main (574ad3fb9) vs feat/nnx-post-train-fixes (574ad3fb9)| Metric | main 574ad3fb9 | feat/nnx-post-train-fixes 574ad3fb9 | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 5.8920 | 5.8920 | 0 |
| TFLOP/s | 58.325 | 57.459 | -0.866 |
| Tok/s | 8791.4 | 8660.8 | -130.569 |
| Avg s/step | 3.511 | 3.408 | -0.103 |
| 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 18 18:38:26 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-18 18:38:50.272318: 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) I0418 18:38:50.453994 133296776742720 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 18:38:59.495202 133296776742720 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 18:38:59.497480 133296776742720 distributed.py:172] Connecting to JAX distributed service on mt-09-pdb-lt-1-j8p7j-slice-job-0-0.mt-09-pdb-lt-1-j8p7j:8482 I0418 18:39:00.766492 133296776742720 max_utils.py:284] Jax distributed system initialized! I0418 18:39:06.995179 133296776742720 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 18:39:06.995287 133296776742720 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 18:39:06.995349 133296776742720 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 I0418 18:39:06.995398 133296776742720 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 18:39:07.689579 133296776742720 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 8, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0418 18:39:07.689859 133296776742720 checkpointing.py:677] Setting up checkpoint logger... I0418 18:39:07.689911 133296776742720 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 18:39:07.689954 133296776742720 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 18:39:07.690274 133296776742720 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 0x793af71140e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 18:39:11.441272 133296776742720 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0418 18:39:11.441528 133296776742720 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7926d01968a0>}, handler_registry=None I0418 18:39:11.441766 133296776742720 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7926d01968a0>` 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`. I0418 18:39:11.441815 133296776742720 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7926d0196120>` 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`. I0418 18:39:11.441850 133296776742720 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 0x7926d01968a0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7926d01968a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7926d0196120>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7926d0196120>}). I0418 18:39:11.442166 133296776742720 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 18:39:11.442238 133296776742720 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7926b818be20> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 18:39:13.516368 133296776742720 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints I0418 18:39:13.932364 133296776742720 checkpoint_manager.py:929] [process=5][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=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7926d0393650> I0418 18:39:13.932564 133296776742720 checkpointing.py:301] Checkpoint manager created! I0418 18:39:14.849104 133296776742720 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 18:39:14.849216 133296776742720 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0418 18:39:15.224098 133296776742720 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 18:39:15.224185 133296776742720 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0418 18:39:15.240371 133296776742720 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 18:39:15.240431 133296776742720 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0418 18:39:15.264005 133296776742720 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:39:15.264069 133296776742720 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0418 18:39:15.280211 133296776742720 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:39:15.280270 133296776742720 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0418 18:39:15.296475 133296776742720 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:39:15.296538 133296776742720 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0418 18:39:15.321162 133296776742720 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 18:39:15.321228 133296776742720 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0418 18:39:15.342195 133296776742720 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 18:39:15.342258 133296776742720 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, 'tensor'). I0418 18:39:15.557906 133296776742720 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0418 18:39:15.558013 133296776742720 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 8 tensor: 4 I0418 18:39:16.997103 133296776742720 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: ('tensor',) I0418 18:39:16.997227 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, 'tensor') I0418 18:39:16.997279 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, 'tensor') I0418 18:39:16.997337 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0418 18:39:16.997388 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0418 18:39:16.997428 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0418 18:39:16.997493 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, 'tensor', None) I0418 18:39:16.997549 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: ('tensor', None, None, 'fsdp') I0418 18:39:16.997589 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, 'tensor', None) I0418 18:39:16.997626 133296776742720 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, 'tensor', None) I0418 18:39:16.997673 133296776742720 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0418 18:39:16.997725 