main (59e0f1759) vs feat/nnx-trainstate-and-training-loop (77202a8e9)| Metric | main 59e0f1759 | feat/nnx-trainstate-and-training-loop 77202a8e9 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 6.9810 | 6.9810 | 0 |
| TFLOP/s | 11.354 | 11.369 | +0.015 |
| Tok/s | 1814.7 | 1817.1 | +2.396 |
| Avg s/step | 1.272 | 1.344 | +0.072 |
| Memory % | 1.31 | 1.31 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Sat Apr 25 07:20:19 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 07:20:44.686470: 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 07:20:44.895111 140390170629952 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 07:20:53.937241 140390170629952 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 07:20:53.939524 140390170629952 distributed.py:172] Connecting to JAX distributed service on mt-03-dropout-963zw-slice-job-0-0.mt-03-dropout-963zw:8482 I0425 07:20:55.914268 140390170629952 max_utils.py:284] Jax distributed system initialized! I0425 07:21:02.257855 140390170629952 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 07:21:02.257958 140390170629952 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 07:21:02.257998 140390170629952 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 07:21:02.258034 140390170629952 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 07:21:02.952807 140390170629952 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1) I0425 07:21:02.953090 140390170629952 checkpointing.py:677] Setting up checkpoint logger... I0425 07:21:02.953162 140390170629952 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 07:21:02.953207 140390170629952 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 07:21:02.953548 140390170629952 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 0x7fae9af1fd40>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 07:21:05.794047 140390170629952 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0425 07:21:05.794309 140390170629952 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 0x7f9a54246510>}, handler_registry=None I0425 07:21:05.794550 140390170629952 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f9a54246510>` 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 07:21:05.794599 140390170629952 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f9a7c242cc0>` 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 07:21:05.794635 140390170629952 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 0x7f9a54246510>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f9a54246510>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f9a7c242cc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f9a7c242cc0>}). I0425 07:21:05.794959 140390170629952 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 07:21:05.795027 140390170629952 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f9a04769b20> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 07:21:06.519822 140390170629952 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_03_dropout/checkpoints I0425 07:21:06.974756 140390170629952 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=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_20260425_071456/linen_xpk_main_20260425_071456_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7fad59632600> I0425 07:21:06.974930 140390170629952 checkpointing.py:301] Checkpoint manager created! I0425 07:21:07.324936 140390170629952 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 07:21:07.325055 140390170629952 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0425 07:21:07.706725 140390170629952 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,128,2048]....................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 07:21:07.706823 140390170629952 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0425 07:21:07.723530 140390170629952 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,128,2048]....................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 07:21:07.723590 140390170629952 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0425 07:21:07.747732 140390170629952 attentions.py:1154] attentions/query Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 07:21:07.747805 140390170629952 attentions.py:1154] attentions/query Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0425 07:21:07.764520 140390170629952 attentions.py:1155] attentions/key Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 07:21:07.764585 140390170629952 attentions.py:1155] attentions/key Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0425 07:21:07.781364 140390170629952 attentions.py:1156] attentions/value Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 07:21:07.781425 140390170629952 attentions.py:1156] attentions/value Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0425 07:21:07.806515 140390170629952 attentions.py:1198] attentions/out Logical: bfloat16[32,128,16,128]..................