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 | 5.3910 | 5.3910 | 0 |
| TFLOP/s | 92.204 | 92.269 | +0.065 |
| Tok/s | 13898.0 | 13907.8 | +9.721 |
| Avg s/step | 8.362 | 8.399 | +0.037 |
| 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 08:12:30 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 08:12:55.303925: 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 08:12:55.507719 140013575620416 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 08:13:04.549652 140013575620416 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 08:13:04.551973 140013575620416 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-true-rs-ebhla-slice-job-0-0.mt-15-ocdbt-true-rs-ebhla:8482 I0425 08:13:05.701002 140013575620416 max_utils.py:284] Jax distributed system initialized! I0425 08:13:11.746286 140013575620416 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 08:13:11.746393 140013575620416 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 08:13:11.746433 140013575620416 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 08:13:11.746467 140013575620416 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 08:13:12.433716 140013575620416 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1) I0425 08:13:12.434021 140013575620416 checkpointing.py:677] Setting up checkpoint logger... I0425 08:13:12.434090 140013575620416 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 08:13:12.434168 140013575620416 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 08:13:12.434542 140013575620416 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7f56d77181a0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 08:13:13.691033 140013575620416 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0425 08:13:13.691293 140013575620416 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 0x7f42c444b350>}, handler_registry=None I0425 08:13:13.691539 140013575620416 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f42c444b350>` 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 08:13:13.691588 140013575620416 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f42c444cf20>` 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 08:13:13.691623 140013575620416 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 0x7f42c444b350>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f42c444b350>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f42c444cf20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f42c444cf20>}). I0425 08:13:13.691994 140013575620416 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 08:13:13.692076 140013575620416 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f42c41d5a80> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 08:13:15.185477 139892612900608 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777104629642286619, 'commit_timestamp_nsecs': 1777104674832006187, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints/9/_CHECKPOINT_METADATA I0425 08:13:15.457513 139892621293312 checkpoint_manager.py:1871] Missing metrics for step 5 I0425 08:13:15.865164 140013575620416 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints I0425 08:13:15.903348 140013575620416 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_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f42c444c620> I0425 08:13:15.903490 140013575620416 checkpointing.py:301] Checkpoint manager created! I0425 08:13:16.840279 140013575620416 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 08:13:16.840418 140013575620416 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 08:13:17.220350 140013575620416 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 08:13:17.220443 140013575620416 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 08:13:17.237061 140013575620416 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 08:13:17.237164 140013575620416 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 08:13:17.261290 140013575620416 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 08:13:17.261359 140013575620416 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 08:13:17.278159 140013575620416 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 08:13:17.278223 140013575620416 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 08:13:17.294834 140013575620416 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 08:13:17.294896 140013575620416 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 08:13:17.320255 140013575620416 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 08:13:17.320327 140013575620416 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 08:13:17.341977 140013575620416 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 08:13:17.342041 140013575620416 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 08:13:17.555547 140013575620416 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0425 08:13:17.555659 140013575620416 checkpointing.py:581] restoring from this run's directory step 9 I0425 08:13:17.807362 140013575620416 event_tracking.py:70] [process=6] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints/9. I0425 08:13:18.048168 140013575620416 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints/9. I0425 08:13:18.048625 140013575620416 event_tracking.py:125] [process=6] [sync] Finished blocking load in 0.24 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints/9. I0425 08:13:19.446868 1451 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 08:13:25.163980 140013575620416 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 276.129 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 5.722959041595459 s) (per-host) I0425 08:13:25.169765 140013575620416 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 257.495 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.146122217178345 s) (per-host) I0425 08:13:25.448882 140013575620416 event_tracking.py:138] [process=6] [sync] Finished load in 7.64 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints/9 I0425 08:13:25.449040 140013575620416 