133296776742720 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0418 18:39:17.498576 133296776742720 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0418 18:39:17.498669 133296776742720 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0418 18:39:17.514003 133296776742720 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0418 18:39:17.514061 133296776742720 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0418 18:39:32.748797 133296776742720 max_utils.py:791] Total memory size: 0.9 GB, Output size: 0.4 GB, Temp size: 0.5 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0418 18:39:32.749679 133296776742720 metric_logger.py:301] number parameters: 1.104 billion I0418 18:39:33.938663 133296776742720 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0418 18:39:48.310223 133296776742720 checkpointing.py:776] Waited 14.371539115905762 seconds for step 0 to finish before starting checkpointing. I0418 18:39:48.312875 133296776742720 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 18:39:48.314876 133296776742720 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0418 18:39:48.316382 133296776742720 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints/0. I0418 18:39:49.490287 133296776742720 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 18:39:49.491233 133296776742720 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 18:39:49.491290 133296776742720 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 I0418 18:39:49.754537 133296776742720 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.264323s I0418 18:39:49.754716 133296776742720 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.715 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2699153423309326 s) (per-host) I0418 18:39:49.754768 133296776742720 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.269979s (batch_requests_ready=0.002328s, total_serialization_initiated=0.267578s, others=0.000073s) I0418 18:39:49.754859 133296776742720 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.274095s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.274078) I0418 18:39:49.755639 133296776742720 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.44 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints/0. I0418 18:39:49.755967 133168650172160 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 18:59:49.755929 I0418 18:39:49.765867 133296776742720 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:39:49.766180 133168136972032 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:39:49.766334 133296776742720 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776537588.3128452, 'wait_for_prev_duration_secs': 0.00011754035949707031, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776537588.3149154, 'checkpointer_blocking_duration_secs': 1.4412047863006592, 'get_old_steps_start_time': 1776537589.7561464, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776537588.3108091, 'checkpoint_manager_blocking_duration_secs': 1.4554872512817383} I0418 18:39:49.766437 133296776742720 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0418 18:39:49.766505 133296776742720 max_utils.py:750] Memstats: After params initialized: I0418 18:39:49.766558 133296776742720 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0)) I0418 18:39:49.766589 133296776742720 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0418 18:39:49.766616 133296776742720 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0418 18:39:49.766641 133296776742720 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0418 18:39:50.084203 133296776742720 metric_logger.py:196] completed step: 0, seconds: 1.189, TFLOP/s/device: 2.857, Tokens/s/device: 430.664, total_weights: 16384, loss: 10.862, lm_loss: 10.862, perplexity: 52153.723 I0418 18:39:50.164539 133296776742720 metric_logger.py:196] completed step: 1, seconds: 16.144, TFLOP/s/device: 0.210, Tokens/s/device: 31.714, total_weights: 16384, loss: 10.862, lm_loss: 10.862, perplexity: 52153.723 I0418 18:39:50.558382 133296776742720 metric_logger.py:196] completed step: 2, seconds: 0.022, TFLOP/s/device: 155.062, Tokens/s/device: 23372.592, total_weights: 16384, loss: 9.763, lm_loss: 9.763, perplexity: 17373.129 I0418 18:39:50.616731 133296776742720 metric_logger.py:196] completed step: 3, seconds: 0.394, TFLOP/s/device: 8.618, Tokens/s/device: 1299.021, total_weights: 16384, loss: 8.823, lm_loss: 8.823, perplexity: 6791.817 I0418 18:39:50.733727 133296776742720 metric_logger.py:196] completed step: 4, seconds: 0.064, TFLOP/s/device: 53.239, Tokens/s/device: 8024.827, total_weights: 16384, loss: 7.947, lm_loss: 7.947, perplexity: 2828.417 I0418 18:39:50.739140 133296776742720 metric_logger.py:196] completed step: 5, seconds: 0.058, TFLOP/s/device: 58.430, Tokens/s/device: 8807.238, total_weights: 16384, loss: 7.212, lm_loss: 7.212, perplexity: 1355.339 I0418 18:39:54.127589 2639 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 18:39:56.490655 133168145364736 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_5 I0418 18:40:05.479774 133296776742720 metric_logger.py:196] completed step: 6, seconds: 0.117, TFLOP/s/device: 28.910, Tokens/s/device: 4357.595, total_weights: 16384, loss: 6.652, lm_loss: 6.652, perplexity: 774.659 I0418 18:40:05.538325 133296776742720 metric_logger.py:196] completed step: 7, seconds: 14.683, TFLOP/s/device: 0.231, Tokens/s/device: 34.871, total_weights: 16384, loss: 6.269, lm_loss: 6.269, perplexity: 527.696 I0418 18:40:05.597030 133296776742720 metric_logger.py:196] completed step: 8, seconds: 0.063, TFLOP/s/device: 53.771, Tokens/s/device: 8104.985, total_weights: 16384, loss: 6.031, lm_loss: 6.031, perplexity: 416.165 I0418 18:40:05.654812 133296776742720 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0418 18:40:05.655410 133296776742720 checkpointing.py:776] Waited 0.0006153583526611328 seconds for step 9 to finish before starting checkpointing. I0418 