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 07:21:07.806584 140390170629952 attentions.py:1198] attentions/out Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0425 07:21:07.833819 140390170629952 linears.py:525] linears/x Logical: bfloat16[32,128,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 07:21:07.833888 140390170629952 linears.py:525] linears/x Physical: bfloat16[32,128,7168]....................................... ('fsdp', None, None). I0425 07:21:08.055362 140390170629952 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0425 07:21:08.055476 140390170629952 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 07:21:09.525898 140390170629952 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 07:21:09.526029 140390170629952 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 07:21:09.526083 140390170629952 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 07:21:09.526162 140390170629952 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 07:21:09.526221 140390170629952 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 07:21:09.526262 140390170629952 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 07:21:09.526314 140390170629952 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 07:21:09.526366 140390170629952 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 07:21:09.526407 140390170629952 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 07:21:09.526444 140390170629952 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 07:21:09.526492 140390170629952 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 07:21:09.526541 140390170629952 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 07:21:10.050785 140390170629952 train.py:155] train/xent Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0425 07:21:10.050883 140390170629952 train.py:155] train/xent Physical: float32[32,128]............................................. ('fsdp', None). I0425 07:21:10.066601 140390170629952 train.py:162] train/z_loss Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0425 07:21:10.066659 140390170629952 train.py:162] train/z_loss Physical: float32[32,128]............................................. ('fsdp', None). I0425 07:21:13.682776 140390170629952 max_utils.py:791] Total memory size: 0.8 GB, Output size: 0.4 GB, Temp size: 0.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0425 07:21:13.683552 140390170629952 metric_logger.py:301] number parameters: 1.104 billion I0425 07:21:18.243298 140390170629952 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0425 07:21:18.318222 140390170629952 checkpointing.py:776] Waited 0.0749046802520752 seconds for step 0 to finish before starting checkpointing. I0425 07:21:18.320603 140390170629952 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 07:21:18.322906 140390170629952 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0425 07:21:18.324234 140390170629952 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_03_dropout/checkpoints/0. I0425 07:21:18.623349 140390170629952 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 07:21:18.624282 140390170629952 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 07:21:18.624342 140390170629952 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 07:21:18.902518 140390170629952 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279243s I0425 07:21:18.902692 140390170629952 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.419 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2846674919128418 s) (per-host) I0425 07:21:18.902746 140390170629952 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.284731s (batch_requests_ready=0.002229s, total_serialization_initiated=0.282430s, others=0.000071s) I0425 07:21:18.902840 140390170629952 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288963s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.288945) I0425 07:21:18.903664 140390170629952 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.58 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_03_dropout/checkpoints/0. I0425 07:21:18.903962 140261683451648 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 07:41:18.903931 I0425 07:21:19.352221 140390170629952 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 07:21:19.352600 140261111879424 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 07:21:19.352764 140390170629952 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777101678.3205843, 'wait_for_prev_duration_secs': 6.270408630371094e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777101678.322947, 'checkpointer_blocking_duration_secs': 0.5811717510223389, 'get_old_steps_start_time': 1777101678.9041443, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1777101678.3187644, 'checkpoint_manager_blocking_duration_secs': 1.0339574813842773} I0425 07:21:19.352874 140390170629952 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0425 07:21:19.352930 140390170629952 max_utils.py:750] Memstats: After params initialized: I0425 07:21:19.352989 140390170629952 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_24(process=6,(0,6,0,0)) I0425 07:21:19.353022 140390170629952 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_25(process=6,(1,6,0,0)) I0425 07:21:19.353049 140390170629952 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_28(process=6,(0,7,0,0)) I0425 07:21:19.353075 140390170629952 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_29(process=6,(1,7,0,0)) I0425 07:21:19.672222 140390170629952 metric_logger.py:196] completed step: 0, seconds: 4.560, TFLOP/s/device: 0.176, Tokens/s/device: 28.072, total_weights: 4096, loss: 10.889, lm_loss: 10.889, perplexity: 53560.699 I0425 07:21:19.770174 140390170629952 metric_logger.py:196] completed step: 1, seconds: 1.427, TFLOP/s/device: 0.561, Tokens/s/device: 89.674, total_weights: 4096, loss: 10.902, lm_loss: 10.902, perplexity: 54298.523 I0425 07:21:20.167889 140390170629952 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 28.893, Tokens/s/device: 4617.771, total_weights: 4096, loss: 9.956, lm_loss: 9.956, perplexity: 21082.051 I0425 07:21:20.238533 140390170629952 metric_logger.py:196] completed step: 3, seconds: 0.399, TFLOP/s/device: 2.007, Tokens/s/device: 320.841, total_weights: 4096, loss: 9.139, lm_loss: 9.139, perplexity: 9314.691 I0425 07:21:20.380358 140390170629952 metric_logger.py:196] completed step: 4, seconds: 0.075, TFLOP/s/device: 10.681, Tokens/s/device: 1707.054, total_weights: 4096, loss: 8.459, lm_loss: 8.459, perplexity: 4715.493 I0425 07:21:20.386507 140390170629952 metric_logger.py:196] completed