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints', 'checkpointer_start_time': 1777104797.8072608, 'checkpointer_duration_secs': 7.641745328903198, 'checkpoint_manager_start_time': 1777104797.5559042, 'checkpoint_manager_duration_secs': 7.893104314804077} fsdp: 32 I0425 08:13:25.871117 140013575620416 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 08:13:25.871251 140013575620416 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 08:13:25.871307 140013575620416 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 08:13:25.871368 140013575620416 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 08:13:25.871422 140013575620416 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 08:13:25.871460 140013575620416 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 08:13:25.871513 140013575620416 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 08:13:25.871567 140013575620416 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 08:13:25.871607 140013575620416 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 08:13:25.871642 140013575620416 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 08:13:25.871690 140013575620416 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 08:13:25.871738 140013575620416 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 08:13:26.362912 140013575620416 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 08:13:26.363008 140013575620416 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0425 08:13:26.378667 140013575620416 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 08:13:26.378726 140013575620416 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0425 08:13:37.392528 140013575620416 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 08:13:37.393362 140013575620416 metric_logger.py:301] number parameters: 1.104 billion I0425 08:13:48.978393 140013575620416 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0425 08:13:49.174041 140013575620416 checkpointing.py:776] Waited 0.19562983512878418 seconds for step 10 to finish before starting checkpointing. I0425 08:13:49.177947 140013575620416 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 08:13:49.179471 140013575620416 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10 I0425 08:13:49.180994 140013575620416 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_true/checkpoints/10. I0425 08:13:49.497717 140013575620416 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 08:13:49.498860 140013575620416 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 08:13:49.498916 140013575620416 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 08:13:49.781090 140013575620416 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.283448s I0425 08:13:49.781280 140013575620416 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.369 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28731369972229004 s) (per-host) I0425 08:13:49.781332 140013575620416 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.287374s (batch_requests_ready=0.002158s, total_serialization_initiated=0.285148s, others=0.000068s) I0425 08:13:49.781419 140013575620416 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.291467s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.291449) I0425 08:13:49.782164 140013575620416 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_true/checkpoints/10. I0425 08:13:49.782464 139892017313536 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:33:49.782430 I0425 08:13:49.791451 140013575620416 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0425 08:13:49.791774 139890817746688 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 08:13:49.791932 140013575620416 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777104829.1779282, 'wait_for_prev_duration_secs': 6.389617919921875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777104829.1795151, 'checkpointer_blocking_duration_secs': 0.6031055450439453, 'get_old_steps_start_time': 1777104829.7826457, 'get_old_steps_duration_secs': 3.24249267578125e-05, 'checkpoint_manager_blocking_start_time': 1777104829.1745505, 'checkpoint_manager_blocking_duration_secs': 0.6173436641693115} I0425 08:13:49.792044 140013575620416 checkpointing.py:408] Started an asynchronous checkpoint save for step 10 I0425 08:13:49.792108 140013575620416 max_utils.py:750] Memstats: After params initialized: I0425 08:13:49.792158 140013575620416 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0425 08:13:49.792190 140013575620416 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0425 08:13:49.792216 140013575620416 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0425 08:13:49.792241 140013575620416 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0425 08:13:50.104292 140013575620416 metric_logger.py:196] completed step: 10, seconds: 11.585, TFLOP/s/device: 1.173, Tokens/s/device: 176.781, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869 I0425 08:13:50.284712 140013575620416 metric_logger.py:196] completed step: 11, seconds: 1.124, TFLOP/s/device: 12.085, Tokens/s/device: 1821.638, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659 I0425 08:13:50.432295 140013575620416 metric_logger.py:196] completed step: 12, seconds: 0.033, TFLOP/s/device: 410.215, Tokens/s/device: 61832.015, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068 I0425 08:13:50.579753 140013575620416 metric_logger.py:196] completed step: 13, seconds: 0.154, TFLOP/s/device: 88.468, Tokens/s/device: 13334.809, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001 I0425 08:13:50.584292 140013575620416 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0425 08:13:50.873699 140013575620416 checkpointing.py:776] Waited 0.28937244415283203 seconds for step 15 to finish before starting checkpointing. I0425 08:13:50.876801 140013575620416 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 08:13:55.035033 139890922600192 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_true/checkpoints/10/items/array_metadatas/process_6 I0425 08:14:31.083988 139892017313536 