18:40:05.657664 133296776742720 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:40:27.347978 133168650172160 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 41.719 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.86314153671265 s) (per-host) I0418 18:40:27.348108 133168650172160 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.592027s. I0418 18:40:35.059599 133168650172160 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 45.303503s. I0418 18:40:35.059864 133168136972032 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:40:35.059994 133168136972032 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:40:35.060049 133168136972032 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:40:35.061401 133168136972032 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 18:40:35.061600 133296776742720 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0418 18:40:35.061740 133296776742720 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.404074 seconds. If this number is high, consider checkpointing less frequently. I0418 18:40:35.063609 133296776742720 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0418 18:40:35.065620 133296776742720 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints/9. I0418 18:40:36.217618 133296776742720 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 18:40:36.217706 133296776742720 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 I0418 18:40:36.243551 133296776742720 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026923s I0418 18:40:36.243697 133296776742720 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 50.685 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.030434370040893555 s) (per-host) I0418 18:40:36.243749 133296776742720 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.030495s (batch_requests_ready=0.001891s, total_serialization_initiated=0.028535s, others=0.000069s) I0418 18:40:36.243839 133296776742720 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.034483s (all_items=0.000013s, per_item={'items': '0.00001287'}, temp_paths=0.034471) I0418 18:40:36.244506 133296776742720 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.18 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints/9. I0418 18:40:36.244837 133168136972032 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 19:00:36.244797 I0418 18:40:36.248429 133296776742720 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:40:36.248728 133167584974592 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:40:36.248891 133296776742720 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776537605.6576347, 'wait_for_prev_duration_secs': 29.4040744304657, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776537635.063648, 'checkpointer_blocking_duration_secs': 1.18135404586792, 'get_old_steps_start_time': 1776537636.2450247, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776537605.6556537, 'checkpoint_manager_blocking_duration_secs': 30.593202352523804} I0418 18:40:36.249004 133296776742720 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0418 18:40:36.249044 133296776742720 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:40:43.113481 133168631879424 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_5 I0418 18:41:17.216261 133168136972032 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 38.524 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.00296330451965 s) (per-host) I0418 18:41:17.216373 133168136972032 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.971422s. I0418 18:41:26.597702 133168136972032 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.352736s. I0418 18:41:26.597987 133167584974592 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:41:26.598106 133167584974592 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:41:26.598153 133167584974592 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:41:26.599804 133167584974592 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 18:41:26.599957 133296776742720 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0418 18:41:26.600106 133296776742720 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 18:41:26.601119 133296776742720 metric_logger.py:196] completed step: 9, seconds: 0.058, TFLOP/s/device: 58.325, Tokens/s/device: 8791.360, total_weights: 16384, loss: 5.892, lm_loss: 5.892, perplexity: 361.955 Per train step: Total TFLOPs: 3.40 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sat Apr 18 18:41:37 UTC 2026 EXIT_CODE=0
XPK Start: Sun Apr 19 00:18:55 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-19 00:19:19.769349: 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) I0419 00:19:19.946853 139870164690752 max_utils.py:273] Attempting to initialize the jax distributed system... I0419 00:19:28.988790 139870164690752 distributed.py:149] Starting JAX distributed service on [::]:8482 I0419 00:19:28.991202 139870164690752 distributed.py:172] Connecting to JAX distributed service on mt-09-pdb-lt-1-bngaa-slice-job-0-0.mt-09-pdb-lt-1-bngaa:8482 I0419 00:19:31.194670 139870164690752 max_utils.py:284] Jax distributed system initialized! I0419 00:19:37.429256 139870164690752 max_utils.py:800] System Information: Jax Version: 0.9.2 I0419 00:19:37.429354 139870164690752 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0419 00:19:37.429396 139870164690752 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 I0419 00:19:37.429431 139870164690752 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0419 00:19:38.473726 139870164690752 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 8, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0419 00:19:38.474320 139870164690752 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 8, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0419 00:19:38.474499 139870164690752 checkpointing.py:688] Setting up checkpoint logger... I0419 00:19:38.474548 139870164690752 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0419 