step: 5, seconds: 0.071, TFLOP/s/device: 11.235, Tokens/s/device: 1795.584, total_weights: 4096, loss: 7.924, lm_loss: 7.924, perplexity: 2762.373 I0425 07:21:21.983175 2788 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 07:21:23.993351 140261120272128 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_03_dropout/checkpoints/0/items/array_metadatas/process_6 I0425 07:21:29.613523 140390170629952 metric_logger.py:196] completed step: 6, seconds: 0.142, TFLOP/s/device: 5.644, Tokens/s/device: 902.056, total_weights: 4096, loss: 7.553, lm_loss: 7.553, perplexity: 1906.617 I0425 07:21:29.684189 140390170629952 metric_logger.py:196] completed step: 7, seconds: 9.158, TFLOP/s/device: 0.087, Tokens/s/device: 13.977, total_weights: 4096, loss: 7.267, lm_loss: 7.267, perplexity: 1431.653 I0425 07:21:29.754827 140390170629952 metric_logger.py:196] completed step: 8, seconds: 0.075, TFLOP/s/device: 10.722, Tokens/s/device: 1713.635, total_weights: 4096, loss: 7.111, lm_loss: 7.111, perplexity: 1225.983 I0425 07:21:29.824877 140390170629952 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0425 07:21:29.825550 140390170629952 checkpointing.py:776] Waited 0.0006933212280273438 seconds for step 9 to finish before starting checkpointing. I0425 07:21:29.827609 140390170629952 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 07:21:51.247332 140261683451648 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 48.410 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 32.629258155822754 s) (per-host) I0425 07:21:51.247457 140261683451648 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.343384s. I0425 07:21:58.931498 140261683451648 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 40.027407s. I0425 07:21:58.931768 140261111879424 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 07:21:58.931897 140261111879424 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 07:21:58.931949 140261111879424 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 07:21:58.933577 140261111879424 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 07:21:58.933748 140390170629952 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0425 07:21:58.933881 140390170629952 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.106269 seconds. If this number is high, consider checkpointing less frequently. I0425 07:21:58.935819 140390170629952 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0425 07:21:58.937827 140390170629952 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_03_dropout/checkpoints/9. I0425 07:21:59.233667 140390170629952 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 07:21:59.233764 140390170629952 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 07:21:59.270801 140390170629952 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038932s I0425 07:21:59.270971 140390170629952 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 36.394 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04238486289978027 s) (per-host) I0425 07:21:59.271023 140390170629952 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.042446s (batch_requests_ready=0.001807s, total_serialization_initiated=0.040570s, others=0.000069s) I0425 07:21:59.271134 140390170629952 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.046740s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.046725) I0425 07:21:59.271816 140390170629952 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_03_dropout/checkpoints/9. I0425 07:21:59.272157 140261111879424 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 07:41:59.272122 I0425 07:21:59.274038 140390170629952 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0425 07:21:59.274321 140255202092800 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 07:21:59.274480 140390170629952 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777101689.8275807, 'wait_for_prev_duration_secs': 29.106268882751465, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777101718.9358585, 'checkpointer_blocking_duration_secs': 0.33645081520080566, 'get_old_steps_start_time': 1777101719.2723317, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1777101689.8257768, 'checkpoint_manager_blocking_duration_secs': 29.44866943359375} I0425 07:21:59.274594 140390170629952 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0425 07:21:59.274642 140390170629952 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 07:22:04.734699 140237703386880 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_03_dropout/checkpoints/9/items/array_metadatas/process_6 I0425 07:22:39.949577 140261111879424 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.791 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.720954179763794 s) (per-host) I0425 07:22:39.949693 140261111879424 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.677418s. I0425 07:22:48.287503 140261111879424 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.015207s. I0425 07:22:48.287778 140255202092800 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 07:22:48.287894 140255202092800 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 07:22:48.287944 140255202092800 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0425 07:22:48.289530 140255202092800 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0425 07:22:48.289730 140390170629952 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0425 07:22:48.289877 140390170629952 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 07:22:48.290798 140390170629952 metric_logger.py:196] completed step: 9, seconds: 0.071, TFLOP/s/device: 11.354, Tokens/s/device: 1814.728, total_weights: 4096, loss: 6.981, lm_loss: 6.981, perplexity: 1075.867 Per train step: Total TFLOPs: 0.80 split as 99.60% learnable weight flops and 0.40% attention flops XPK End: Sat Apr 25 07:23:00 UTC 2026 EXIT_CODE=0