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.980 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.58997464179993 s) (per-host) I0425 08:14:31.084137 139892017313536 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.301556s. I0425 08:14:39.716211 139892017313536 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.933614s. I0425 08:14:39.716521 139890817746688 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 08:14:39.716644 139890817746688 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 08:14:39.716692 139890817746688 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0425 08:14:39.718217 139890817746688 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0425 08:14:39.718410 140013575620416 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0425 08:14:39.718546 140013575620416 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.841743 seconds. If this number is high, consider checkpointing less frequently. I0425 08:14:39.720525 140013575620416 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15 I0425 08:14:39.723027 140013575620416 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_true/checkpoints/15. I0425 08:14:40.009442 140013575620416 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 08:14:40.009541 140013575620416 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 08:14:40.036039 140013575620416 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027772s I0425 08:14:40.036220 140013575620416 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 49.414 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.031217575073242188 s) (per-host) I0425 08:14:40.036273 140013575620416 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031279s (batch_requests_ready=0.001767s, total_serialization_initiated=0.029442s, others=0.000070s) I0425 08:14:40.036365 140013575620416 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.036232s (all_items=0.000014s, per_item={'items': '0.00001359'}, temp_paths=0.036219) I0425 08:14:40.037148 140013575620416 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints/15. I0425 08:14:40.037475 139890817746688 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:34:40.037439 I0425 08:14:40.048077 140013575620416 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0425 08:14:40.048379 139888244672256 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 08:14:40.048518 140013575620416 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777104830.8767717, 'wait_for_prev_duration_secs': 48.8417432308197, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777104879.720565, 'checkpointer_blocking_duration_secs': 0.31705522537231445, 'get_old_steps_start_time': 1777104880.0376453, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1777104830.8743536, 'checkpoint_manager_blocking_duration_secs': 49.174129486083984} I0425 08:14:40.048640 140013575620416 checkpointing.py:408] Started an asynchronous checkpoint save for step 15 I0425 08:14:40.049455 140013575620416 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 92.062, Tokens/s/device: 13876.561, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193 I0425 08:14:40.055599 140013575620416 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.278, Tokens/s/device: 13909.075, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661 I0425 08:14:45.281128 139890922600192 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_true/checkpoints/15/items/array_metadatas/process_6 I0425 08:15:00.863524 140013575620416 metric_logger.py:196] completed step: 16, seconds: 49.470, TFLOP/s/device: 0.275, Tokens/s/device: 41.399, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872 I0425 08:15:01.010947 140013575620416 metric_logger.py:196] completed step: 17, seconds: 20.662, TFLOP/s/device: 0.658, Tokens/s/device: 99.118, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975 I0425 08:15:01.158237 140013575620416 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.405, Tokens/s/device: 13476.167, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333 I0425 08:15:01.304883 140013575620416 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0425 08:15:01.305622 140013575620416 checkpointing.py:776] Waited 0.0007557868957519531 seconds for step 19 to finish before starting checkpointing. I0425 08:15:01.307568 140013575620416 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 08:15:22.038301 139890817746688 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.580 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.033257246017456 s) (per-host) I0425 08:15:22.038427 139890817746688 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.000835s. I0425 08:15:29.449246 139890817746688 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.411636s. I0425 08:15:29.449481 139888244672256 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 08:15:29.449544 139888244672256 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 08:15:29.449596 139888244672256 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0425 08:15:29.451156 139888244672256 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0425 08:15:29.451326 140013575620416 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15. W0425 08:15:29.451451 140013575620416 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.143890 seconds. If this number is high, consider checkpointing less frequently. I0425 08:15:29.453054 140013575620416 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19 I0425 08:15:29.455247 140013575620416 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_true/checkpoints/19. I0425 08:15:30.121649 140013575620416 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 08:15:30.121741 140013575620416 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 08:15:30.148143 140013575620416 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027480s I0425 08:15:30.148318 140013575620416 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 49.764 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.030997514724731445 s) (per-host) I0425 