00:19:38.474591 139870164690752 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0419 00:19:38.474939 139870164690752 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 0x7f3571110350>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0419 00:19:41.764345 139870164690752 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0419 00:19:41.764681 139870164690752 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f344b7f7da0>}, handler_registry=None I0419 00:19:41.764930 139870164690752 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f344b7f7da0>` 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`. I0419 00:19:41.764978 139870164690752 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f344b7f6c60>` 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`. I0419 00:19:41.765014 139870164690752 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 0x7f344b7f7da0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f344b7f7da0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f344b7f6c60>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f344b7f6c60>}). I0419 00:19:41.765397 139870164690752 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0419 00:19:41.765467 139870164690752 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f2110286c00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0419 00:19:43.169146 139870164690752 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints I0419 00:19:44.042864 139870164690752 checkpoint_manager.py:929] [process=0][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f344b7f5e20> I0419 00:19:44.043031 139870164690752 checkpointing.py:302] Checkpoint manager created! I0419 00:19:44.954543 139870164690752 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0419 00:19:44.954644 139870164690752 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0419 00:19:45.334937 139870164690752 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 00:19:45.335027 139870164690752 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0419 00:19:45.351174 139870164690752 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 00:19:45.351234 139870164690752 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0419 00:19:45.374916 139870164690752 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 00:19:45.374979 139870164690752 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0419 00:19:45.391203 139870164690752 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 00:19:45.391265 139870164690752 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0419 00:19:45.407403 139870164690752 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 00:19:45.407464 139870164690752 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0419 00:19:45.432678 139870164690752 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0419 00:19:45.432749 139870164690752 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0419 00:19:45.453413 139870164690752 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0419 00:19:45.453473 139870164690752 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, 'tensor'). I0419 00:19:45.667032 139870164690752 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0419 00:19:45.667141 139870164690752 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 8 tensor: 4 I0419 00:19:47.101045 139870164690752 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: ('tensor',) I0419 00:19:47.101169 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, 'tensor') I0419 00:19:47.101222 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, 'tensor') I0419 00:19:47.101278 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0419 00:19:47.101336 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0419 00:19:47.101376 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0419 00:19:47.101435 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, 'tensor', None) I0419 00:19:47.101491 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: ('tensor', None, None, 'fsdp') I0419 00:19:47.101532 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, 'tensor', None) I0419 00:19:47.101570 139870164690752 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, 'tensor', None) I0419 00:19:47.101618 139870164690752 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0419 00:19:47.101665 139870164690752 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0419 00:19:47.591997 139870164690752 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0419 00:19:47.592092 139870164690752 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0419 00:19:47.607369 139870164690752 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0419 00:19:47.607430 139870164690752 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0419 00:20:02.733687 139870164690752 max_utils.py:791] Total memory size: 0.9 GB, Output size: 0.4 GB, Temp size: 0.5 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0419 00:20:02.741709 139870164690752 metric_logger.py:301] number parameters: 1.104 billion I0419 00:20:04.271446 139870164690752 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0419 00:20:18.420612 139870164690752 checkpointing.py:798] Waited 14.14914321899414 seconds for step 0 to finish before starting checkpointing. I0419 00:20:18.423332 139870164690752 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 00:20:18.425452 139870164690752 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0 I0419 00:20:18.426996 139870164690752 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0. I0419 00:20:19.127028 139870164690752 signaling_client.py:364] Using JaxDistributedSignalingClient I0419 00:20:19.142987 139870164690752 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 00:20:19.143057 139870164690752 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 I0419 00:20:19.397565 139742408759040 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0 I0419 00:20:19.400232 139870164690752 