main (59e0f1759) vs feat/nnx-post-train-fixes (28b1e4a16)| Metric | main 59e0f1759 | feat/nnx-post-train-fixes 28b1e4a16 | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 7.1810 | 7.1810 | 0 |
| TFLOP/s | 92.061 | 92.139 | +0.078 |
| Tok/s | 13876.5 | 13888.1 | +11.668 |
| Avg s/step | 4.936 | 4.486 | -0.45 |
| 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: Sat Apr 25 08:17:34 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. 2026-04-25 08:17:59.354071: 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) I0425 08:17:59.558896 134917365917504 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 08:18:08.600152 134917365917504 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 08:18:08.602445 134917365917504 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-2d7ta-slice-job-0-0.mt-15-ocdbt-false-sv-2d7ta:8482 I0425 08:18:09.887225 134917365917504 max_utils.py:284] Jax distributed system initialized! I0425 08:18:14.992676 134917365917504 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 08:18:14.992781 134917365917504 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 08:18:14.992820 134917365917504 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 I0425 08:18:14.992854 134917365917504 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 08:18:15.684124 134917365917504 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1) I0425 08:18:15.684412 134917365917504 checkpointing.py:677] Setting up checkpoint logger... I0425 08:18:15.684466 134917365917504 checkpointing.py:233] Creating checkpoint manager with ocdbt=False and zarr3=True I0425 08:18:15.684512 134917365917504 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 08:18:15.684863 134917365917504 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 0x7ab451b14ef0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 08:18:18.550119 134917365917504 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0425 08:18:18.550360 134917365917504 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7aa0a074ba70>}, handler_registry=None I0425 08:18:18.550592 134917365917504 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7aa0a074ba70>` 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`. I0425 08:18:18.550642 134917365917504 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa0a074db50>` 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`. I0425 08:18:18.550680 134917365917504 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 0x7aa0a074ba70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7aa0a074ba70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa0a074db50>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa0a074db50>}). I0425 08:18:18.551014 134917365917504 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 08:18:18.551084 134917365917504 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7aa0a04cda80> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 08:18:19.418244 134917365917504 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints I0425 08:18:19.429821 134917365917504 checkpoint_manager.py:929] [process=6][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_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7aa0a074c3e0> I0425 08:18:19.429942 134917365917504 checkpointing.py:301] Checkpoint manager created! I0425 08:18:20.360332 134917365917504 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 08:18:20.360459 134917365917504 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 08:18:20.750126 134917365917504 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 08:18:20.750226 134917365917504 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 08:18:20.766751 134917365917504 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 08:18:20.766812 134917365917504 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 08:18:20.790678 134917365917504 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 08:18:20.790746 134917365917504 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 08:18:20.807266 134917365917504 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 08:18:20.807326 134917365917504 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 08:18:20.823752 134917365917504 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 08:18:20.823813 134917365917504 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 08:18:20.848592 134917365917504 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 08:18:20.848663 134917365917504 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 08:18:20.870069 134917365917504 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 08:18:20.870150 134917365917504 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 08:18:21.079671 134917365917504 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0425 08:18:21.079780 134917365917504 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 08:18:22.508904 134917365917504 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 08:18:22.509029 134917365917504 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) I0425 08:18:22.509082 134917365917504 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) I0425 08:18:22.509165 134917365917504 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') I0425 08:18:22.509219 134917365917504 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) I0425 08:18:22.509271 134917365917504 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) I0425 08:18:22.509327 134917365917504 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) I0425 08:18:22.509378 134917365917504 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') I0425 08:18:22.509418 134917365917504 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) I0425 08:18:22.509454 134917365917504 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) I0425 08:18:22.509501 134917365917504 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 08:18:22.509549 134917365917504 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 08:18:23.004960 134917365917504 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 08:18:23.005054 134917365917504 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0425 08:18:23.021314 134917365917504 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 08:18:23.021373 134917365917504 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0425 