08:15:30.148371 140013575620416 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031062s (batch_requests_ready=0.001640s, total_serialization_initiated=0.029353s, others=0.000069s) I0425 08:15:30.148469 140013575620416 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035131s (all_items=0.000012s, per_item={'items': '0.00001240'}, temp_paths=0.035118) I0425 08:15:30.149201 140013575620416 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.70 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints/19. I0425 08:15:30.149479 139888244672256 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 08:35:30.149452 I0425 08:15:30.151519 140013575620416 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0425 08:15:30.151807 139885068506880 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 08:15:30.151968 140013575620416 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777104901.3075383, 'wait_for_prev_duration_secs': 28.143890142440796, 'time_between_consecutive_saves_sec': 21.5892813205719, 'checkpointer_blocking_start_time': 1777104929.4531126, 'checkpointer_blocking_duration_secs': 0.6964943408966064, 'get_old_steps_start_time': 1777104930.149625, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1777104901.3058581, 'checkpoint_manager_blocking_duration_secs': 28.84607744216919} I0425 08:15:30.152107 140013575620416 checkpointing.py:408] Started an asynchronous checkpoint save for step 19 I0425 08:15:30.152168 140013575620416 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 08:15:35.204794 139890922600192 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_true/checkpoints/19/items/array_metadatas/process_6 I0425 08:16:11.464193 139888244672256 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.204 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.346834659576416 s) (per-host) I0425 08:16:11.464316 139888244672256 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.314742s. I0425 08:16:18.975395 139888244672256 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.825805s. I0425 08:16:18.975687 139885068506880 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 08:16:18.975805 139885068506880 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 08:16:18.975845 139885068506880 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0425 08:16:18.977508 139885068506880 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0425 08:16:18.977647 140013575620416 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0425 08:16:18.977737 140013575620416 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 08:16:18.978688 140013575620416 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.204, Tokens/s/device: 13898.031, total_weights: 65536, loss: 5.391, lm_loss: 5.391, perplexity: 219.349 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sat Apr 25 08:16:31 UTC 2026 EXIT_CODE=0
XPK Start: Sat Apr 25 13:08:33 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 13:08:58.258057: 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 13:08:58.468610 133170514949952 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 13:09:07.510421 133170514949952 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 13:09:07.512813 133170514949952 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-true-rs-p031g-slice-job-0-0.mt-15-ocdbt-true-rs-p031g:8482 I0425 13:09:08.113279 133170514949952 max_utils.py:284] Jax distributed system initialized! I0425 13:09:14.157874 133170514949952 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 13:09:14.157981 133170514949952 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 13:09:14.158022 133170514949952 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 13:09:14.158058 133170514949952 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 13:09:14.856062 133170514949952 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 13:09:14.856666 133170514949952 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 13:09:14.856847 133170514949952 checkpointing.py:688] Setting up checkpoint logger... I0425 13:09:14.856898 133170514949952 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 13:09:14.856942 133170514949952 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 13:09:14.857300 133170514949952 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x791d8c31c2c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 13:09:16.117714 133170514949952 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0425 13:09:16.117952 133170514949952 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 0x790918406a20>}, handler_registry=None I0425 13:09:16.118328 133170514949952 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x790918406a20>` 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 13:09:16.118477 133170514949952 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79091840ee10>` 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 13:09:16.118520 133170514949952 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 0x790918406a20>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x790918406a20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79091840ee10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79091840ee10>}). I0425 13:09:16.118876 133170514949952 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 13:09:16.118968 133170514949952 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7908347d1d00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 13:09:17.624232 133048406365952 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777122393448164617, 'commit_timestamp_nsecs': 1777122438604361521, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_true/checkpoints/9/_CHECKPOINT_METADATA I0425 13:09:17.868127 133048406365952 checkpoint_manager.py:1871] Missing metrics for step 9 I0425 13:09:17.897413 133170514949952 checkpoint_manager.py:1812] Found 3 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_true/checkpoints I0425 13:09:18.720916 133170514949952 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_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x790918406030> I0425 