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.258073s I0419 00:20:19.401052 139870164690752 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.833 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.26446032524108887 s) (per-host) I0419 00:20:19.401160 139870164690752 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.264588s (batch_requests_ready=0.002219s, total_serialization_initiated=0.261599s, others=0.000770s) I0419 00:20:19.401243 139870164690752 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.268863s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.268845) I0419 00:20:19.401983 139870164690752 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.98 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0. I0419 00:20:19.402340 139741364877056 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:40:19.402297 I0419 00:20:19.417606 139870164690752 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0419 00:20:19.417875 139742417151744 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0419 00:20:19.418016 139870164690752 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776558018.423315, 'wait_for_prev_duration_secs': 5.8650970458984375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776558018.4254906, 'checkpointer_blocking_duration_secs': 0.977003812789917, 'get_old_steps_start_time': 1776558019.4025195, 'get_old_steps_duration_secs': 2.86102294921875e-05, 'checkpoint_manager_blocking_start_time': 1776558018.4211788, 'checkpoint_manager_blocking_duration_secs': 0.9967994689941406} I0419 00:20:19.418118 139870164690752 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0419 00:20:19.418167 139870164690752 max_utils.py:750] Memstats: After params initialized: I0419 00:20:19.418214 139870164690752 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0)) I0419 00:20:19.418247 139870164690752 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0)) I0419 00:20:19.418274 139870164690752 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0)) I0419 00:20:19.418298 139870164690752 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0)) I0419 00:20:19.771278 139870164690752 metric_logger.py:196] completed step: 0, seconds: 1.182, TFLOP/s/device: 2.873, Tokens/s/device: 433.056, total_weights: 16384, loss: 10.862, lm_loss: 10.862, perplexity: 52153.723 I0419 00:20:19.774299 139870164690752 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/tensorboard/' I0419 00:20:20.154483 139870164690752 metric_logger.py:196] completed step: 1, seconds: 15.498, TFLOP/s/device: 0.219, Tokens/s/device: 33.036, total_weights: 16384, loss: 10.862, lm_loss: 10.862, perplexity: 52153.723 I0419 00:20:20.211452 139870164690752 metric_logger.py:196] completed step: 2, seconds: 0.384, TFLOP/s/device: 8.848, Tokens/s/device: 1333.708, total_weights: 16384, loss: 9.763, lm_loss: 9.763, perplexity: 17373.129 I0419 00:20:20.269959 139870164690752 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 384.208, Tokens/s/device: 57912.001, total_weights: 16384, loss: 8.823, lm_loss: 8.823, perplexity: 6791.817 I0419 00:20:21.419551 139741911570176 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0/items I0419 00:20:21.571717 139740853491456 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776558021159438957, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776558021159438957, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0419 00:20:23.139748 2604 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0419 00:20:24.890309 139741381662464 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_0 I0419 00:20:34.807752 139870164690752 metric_logger.py:196] completed step: 4, seconds: 0.058, TFLOP/s/device: 58.945, Tokens/s/device: 8884.878, total_weights: 16384, loss: 7.947, lm_loss: 7.947, perplexity: 2828.417 I0419 00:20:34.816510 139870164690752 metric_logger.py:196] completed step: 5, seconds: 0.059, TFLOP/s/device: 57.990, Tokens/s/device: 8740.930, total_weights: 16384, loss: 7.212, lm_loss: 7.212, perplexity: 1355.339 I0419 00:20:34.873698 139870164690752 metric_logger.py:196] completed step: 6, seconds: 14.537, TFLOP/s/device: 0.234, Tokens/s/device: 35.220, total_weights: 16384, loss: 6.652, lm_loss: 6.652, perplexity: 774.659 I0419 00:20:34.932271 139870164690752 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 510.411, Tokens/s/device: 76934.636, total_weights: 16384, loss: 6.269, lm_loss: 6.269, perplexity: 527.696 I0419 00:20:34.990733 139870164690752 metric_logger.py:196] completed step: 8, seconds: 0.058, TFLOP/s/device: 58.745, Tokens/s/device: 8854.761, total_weights: 16384, loss: 6.031, lm_loss: 6.031, perplexity: 416.165 I0419 00:20:35.048341 139870164690752 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0419 00:20:35.049016 139870164690752 checkpointing.py:798] Waited 0.0006883144378662109 seconds for step 9 to finish before starting checkpointing. I0419 00:20:35.051012 139870164690752 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 00:20:59.376647 139741373269760 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 36.243348s (commit=35.099472s, array_metadata_write=1.143876s) I0419 00:20:59.377653 139741364877056 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 39.253 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.241034269332886 s) (per-host) I0419 00:20:59.377779 139741364877056 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.975322s. I0419 00:21:00.038064 139741364877056 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776558021159438957, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0419 00:21:00.713512 139741364877056 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000203s. I0419 00:21:01.193129 139740853491456 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776558021159438957, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0/_CHECKPOINT_METADATA I0419 00:21:01.400790 139741364877056 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0419 