08:18:34.099685 134917365917504 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. I0425 08:18:34.100540 134917365917504 metric_logger.py:301] number parameters: 1.104 billion I0425 08:18:45.802470 134917365917504 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0425 08:18:45.954874 134917365917504 checkpointing.py:776] Waited 0.1523878574371338 seconds for step 0 to finish before starting checkpointing. I0425 08:18:45.957279 134917365917504 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 08:18:45.958943 134917365917504 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0425 08:18:45.960393 134917365917504 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/0. I0425 08:18:46.285557 134917365917504 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 08:18:46.286504 134917365917504 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 08:18:46.286566 134917365917504 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 I0425 08:18:46.558972 134917365917504 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273491s I0425 08:18:46.559165 134917365917504 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.525 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2792017459869385 s) (per-host) I0425 08:18:46.559219 134917365917504 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279266s (batch_requests_ready=0.002394s, total_serialization_initiated=0.276800s, others=0.000072s) I0425 08:18:46.559320 134917365917504 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283444s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.283426) I0425 08:18:46.560217 134917365917504 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.60 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/0. I0425 08:18:46.560570 134790082713344 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:38:46.560526 I0425 08:18:46.573541 134917365917504 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 08:18:46.573805 134787405178624 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 08:18:46.573955 134917365917504 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777105125.95726, 'wait_for_prev_duration_secs': 6.270408630371094e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777105125.95898, 'checkpointer_blocking_duration_secs': 0.6017448902130127, 'get_old_steps_start_time': 1777105126.5607505, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1777105125.9553263, 'checkpoint_manager_blocking_duration_secs': 0.6185917854309082} I0425 08:18:46.574061 134917365917504 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0425 08:18:46.574123 134917365917504 max_utils.py:750] Memstats: After params initialized: I0425 08:18:46.574172 134917365917504 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0425 08:18:46.574208 134917365917504 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0425 08:18:46.574233 134917365917504 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0425 08:18:46.574256 134917365917504 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0425 08:18:46.890147 134917365917504 metric_logger.py:196] completed step: 0, seconds: 11.702, TFLOP/s/device: 1.161, Tokens/s/device: 175.016, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0425 08:18:47.065730 134917365917504 metric_logger.py:196] completed step: 1, seconds: 1.086, TFLOP/s/device: 12.510, Tokens/s/device: 1885.615, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0425 08:18:47.478807 134917365917504 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 478.386, Tokens/s/device: 72107.598, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0425 08:18:47.626318 134917365917504 metric_logger.py:196] completed step: 3, seconds: 0.413, TFLOP/s/device: 32.873, Tokens/s/device: 4954.915, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0425 08:18:47.630716 134917365917504 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0425 08:18:47.920388 134917365917504 checkpointing.py:776] Waited 0.28963494300842285 seconds for step 5 to finish before starting checkpointing. I0425 08:18:47.923334 134917365917504 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 08:18:54.684355 2800 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 08:18:57.115932 134787413571328 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_6 I0425 08:19:01.708397 134790082713344 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 102.382 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 15.428394794464111 s) (per-host) I0425 08:19:01.708522 134790082713344 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 15.147842s. I0425 08:19:06.892085 134790082713344 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 20.331388s. I0425 08:19:06.892387 134787405178624 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 08:19:06.892522 134787405178624 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 08:19:06.892572 134787405178624 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 08:19:06.894491 134787405178624 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 08:19:06.894676 134917365917504 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0425 08:19:06.894817 134917365917504 checkpoint_manager.py:1452] Waiting for previous save to complete took 18.971490 seconds. If this number is high, consider checkpointing less frequently. I0425 08:19:06.896965 134917365917504 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5 I0425 08:19:06.899485 134917365917504 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/5. I0425 08:19:07.601954 134917365917504 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 08:19:07.602044 134917365917504 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 I0425 08:19:07.642388 134917365917504 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041609s I0425 08:19:07.642556 134917365917504 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 34.126 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04520273208618164 s) (per-host) I0425 08:19:07.642605 134917365917504 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045261s (batch_requests_ready=0.001933s, total_serialization_initiated=0.043263s, others=0.000066s) I0425 08:19:07.642697 134917365917504 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050323s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.050308) I0425 