13:09:18.721113 133170514949952 checkpointing.py:302] Checkpoint manager created! I0425 13:09:19.654522 133170514949952 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 13:09:19.654633 133170514949952 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:09:20.041081 133170514949952 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 13:09:20.041183 133170514949952 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:09:20.057709 133170514949952 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 13:09:20.057771 133170514949952 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:09:20.081936 133170514949952 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 13:09:20.082005 133170514949952 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:09:20.098621 133170514949952 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 13:09:20.098683 133170514949952 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:09:20.115325 133170514949952 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 13:09:20.115386 133170514949952 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:09:20.140420 133170514949952 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 13:09:20.140491 133170514949952 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:09:20.161506 133170514949952 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 13:09:20.161578 133170514949952 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 13:09:20.381487 133170514949952 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0425 13:09:20.381599 133170514949952 checkpointing.py:582] restoring from this run's directory step 9 I0425 13:09:20.628852 133170514949952 event_tracking.py:70] [process=6] [sync] Started load 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_true/checkpoints/9. I0425 13:09:20.872676 133170514949952 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_true/checkpoints/9. I0425 13:09:20.873134 133170514949952 event_tracking.py:125] [process=6] [sync] Finished blocking load in 0.24 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_true/checkpoints/9. I0425 13:09:22.286661 1462 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 13:09:29.146790 133170514949952 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 230.169 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.865715980529785 s) (per-host) I0425 13:09:29.151513 133170514949952 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 217.048 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 7.291437387466431 s) (per-host) I0425 13:09:29.443798 133170514949952 event_tracking.py:138] [process=6] [sync] Finished load in 8.81 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_15_ocdbt_true/checkpoints/9 I0425 13:09:29.443954 133170514949952 standard_logger.py:34] {'step': 9, 'event_type': 'restore', '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_true/checkpoints', 'checkpointer_start_time': 1777122560.6287541, 'checkpointer_duration_secs': 8.815168380737305, 'checkpoint_manager_start_time': 1777122560.3818507, 'checkpoint_manager_duration_secs': 9.062073707580566} fsdp: 32 I0425 13:09:29.870011 133170514949952 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 13:09:29.870156 133170514949952 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 13:09:29.870213 133170514949952 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 13:09:29.870277 133170514949952 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 13:09:29.870330 133170514949952 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 13:09:29.870367 133170514949952 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 13:09:29.870419 133170514949952 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 13:09:29.870471 133170514949952 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 13:09:29.870508 133170514949952 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 13:09:29.870544 133170514949952 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 13:09:29.870589 133170514949952 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 13:09:29.870635 133170514949952 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 13:09:30.367799 133170514949952 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 13:09:30.367897 133170514949952 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0425 13:09:30.383558 133170514949952 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 13:09:30.383619 133170514949952 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0425 13:09:41.545609 133170514949952 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 13:09:41.546595 133170514949952 metric_logger.py:301] number parameters: 1.104 billion I0425 13:09:53.065339 133170514949952 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0425 13:09:53.218601 133170514949952 checkpointing.py:798] Waited 0.1532423496246338 seconds for step 10 to finish before starting checkpointing. I0425 13:09:53.221188 133170514949952 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 13:09:53.223259 133170514949952 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10 I0425 13:09:53.224864 133170514949952 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_true/checkpoints/10. I0425 13:09:53.548253 133170514949952 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 13:09:53.549262 133170514949952 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 13:09:53.549323 133170514949952 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 13:09:53.828047 133170514949952 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279868s I0425 13:09:53.828229 133170514949952 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.438 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2836735248565674 s) (per-host) I0425 13:09:53.828283 133170514949952 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.283739s (batch_requests_ready=0.002129s, total_serialization_initiated=0.281537s, others=0.000073s) I0425 13:09:53.828384 133170514949952 