77202a8e9
· feat_nnx_trainstate_and_training_loop_20260425_092114
· full log
XPK Start: Sat Apr 25 09:26:32 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 09:26:56.791383: 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 09:26:57.360384 136863681455936 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 09:27:06.400663 136863681455936 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 09:27:06.403123 136863681455936 distributed.py:172] Connecting to JAX distributed service on mt-03-dropout-d4dsg-slice-job-0-0.mt-03-dropout-d4dsg:8482 I0425 09:27:07.709009 136863681455936 max_utils.py:284] Jax distributed system initialized! I0425 09:27:13.917650 136863681455936 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 09:27:13.917756 136863681455936 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 09:27:13.917796 136863681455936 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 09:27:13.917832 136863681455936 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 09:27:14.609410 136863681455936 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 09:27:14.609999 136863681455936 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 09:27:14.610180 136863681455936 checkpointing.py:688] Setting up checkpoint logger... I0425 09:27:14.610232 136863681455936 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 09:27:14.610275 136863681455936 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 09:27:14.610635 136863681455936 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 0x7c798538bd70>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 09:27:17.742618 136863681455936 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0425 09:27:17.742811 136863681455936 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 0x7c64ec64ee70>}, handler_registry=None I0425 09:27:17.743109 136863681455936 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c64ec64ee70>` 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 09:27:17.743162 136863681455936 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c64ec651fa0>` 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 09:27:17.743200 136863681455936 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 0x7c64ec64ee70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c64ec64ee70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c64ec651fa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c64ec651fa0>}). I0425 09:27:17.743531 136863681455936 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 09:27:17.743601 136863681455936 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c64ec3f1a80> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 09:27:18.438584 136863681455936 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints I0425 09:27:19.295921 136863681455936 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_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c64ec64de50> I0425 09:27:19.296095 136863681455936 checkpointing.py:302] Checkpoint manager created! I0425 09:27:19.648550 136863681455936 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 09:27:19.648659 136863681455936 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0425 09:27:20.027095 136863681455936 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,128,2048]....................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 09:27:20.027189 136863681455936 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0425 09:27:20.043751 136863681455936 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,128,2048]....................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 09:27:20.043811 136863681455936 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0425 09:27:20.067381 136863681455936 attentions.py:1154] attentions/query Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:27:20.067458 136863681455936 attentions.py:1154] attentions/query Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0425 09:27:20.084131 136863681455936 attentions.py:1155] attentions/key Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:27:20.084200 136863681455936 attentions.py:1155] attentions/key Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0425 09:27:20.100783 136863681455936 attentions.py:1156] attentions/value Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:27:20.100848 136863681455936 attentions.py:1156] attentions/value Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0425 09:27:20.125864 136863681455936 attentions.py:1198] attentions/out Logical: bfloat16[32,128,16,128]..................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 09:27:20.125944 136863681455936 attentions.py:1198] attentions/out Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0425 09:27:20.152766 136863681455936 linears.py:525] linears/x Logical: bfloat16[32,128,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 09:27:20.152833 136863681455936 linears.py:525] linears/x Physical: bfloat16[32,128,7168]....................................... ('fsdp', None, None). I0425 09:27:20.381170 136863681455936 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0425 09:27:20.381281 136863681455936 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 09:27:21.824726 136863681455936 maxtext_utils.py:1874] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 09:27:21.824853 136863681455936 maxtext_utils.py:1874] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0425 09:27:21.824908 136863681455936 maxtext_utils.py:1874] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0425 09:27:21.824965 136863681455936 maxtext_utils.py:1874] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0425 09:27:21.825018 136863681455936 