08:19:07.643397 134917365917504 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/5. I0425 08:19:07.643749 134787405178624 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:39:07.643712 I0425 08:19:07.645755 134917365917504 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0425 08:19:07.646052 134790065927936 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 08:19:07.646230 134917365917504 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777105127.9232957, 'wait_for_prev_duration_secs': 18.971489906311035, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777105146.8970053, 'checkpointer_blocking_duration_secs': 0.7468869686126709, 'get_old_steps_start_time': 1777105147.6439142, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1777105127.9209979, 'checkpoint_manager_blocking_duration_secs': 19.725197315216064} I0425 08:19:07.646346 134917365917504 checkpointing.py:408] Started an asynchronous checkpoint save for step 5 I0425 08:19:07.647174 134917365917504 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.833, Tokens/s/device: 13389.909, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0425 08:19:07.653136 134917365917504 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.206, Tokens/s/device: 13898.220, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0425 08:19:12.784091 134787413571328 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_6 I0425 08:19:18.197857 134787405178624 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 149.012 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.600464105606079 s) (per-host) I0425 08:19:18.197981 134787405178624 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.554120s. I0425 08:19:23.787384 134787405178624 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 16.143506s. I0425 08:19:23.787688 134790065927936 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 08:19:23.787810 134790065927936 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 08:19:23.787861 134790065927936 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0425 08:19:23.789408 134790065927936 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0425 08:19:30.072378 134917365917504 metric_logger.py:196] completed step: 6, seconds: 20.021, TFLOP/s/device: 0.679, Tokens/s/device: 102.293, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0425 08:19:30.220152 134917365917504 metric_logger.py:196] completed step: 7, seconds: 22.273, TFLOP/s/device: 0.610, Tokens/s/device: 91.949, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0425 08:19:30.367482 134917365917504 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.194, Tokens/s/device: 13444.231, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0425 08:19:30.514395 134917365917504 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0425 08:19:30.515083 134917365917504 checkpointing.py:776] Waited 0.0007090568542480469 seconds for step 9 to finish before starting checkpointing. I0425 08:19:30.517226 134917365917504 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 08:19:30.519060 134917365917504 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0425 08:19:30.520513 134917365917504 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/9. I0425 08:19:30.864992 134917365917504 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 08:19:30.865087 134917365917504 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 I0425 08:19:30.906602 134917365917504 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042471s I0425 08:19:30.906764 134917365917504 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 33.475 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.046080827713012695 s) (per-host) I0425 08:19:30.906815 134917365917504 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046142s (batch_requests_ready=0.001791s, total_serialization_initiated=0.044286s, others=0.000066s) I0425 08:19:30.906905 134917365917504 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050250s (all_items=0.000010s, per_item={'items': '0.00001025'}, temp_paths=0.050240) I0425 08:19:30.907608 134917365917504 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.39 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/9. I0425 08:19:30.907888 134790065927936 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:39:30.907865 I0425 08:19:30.909758 134917365917504 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0425 08:19:30.910025 134763826329344 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 08:19:30.910170 134917365917504 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777105170.517197, 'wait_for_prev_duration_secs': 7.05718994140625e-05, 'time_between_consecutive_saves_sec': 6.727750062942505, 'checkpointer_blocking_start_time': 1777105170.519111, 'checkpointer_blocking_duration_secs': 0.38887596130371094, 'get_old_steps_start_time': 1777105170.9080057, 'get_old_steps_duration_secs': 2.4318695068359375e-05, 'checkpoint_manager_blocking_start_time': 1777105170.5153325, 'checkpoint_manager_blocking_duration_secs': 0.3948078155517578} I0425 08:19:30.910337 134917365917504 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0425 08:19:30.910387 134917365917504 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 08:19:36.116246 134789554235136 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_6 I0425 08:19:41.608429 134790065927936 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 146.971 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.747706174850464 s) (per-host) I0425 08:19:41.608548 134790065927936 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.700585s. I0425 08:19:47.128921 134790065927936 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 16.220942s. I0425 08:19:47.129259 134763826329344 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 08:19:47.129380 134763826329344 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 08:19:47.129421 134763826329344 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0425 08:19:47.130747 134763826329344 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0425 08:19:47.130929 134917365917504 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0425 08:19:47.131073 134917365917504 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 08:19:47.132110 134917365917504 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.061, Tokens/s/device: 13876.467, 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: Sat Apr 25 08:19:59 UTC 2026 EXIT_CODE=0