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288092s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.288075) I0425 13:09:53.829154 133170514949952 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.61 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_true/checkpoints/10. I0425 13:09:53.829516 133048372795136 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 13:29:53.829474 I0425 13:09:53.835901 133170514949952 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:09:53.836195 133045143873280 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:09:53.836350 133170514949952 standard_logger.py:34] {'step': 10, '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_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777122593.22117, 'wait_for_prev_duration_secs': 6.031990051269531e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777122593.223299, 'checkpointer_blocking_duration_secs': 0.6063716411590576, 'get_old_steps_start_time': 1777122593.8296943, 'get_old_steps_duration_secs': 3.4332275390625e-05, 'checkpoint_manager_blocking_start_time': 1777122593.2191396, 'checkpoint_manager_blocking_duration_secs': 0.617173433303833} I0425 13:09:53.836457 133170514949952 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0425 13:09:53.836508 133170514949952 max_utils.py:750] Memstats: After params initialized: I0425 13:09:53.836558 133170514949952 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0425 13:09:53.836596 133170514949952 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0425 13:09:53.836622 133170514949952 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0425 13:09:53.836646 133170514949952 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0425 13:09:54.153284 133170514949952 metric_logger.py:196] completed step: 10, seconds: 11.519, TFLOP/s/device: 1.180, Tokens/s/device: 177.799, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869 I0425 13:09:54.741090 133170514949952 metric_logger.py:196] completed step: 11, seconds: 1.086, TFLOP/s/device: 12.507, Tokens/s/device: 1885.177, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659 I0425 13:09:54.888436 133170514949952 metric_logger.py:196] completed step: 12, seconds: 0.442, TFLOP/s/device: 30.741, Tokens/s/device: 4633.673, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068 I0425 13:09:55.035784 133170514949952 metric_logger.py:196] completed step: 13, seconds: 0.152, TFLOP/s/device: 89.347, Tokens/s/device: 13467.394, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001 I0425 13:09:55.040460 133170514949952 checkpointing.py:794] Waiting for step 15 to finish before checkpoint... I0425 13:09:55.330340 133170514949952 checkpointing.py:798] Waited 0.28984808921813965 seconds for step 15 to finish before starting checkpointing. I0425 13:09:55.333395 133170514949952 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 13:09:58.790164 133047433271040 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_true/checkpoints/10/items/array_metadatas/process_6 I0425 13:10:35.685901 133048372795136 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.483 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.14130616188049 s) (per-host) I0425 13:10:35.686024 133048372795136 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.856392s. I0425 13:10:44.017386 133048372795136 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.187737s. I0425 13:10:44.017671 133045143873280 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:10:44.017793 133045143873280 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:10:44.017842 133045143873280 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:10:44.020024 133045143873280 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0425 13:10:44.020208 133170514949952 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0425 13:10:44.020345 133170514949952 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.686950 seconds. If this number is high, consider checkpointing less frequently. I0425 13:10:44.022529 133170514949952 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15 I0425 13:10:44.024963 133170514949952 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_true/checkpoints/15. I0425 13:10:44.332590 133170514949952 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 13:10:44.332684 133170514949952 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 13:10:44.358454 133170514949952 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027082s I0425 13:10:44.358604 133170514949952 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 50.684 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03043508529663086 s) (per-host) I0425 13:10:44.358655 133170514949952 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.030494s (batch_requests_ready=0.001730s, total_serialization_initiated=0.028699s, others=0.000066s) I0425 13:10:44.358742 133170514949952 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035559s (all_items=0.000014s, per_item={'items': '0.00001383'}, temp_paths=0.035545) I0425 13:10:44.359529 133170514949952 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_true/checkpoints/15. I0425 13:10:44.359798 133047886280448 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 13:30:44.359774 I0425 13:10:44.372831 133170514949952 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:10:44.373089 133045143873280 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:10:44.373288 133170514949952 standard_logger.py:34] {'step': 15, '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_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777122595.333366, 'wait_for_prev_duration_secs': 48.68694996833801, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777122644.0225728, 'checkpointer_blocking_duration_secs': 0.33731865882873535, 'get_old_steps_start_time': 1777122644.359912, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1777122595.3309052, 'checkpoint_manager_blocking_duration_secs': 49.04234218597412} I0425 13:10:44.373424 133170514949952 checkpointing.py:409] Started an