 13353.590, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0423 13:57:41.056166 132816453080896 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.218, Tokens/s/device: 13900.107, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0423 13:57:46.475088 132687394469632 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_3 I0423 13:57:51.901888 132687386076928 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 144.888 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.902162313461304 s) (per-host) I0423 13:57:51.902020 132687386076928 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.862424s. I0423 13:57:58.396356 132687386076928 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 17.356744s. I0423 13:57:58.396624 132687436179200 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 13:57:58.396749 132687436179200 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0423 13:57:58.396796 132687436179200 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0423 13:57:58.398159 132687436179200 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0423 13:58:04.009137 132816453080896 metric_logger.py:196] completed step: 6, seconds: 16.001, TFLOP/s/device: 0.849, Tokens/s/device: 127.989, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0423 13:58:04.156649 132816453080896 metric_logger.py:196] completed step: 7, seconds: 22.807, TFLOP/s/device: 0.596, Tokens/s/device: 89.797, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0423 13:58:04.304018 132816453080896 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.296, Tokens/s/device: 13459.693, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0423 13:58:04.450999 132816453080896 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0423 13:58:04.451682 132816453080896 checkpointing.py:798] Waited 0.0007052421569824219 seconds for step 9 to finish before starting checkpointing. I0423 13:58:04.453943 132816453080896 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 13:58:04.455588 132816453080896 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 9 I0423 13:58:04.456826 132816453080896 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/9. I0423 13:58:04.761214 132816453080896 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 13:58:04.761312 132816453080896 replica_slices.py:424] 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 I0423 13:58:04.797045 132816453080896 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036969s I0423 13:58:04.797209 132816453080896 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 38.201 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.040380001068115234 s) (per-host) I0423 13:58:04.797259 132816453080896 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040440s (batch_requests_ready=0.001755s, total_serialization_initiated=0.038620s, others=0.000064s) I0423 13:58:04.797349 132816453080896 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044447s (all_items=0.000010s, per_item={'items': '0.00001049'}, temp_paths=0.044436) I0423 13:58:04.798033 132816453080896 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/9. I0423 13:58:04.798362 132687436179200 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:18:04.798326 I0423 13:58:04.800292 132816453080896 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0423 13:58:04.800503 132665122465536 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0423 13:58:04.800616 132816453080896 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776952684.4539015, 'wait_for_prev_duration_secs': 8.153915405273438e-05, 'time_between_consecutive_saves_sec': 6.055705308914185, 'checkpointer_blocking_start_time': 1776952684.4556258, 'checkpointer_blocking_duration_secs': 0.342881441116333, 'get_old_steps_start_time': 1776952684.7985244, 'get_old_steps_duration_secs': 2.3603439331054688e-05, 'checkpoint_manager_blocking_start_time': 1776952684.4519055, 'checkpoint_manager_blocking_duration_secs': 0.3486802577972412} I0423 13:58:04.800719 132816453080896 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0423 13:58:04.800771 132816453080896 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 13:58:09.842188 132665399293696 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_3 I0423 13:58:15.247792 132687436179200 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 150.568 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.490924835205078 s) (per-host) I0423 13:58:15.247915 132687436179200 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.449442s. I0423 13:58:21.540500 132687436179200 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 16.742012s. I0423 13:58:21.540725 132665122465536 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 13:58:21.540779 132665122465536 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0423 13:58:21.540818 132665122465536 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0423 13:58:21.542398 132665122465536 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0423 13:58:21.542561 132816453080896 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0423 13:58:21.542717 132816453080896 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 13:58:21.543678 132816453080896 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.220, Tokens/s/device: 13900.390, total_weights: 65536, loss: 7.181, lm_loss: 7.181, perplexity: 1314.495 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Thu Apr 23 13:58:29 UTC 2026 EXIT_CODE=0