XPK Start: Sat Apr 25 13:13:37 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. 2026-04-25 13:14:02.435447: 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) I0425 13:14:02.647994 138737276741440 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 13:14:11.688721 138737276741440 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 13:14:11.694117 138737276741440 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-i6lf3-slice-job-0-0.mt-15-ocdbt-false-sv-i6lf3:8482 I0425 13:14:12.902539 138737276741440 max_utils.py:284] Jax distributed system initialized! I0425 13:14:18.815157 138737276741440 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 13:14:18.815264 138737276741440 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 13:14:18.815306 138737276741440 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 I0425 13:14:18.815343 138737276741440 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 13:14:19.520930 138737276741440 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 13:14:19.521543 138737276741440 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 13:14:19.521728 138737276741440 checkpointing.py:688] Setting up checkpoint logger... I0425 13:14:19.521780 138737276741440 checkpointing.py:234] Creating checkpoint manager with ocdbt=False and zarr3=True I0425 13:14:19.521823 138737276741440 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 13:14:19.522173 138737276741440 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 0x7e2db39144d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 13:14:22.451428 138737276741440 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0425 13:14:22.451665 138737276741440 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e1a640ddfd0>}, handler_registry=None I0425 13:14:22.451907 138737276741440 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e1a640ddfd0>` 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`. I0425 13:14:22.451957 138737276741440 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e1a2c3af2c0>` 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`. I0425 13:14:22.451993 138737276741440 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 0x7e1a640ddfd0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e1a640ddfd0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e1a2c3af2c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e1a2c3af2c0>}). I0425 13:14:22.452339 138737276741440 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 13:14:22.452411 138737276741440 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e1a2c37de40> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 13:14:24.109220 138737276741440 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints I0425 13:14:24.111525 138737276741440 checkpoint_manager.py:929] [process=6][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_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e1a2c3aacc0> I0425 13:14:24.111639 138737276741440 checkpointing.py:302] Checkpoint manager created! I0425 13:14:25.061048 138737276741440 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 13:14:25.061187 138737276741440 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:14:25.443875 138737276741440 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 13:14:25.443973 138737276741440 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:14:25.460458 138737276741440 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 13:14:25.460521 138737276741440 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:14:25.484542 138737276741440 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 13:14:25.484609 138737276741440 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:14:25.501194 138737276741440 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 13:14:25.501264 138737276741440 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:14:25.517747 138737276741440 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 13:14:25.517811 138737276741440 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:14:25.542664 138737276741440 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 13:14:25.542736 138737276741440 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:14:25.563675 138737276741440 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 13:14:25.563741 138737276741440 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 13:14:25.785574 138737276741440 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0425 13:14:25.785682 138737276741440 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 13:14:27.234627 138737276741440 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 13:14:27.234755 138737276741440 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) I0425 13:14:27.234810 138737276741440 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) I0425 13:14:27.234867 138737276741440 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') I0425 13:14:27.234923 138737276741440 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) I0425 13:14:27.234965 138737276741440 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) I0425 13:14:27.235019 138737276741440 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) I0425 13:14:27.235073 138737276741440 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') I0425 13:14:27.235126 138737276741440 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) I0425 13:14:27.235165 138737276741440 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) I0425 13:14:27.235214 138737276741440 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 13:14:27.235266 138737276741440 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 13:14:27.737229 138737276741440 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 13:14:27.737331 138737276741440 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0425 13:14:27.752833 138737276741440 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 13:14:27.752892 138737276741440 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0425 13:14:38.906746 138737276741440 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. I0425 13:14:38.907598 138737276741440 metric_logger.py:301] number