asynchronous checkpoint save for step 15 I0425 13:10:44.374265 133170514949952 metric_logger.py:196] completed step: 14, seconds: 0.147, TFLOP/s/device: 92.230, Tokens/s/device: 13901.899, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193 I0425 13:10:44.380925 133170514949952 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.261, Tokens/s/device: 13906.525, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661 I0425 13:10:49.477465 133043574527744 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_true/checkpoints/15/items/array_metadatas/process_6 I0425 13:11:05.386211 133170514949952 metric_logger.py:196] completed step: 16, seconds: 49.339, TFLOP/s/device: 0.275, Tokens/s/device: 41.509, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872 I0425 13:11:05.533554 133170514949952 metric_logger.py:196] completed step: 17, seconds: 20.859, TFLOP/s/device: 0.651, Tokens/s/device: 98.181, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975 I0425 13:11:05.680781 133170514949952 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.163, Tokens/s/device: 13439.555, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333 I0425 13:11:05.827455 133170514949952 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0425 13:11:05.828171 133170514949952 checkpointing.py:798] Waited 0.0007336139678955078 seconds for step 19 to finish before starting checkpointing. I0425 13:11:05.830281 133170514949952 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 13:11:27.071733 133047886280448 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.955 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.7435200214386 s) (per-host) I0425 13:11:27.071856 133047886280448 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.711988s. I0425 13:11:34.989562 133047886280448 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.629678s. I0425 13:11:34.989834 133045143873280 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:11:34.989951 133045143873280 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:11:34.989998 133045143873280 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:11:34.991503 133045143873280 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0425 13:11:34.991638 133170514949952 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15. W0425 13:11:34.991721 133170514949952 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.161452 seconds. If this number is high, consider checkpointing less frequently. I0425 13:11:34.993370 133170514949952 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19 I0425 13:11:34.995442 133170514949952 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_true/checkpoints/19. I0425 13:11:35.282413 133170514949952 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 13:11:35.282501 133170514949952 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 13:11:35.308283 133170514949952 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026751s I0425 13:11:35.308458 133170514949952 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 51.120 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03017568588256836 s) (per-host) I0425 13:11:35.308508 133170514949952 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.030236s (batch_requests_ready=0.001632s, total_serialization_initiated=0.028538s, others=0.000066s) I0425 13:11:35.308602 133170514949952 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.034338s (all_items=0.000011s, per_item={'items': '0.00001144'}, temp_paths=0.034326) I0425 13:11:35.309255 133170514949952 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_true/checkpoints/19. I0425 13:11:35.309571 133043574527744 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 13:31:35.309531 I0425 13:11:35.312995 133170514949952 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:11:35.313257 133045143873280 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:11:35.313417 133170514949952 standard_logger.py:34] {'step': 19, '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_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777122665.8302505, 'wait_for_prev_duration_secs': 29.161452054977417, 'time_between_consecutive_saves_sec': 21.810187101364136, 'checkpointer_blocking_start_time': 1777122694.9934118, 'checkpointer_blocking_duration_secs': 0.316303014755249, 'get_old_steps_start_time': 1777122695.3097298, 'get_old_steps_duration_secs': 2.5033950805664062e-05, 'checkpoint_manager_blocking_start_time': 1777122665.8284605, 'checkpoint_manager_blocking_duration_secs': 29.4849271774292} I0425 13:11:35.313515 133170514949952 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0425 13:11:35.313555 133170514949952 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 13:11:41.016392 133042509178624 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_true/checkpoints/19/items/array_metadatas/process_6 I0425 13:12:17.643570 133043574527744 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.285 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.36525821685791 s) (per-host) I0425 13:12:17.643680 133043574527744 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.333994s. I0425 13:12:23.549142 133043574527744 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.239413s. I0425 13:12:23.549447 133045143873280 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:12:23.549587 133045143873280 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:12:23.549629 133045143873280 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:12:23.550911 133045143873280 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0425 13:12:23.551087 133170514949952 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0425 13:12:23.551260 133170514949952 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 13:12:23.552245 133170514949952 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.269, Tokens/s/device: 13907.752, total_weights: 65536, loss: 5.391, lm_loss: 5.391, perplexity: 219.349 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sat Apr 25 13:12:35 UTC 2026 EXIT_CODE=0