maxtext_utils.py:1874] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0425 09:27:21.825056 136863681455936 maxtext_utils.py:1874] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0425 09:27:21.825111 136863681455936 maxtext_utils.py:1874] 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 09:27:21.825167 136863681455936 maxtext_utils.py:1874] 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 09:27:21.825208 136863681455936 maxtext_utils.py:1874] 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 09:27:21.825244 136863681455936 maxtext_utils.py:1874] 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 09:27:21.825292 136863681455936 maxtext_utils.py:1874] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 09:27:21.825342 136863681455936 maxtext_utils.py:1874] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 09:27:22.355309 136863681455936 train.py:157] train/xent Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0425 09:27:22.355401 136863681455936 train.py:157] train/xent Physical: float32[32,128]............................................. ('fsdp', None). I0425 09:27:22.371089 136863681455936 train.py:164] train/z_loss Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0425 09:27:22.371151 136863681455936 train.py:164] train/z_loss Physical: float32[32,128]............................................. ('fsdp', None). I0425 09:27:25.884128 136863681455936 max_utils.py:791] Total memory size: 0.8 GB, Output size: 0.4 GB, Temp size: 0.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0425 09:27:25.884943 136863681455936 metric_logger.py:301] number parameters: 1.104 billion I0425 09:27:30.058428 136863681455936 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0425 09:27:30.442641 136863681455936 checkpointing.py:798] Waited 0.38419389724731445 seconds for step 0 to finish before starting checkpointing. I0425 09:27:30.444994 136863681455936 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 09:27:30.446778 136863681455936 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0425 09:27:30.448090 136863681455936 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints/0. I0425 09:27:30.731200 136863681455936 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 09:27:30.732175 136863681455936 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 09:27:30.732230 136863681455936 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 09:27:31.009866 136863681455936 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278737s I0425 09:27:31.010045 136863681455936 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.428 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2842116355895996 s) (per-host) I0425 09:27:31.010096 136863681455936 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.284273s (batch_requests_ready=0.002339s, total_serialization_initiated=0.281865s, others=0.000069s) I0425 09:27:31.010181 136863681455936 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288273s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.288254) I0425 09:27:31.010930 136863681455936 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.56 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints/0. I0425 09:27:31.011284 136734830483200 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 09:47:31.011246 I0425 09:27:31.458025 136863681455936 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 09:27:31.458408 136732154521344 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0425 09:27:31.458593 136863681455936 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777109250.444976, 'wait_for_prev_duration_secs': 5.9604644775390625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777109250.446816, 'checkpointer_blocking_duration_secs': 0.56463623046875, 'get_old_steps_start_time': 1777109251.011478, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1777109250.4430413, 'checkpoint_manager_blocking_duration_secs': 1.0155088901519775} I0425 09:27:31.458702 136863681455936 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0425 09:27:31.458752 136863681455936 max_utils.py:750] Memstats: After params initialized: I0425 09:27:31.458809 136863681455936 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_18(process=5,(2,4,0,0)) I0425 09:27:31.458843 136863681455936 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_19(process=5,(3,4,0,0)) I0425 09:27:31.458872 136863681455936 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_22(process=5,(2,5,0,0)) I0425 09:27:31.458896 136863681455936 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_23(process=5,(3,5,0,0)) I0425 09:27:31.770545 136863681455936 metric_logger.py:196] completed step: 0, seconds: 4.173, TFLOP/s/device: 0.192, Tokens/s/device: 30.671, total_weights: 4096, loss: 10.889, lm_loss: 10.889, perplexity: 53560.699 I0425 09:27:31.877851 136863681455936 metric_logger.py:196] completed step: 1, seconds: 1.711, TFLOP/s/device: 0.468, Tokens/s/device: 74.828, total_weights: 4096, loss: 10.902, lm_loss: 10.902, perplexity: 54298.523 I0425 09:27:32.286062 136863681455936 metric_logger.py:196] completed step: 2, seconds: 0.037, TFLOP/s/device: 21.723, Tokens/s/device: 3471.940, total_weights: 4096, loss: 9.956, lm_loss: 9.956, perplexity: 21082.051 I0425 09:27:32.356645 136863681455936 metric_logger.py:196] completed step: 3, seconds: 0.409, TFLOP/s/device: 1.960, Tokens/s/device: 313.310, total_weights: 4096, loss: 9.139, lm_loss: 9.139, perplexity: 9314.691 I0425 09:27:32.498690 136863681455936 metric_logger.py:196] completed step: 4, seconds: 0.076, TFLOP/s/device: 10.506, Tokens/s/device: 1679.107, total_weights: 4096, loss: 8.459, lm_loss: 8.459, perplexity: 4715.493 I0425 09:27:32.505086 136863681455936 metric_logger.py:196] completed step: 5, seconds: 0.071, TFLOP/s/device: 