parameters: 1.104 billion I0425 13:14:50.377060 138737276741440 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0425 13:14:50.530886 138737276741440 checkpointing.py:798] Waited 0.15381073951721191 seconds for step 0 to finish before starting checkpointing. I0425 13:14:50.533378 138737276741440 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 13:14:50.535158 138737276741440 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0425 13:14:50.537201 138737276741440 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/0. I0425 13:14:51.279805 138737276741440 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 13:14:51.280946 138737276741440 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 13:14:51.281006 138737276741440 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 I0425 13:14:51.558539 138737276741440 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278813s I0425 13:14:51.558721 138737276741440 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 2.190 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.7044885158538818 s) (per-host) I0425 13:14:51.558778 138737276741440 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.704555s (batch_requests_ready=0.422420s, total_serialization_initiated=0.282059s, others=0.000076s) I0425 13:14:51.558871 138737276741440 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.709036s (all_items=0.000048s, per_item={'items': '0.00004792'}, temp_paths=0.708988) I0425 13:14:51.559729 138737276741440 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.02 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/0. I0425 13:14:51.560074 138611744921344 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 13:34:51.560037 I0425 13:14:51.567649 138737276741440 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:14:51.567965 138610091214592 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:14:51.568136 138737276741440 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777122890.5333598, 'wait_for_prev_duration_secs': 6.29425048828125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777122890.5351956, 'checkpointer_blocking_duration_secs': 1.0250425338745117, 'get_old_steps_start_time': 1777122891.5602667, 'get_old_steps_duration_secs': 3.719329833984375e-05, 'checkpoint_manager_blocking_start_time': 1777122890.5313902, 'checkpoint_manager_blocking_duration_secs': 1.036694049835205} I0425 13:14:51.568246 138737276741440 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0425 13:14:51.568302 138737276741440 max_utils.py:750] Memstats: After params initialized: I0425 13:14:51.568352 138737276741440 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0425 13:14:51.568386 138737276741440 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0425 13:14:51.568415 138737276741440 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0425 13:14:51.568440 138737276741440 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0425 13:14:51.881191 138737276741440 metric_logger.py:196] completed step: 0, seconds: 11.469, TFLOP/s/device: 1.185, Tokens/s/device: 178.563, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0425 13:14:52.063807 138737276741440 metric_logger.py:196] completed step: 1, seconds: 1.503, TFLOP/s/device: 9.042, Tokens/s/device: 1362.974, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0425 13:14:52.488410 138737276741440 metric_logger.py:196] completed step: 2, seconds: 0.035, TFLOP/s/device: 384.665, Tokens/s/device: 57980.862, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0425 13:14:52.635706 138737276741440 metric_logger.py:196] completed step: 3, seconds: 0.425, TFLOP/s/device: 31.969, Tokens/s/device: 4818.755, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0425 13:14:52.640249 138737276741440 checkpointing.py:794] Waiting for step 5 to finish before checkpoint... I0425 13:14:52.929895 138737276741440 checkpointing.py:798] Waited 0.2896106243133545 seconds for step 5 to finish before starting checkpointing. I0425 13:14:52.933027 138737276741440 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 13:14:54.056571 2757 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 13:14:56.200458 138610099607296 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_6 I0425 13:15:01.337502 138611744921344 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 150.678 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.483237266540527 s) (per-host) I0425 13:15:01.337613 138611744921344 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 9.777416s. I0425 13:15:07.933146 138611744921344 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 16.372934s. I0425 13:15:07.933433 138610091214592 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:15:07.933567 138610091214592 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:15:07.933618 138610091214592 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:15:07.935198 138610091214592 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 13:15:07.935466 138737276741440 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0425 13:15:07.935603 138737276741440 checkpoint_manager.py:1452] Waiting for previous save to complete took 15.002577 seconds. If this number is high, consider checkpointing less frequently. I0425 13:15:07.937584 138737276741440 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5 I0425 13:15:07.940114 138737276741440 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/5. I0425 13:15:08.221750 138737276741440 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 13:15:08.221847 138737276741440 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 I0425 13:15:08.257265 138737276741440 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036662s I0425 13:15:08.257422 138737276741440 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 38.377 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04019570350646973 s) (per-host) I0425 13:15:08.257474 138737276741440 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040258s (batch_requests_ready=0.001879s, total_serialization_initiated=0.038309s, others=0.000070s) I0425 13:15:08.257563 138737276741440 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.045246s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.045230) I0425 13:15:08.258283 138737276741440 