00:21:02.343543 139741364877056 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.188345s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0/items I0419 00:21:02.345259 139741364877056 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0/items I0419 00:21:02.863425 139741364877056 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0 I0419 00:21:04.646479 139741364877056 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0`. I0419 00:21:04.647361 139741364877056 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 46.22 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/0 I0419 00:21:04.648983 139741364877056 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 45.246523s. I0419 00:21:04.649178 139742417151744 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 00:21:04.649287 139742417151744 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0419 00:21:04.649348 139742417151744 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0419 00:21:04.652003 139742417151744 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0419 00:21:04.652173 139870164690752 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0419 00:21:04.652262 139870164690752 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.601253 seconds. If this number is high, consider checkpointing less frequently. I0419 00:21:04.654103 139870164690752 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9 I0419 00:21:04.656233 139870164690752 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9. I0419 00:21:05.358985 139870164690752 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 00:21:05.359077 139870164690752 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 I0419 00:21:05.385407 139870164690752 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027161s I0419 00:21:05.386128 139870164690752 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 48.766 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.031632423400878906 s) (per-host) I0419 00:21:05.386200 139870164690752 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031725s (batch_requests_ready=0.001923s, total_serialization_initiated=0.029143s, others=0.000659s) I0419 00:21:05.386271 139870164690752 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035674s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.035657) I0419 00:21:05.386961 139870164690752 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.73 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9. I0419 00:21:05.387315 139740811527936 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:41:05.387274 I0419 00:21:05.390013 139870164690752 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0419 00:21:05.390257 139742417151744 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0419 00:21:05.390388 139870164690752 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776558035.0509794, 'wait_for_prev_duration_secs': 29.601253271102905, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776558064.654146, 'checkpointer_blocking_duration_secs': 0.7333106994628906, 'get_old_steps_start_time': 1776558065.387482, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776558035.0492337, 'checkpoint_manager_blocking_duration_secs': 30.341114044189453} I0419 00:21:05.390489 139870164690752 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0419 00:21:05.390531 139870164690752 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 00:21:05.603410 139741364877056 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9 I0419 00:21:07.340346 139741911570176 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9/items I0419 00:21:10.174056 139740828313344 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_0 I0419 00:21:50.032492 139740819920640 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.600429s (commit=40.555518s, array_metadata_write=1.044911s) I0419 00:21:50.033505 139740811527936 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 35.354 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.67897152900696 s) (per-host) I0419 00:21:50.033655 139740811527936 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.646225s. I0419 00:21:51.683750 139740811527936 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0419 00:21:51.837111 139740811527936 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000232s. I0419 00:21:52.762752 139740811527936 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.914867s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9/items I0419 00:21:52.764408 139740811527936 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9/items I0419 00:21:53.291602 139740811527936 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9 I0419 00:21:54.652629 139740811527936 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9`. I0419 00:21:54.653470 139740811527936 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 50.00 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_09_pdb_lt_1/checkpoints/9 I0419 00:21:54.655009 139740811527936 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 49.267576s. I0419 00:21:54.655218 139742417151744 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 00:21:54.655332 139742417151744 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0419 00:21:54.655391 139742417151744 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0419 00:21:54.657634 139742417151744 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0419 00:21:54.657832 139870164690752 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0419 00:21:54.658010 139870164690752 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 00:21:54.659144 139870164690752 metric_logger.py:196] completed step: 9, seconds: 0.059, TFLOP/s/device: 57.459, Tokens/s/device: 8660.791, total_weights: 16384, loss: 5.892, lm_loss: 5.892, perplexity: 361.955 Per train step: Total TFLOPs: 3.40 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sun Apr 19 00:22:06 UTC 2026 EXIT_CODE=0