11.337, Tokens/s/device: 1811.979, total_weights: 4096, loss: 7.924, lm_loss: 7.924, perplexity: 2762.373 I0425 09:27:34.496444 2853 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 09:27:36.502156 136732162914048 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints/0/items/array_metadatas/process_5 I0425 09:27:42.081782 136863681455936 metric_logger.py:196] completed step: 6, seconds: 0.143, TFLOP/s/device: 5.608, Tokens/s/device: 896.371, total_weights: 4096, loss: 7.553, lm_loss: 7.553, perplexity: 1906.617 I0425 09:27:42.152260 136863681455936 metric_logger.py:196] completed step: 7, seconds: 9.507, TFLOP/s/device: 0.084, Tokens/s/device: 13.463, total_weights: 4096, loss: 7.267, lm_loss: 7.267, perplexity: 1431.653 I0425 09:27:42.222902 136863681455936 metric_logger.py:196] completed step: 8, seconds: 0.075, TFLOP/s/device: 10.688, Tokens/s/device: 1708.238, total_weights: 4096, loss: 7.111, lm_loss: 7.111, perplexity: 1225.983 I0425 09:27:42.292975 136863681455936 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0425 09:27:42.293602 136863681455936 checkpointing.py:798] Waited 0.0006449222564697266 seconds for step 9 to finish before starting checkpointing. I0425 09:27:42.295747 136863681455936 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 09:28:02.409803 136734830483200 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 49.855 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 31.683929681777954 s) (per-host) I0425 09:28:02.409924 136734830483200 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 31.398529s. I0425 09:28:12.105286 136734830483200 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 41.093874s. I0425 09:28:12.105626 136732154521344 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 09:28:12.105745 136732154521344 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0425 09:28:12.105797 136732154521344 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 09:28:12.107873 136732154521344 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 09:28:12.108063 136863681455936 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. W0425 09:28:12.108201 136863681455936 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.812451 seconds. If this number is high, consider checkpointing less frequently. I0425 09:28:12.110124 136863681455936 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0425 09:28:12.112226 136863681455936 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints/9. I0425 09:28:12.399632 136863681455936 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 09:28:12.399722 136863681455936 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 09:28:12.433466 136863681455936 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034917s I0425 09:28:12.433632 136863681455936 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 40.305 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03827238082885742 s) (per-host) I0425 09:28:12.433682 136863681455936 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038332s (batch_requests_ready=0.001729s, total_serialization_initiated=0.036538s, others=0.000065s) I0425 09:28:12.433768 136863681455936 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042531s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.042516) I0425 09:28:12.434460 136863681455936 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints/9. I0425 09:28:12.434819 136732154521344 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 09:48:12.434781 I0425 09:28:12.438587 136863681455936 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0425 09:28:12.438886 136732162914048 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0425 09:28:12.439051 136863681455936 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777109262.2957184, 'wait_for_prev_duration_secs': 29.812451124191284, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777109292.1101632, 'checkpointer_blocking_duration_secs': 0.3248019218444824, 'get_old_steps_start_time': 1777109292.4349904, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1777109262.2938313, 'checkpoint_manager_blocking_duration_secs': 30.145184993743896} I0425 09:28:12.439153 136863681455936 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0425 09:28:12.439197 136863681455936 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 09:28:17.570568 136711161939712 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_03_dropout/checkpoints/9/items/array_metadatas/process_5 I0425 09:28:52.583523 136732154521344 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 39.305 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.18812108039856 s) (per-host) I0425 09:28:52.583643 136732154521344 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.148711s. I0425 09:29:01.395312 136732154521344 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 48.960365s. I0425 09:29:01.395594 136732162914048 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 09:29:01.395723 136732162914048 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0425 09:29:01.395770 136732162914048 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0425 09:29:01.397134 136732162914048 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0425 09:29:01.397325 136863681455936 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. I0425 09:29:01.397492 136863681455936 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 09:29:01.398491 136863681455936 metric_logger.py:196] completed step: 9, seconds: 0.070, TFLOP/s/device: 11.369, Tokens/s/device: 1817.124, total_weights: 4096, loss: 6.981, lm_loss: 6.981, perplexity: 1075.867 Per train step: Total TFLOPs: 0.80 split as 99.60% learnable weight flops and 0.40% attention flops XPK End: Sat Apr 25 09:29:12 UTC 2026 EXIT_CODE=0