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/5. I0425 13:15:08.258582 138610091214592 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 13:35:08.258549 I0425 13:15:08.266172 138737276741440 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:15:08.266448 138611728135936 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:15:08.266615 138737276741440 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777122892.9329948, 'wait_for_prev_duration_secs': 15.00257682800293, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777122907.9376237, 'checkpointer_blocking_duration_secs': 0.3210937976837158, 'get_old_steps_start_time': 1777122908.2587395, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1777122892.930534, 'checkpoint_manager_blocking_duration_secs': 15.336041450500488} I0425 13:15:08.266726 138737276741440 checkpointing.py:409] Started an asynchronous checkpoint save for step 5 I0425 13:15:08.267580 138737276741440 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.758, Tokens/s/device: 13378.626, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0425 13:15:08.273395 138737276741440 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.379, Tokens/s/device: 13924.300, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0425 13:15:13.310748 138610099607296 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_6 I0425 13:15:18.605246 138610091214592 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 152.060 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.387979745864868 s) (per-host) I0425 13:15:18.605383 138610091214592 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.346692s. I0425 13:15:24.048469 138610091214592 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 15.789762s. I0425 13:15:24.048762 138611728135936 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:15:24.048882 138611728135936 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:15:24.048931 138611728135936 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:15:24.050504 138611728135936 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0425 13:15:30.595153 138737276741440 metric_logger.py:196] completed step: 6, seconds: 15.632, TFLOP/s/device: 0.869, Tokens/s/device: 131.016, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0425 13:15:30.742929 138737276741440 metric_logger.py:196] completed step: 7, seconds: 22.176, TFLOP/s/device: 0.613, Tokens/s/device: 92.353, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0425 13:15:30.890209 138737276741440 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.391, Tokens/s/device: 13473.950, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0425 13:15:31.037065 138737276741440 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0425 13:15:31.037735 138737276741440 checkpointing.py:798] Waited 0.0006897449493408203 seconds for step 9 to finish before starting checkpointing. I0425 13:15:31.039970 138737276741440 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 13:15:31.042831 138737276741440 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0425 13:15:31.044180 138737276741440 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/9. I0425 13:15:31.348066 138737276741440 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 13:15:31.348174 138737276741440 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 I0425 13:15:31.383061 138737276741440 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036091s I0425 13:15:31.383218 138737276741440 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 38.965 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03958868980407715 s) (per-host) I0425 13:15:31.383267 138737276741440 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.039647s (batch_requests_ready=0.001867s, total_serialization_initiated=0.037717s, others=0.000064s) I0425 13:15:31.383362 138737276741440 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043672s (all_items=0.000011s, per_item={'items': '0.00001121'}, temp_paths=0.043661) I0425 13:15:31.384037 138737276741440 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/9. I0425 13:15:31.384324 138605661050624 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 13:35:31.384292 I0425 13:15:31.386385 138737276741440 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:15:31.386639 138605669443328 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:15:31.386743 138737276741440 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777122931.0399394, 'wait_for_prev_duration_secs': 7.462501525878906e-05, 'time_between_consecutive_saves_sec': 6.989393472671509, 'checkpointer_blocking_start_time': 1777122931.0428712, 'checkpointer_blocking_duration_secs': 0.34160447120666504, 'get_old_steps_start_time': 1777122931.3844993, 'get_old_steps_duration_secs': 2.6702880859375e-05, 'checkpoint_manager_blocking_start_time': 1777122931.0379589, 'checkpoint_manager_blocking_duration_secs': 0.3487508296966553} I0425 13:15:31.386925 138737276741440 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0425 13:15:31.386975 138737276741440 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 13:15:36.500219 138584115877632 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_6 I0425 13:15:41.899813 138605661050624 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 149.638 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.556148290634155 s) (per-host) I0425 13:15:41.899925 138605661050624 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.515486s. I0425 13:15:46.919816 138605661050624 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 15.535357s. I0425 13:15:46.920033 138605669443328 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:15:46.920122 138605669443328 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:15:46.920169 138605669443328 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:15:46.922770 138605669443328 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0425 13:15:46.922965 138737276741440 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0425 13:15:46.923137 138737276741440 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 13:15:46.924076 138737276741440 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.139, Tokens/s/device: 13888.135, 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: Sat Apr 25 13:15:57 UTC 2026 EXIT_CODE=0