main (574ad3fb9) vs feat/nnx-trainstate-and-training-loop (574ad3fb9)| Metric | main 574ad3fb9 | feat/nnx-trainstate-and-training-loop 574ad3fb9 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 6.9810 | 6.9810 | 0 |
| TFLOP/s | 11.299 | 11.111 | -0.188 |
| Tok/s | 1805.9 | 1775.9 | -30.041 |
| Avg s/step | 1.327 | 1.690 | +0.363 |
| 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: Sun Apr 19 18:14:41 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 18:15:05.619172: 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 18:15:06.154783 133482703050560 max_utils.py:273] Attempting to initialize the jax distributed system... I0419 18:15:15.194252 133482703050560 distributed.py:149] Starting JAX distributed service on [::]:8482 I0419 18:15:15.196616 133482703050560 distributed.py:172] Connecting to JAX distributed service on mt-03-dropout-6jpss-slice-job-0-0.mt-03-dropout-6jpss:8482 I0419 18:15:17.591546 133482703050560 max_utils.py:284] Jax distributed system initialized! I0419 18:15:23.695772 133482703050560 max_utils.py:800] System Information: Jax Version: 0.9.2 I0419 18:15:23.695871 133482703050560 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0419 18:15:23.695910 133482703050560 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 18:15:23.695945 133482703050560 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0419 18:15:24.389572 133482703050560 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 18:15:24.389842 133482703050560 checkpointing.py:677] Setting up checkpoint logger... I0419 18:15:24.389894 133482703050560 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0419 18:15:24.389938 133482703050560 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0419 18:15:24.390264 133482703050560 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 0x79662573d190>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0419 18:15:27.663002 133482703050560 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0419 18:15:27.663237 133482703050560 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 0x7951f07a6450>}, handler_registry=None I0419 18:15:27.663494 133482703050560 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7951f07a6450>` 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 18:15:27.663545 133482703050560 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7951f07a6660>` 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 18:15:27.663580 133482703050560 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 0x7951f07a6450>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7951f07a6450>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7951f07a6660>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7951f07a6660>}). I0419 18:15:27.663890 133482703050560 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0419 18:15:27.663960 133482703050560 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7951f03d7920> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0419 18:15:29.345293 133482703050560 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints I0419 18:15:29.351955 133482703050560 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_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7951f07a6270> I0419 18:15:29.352080 133482703050560 checkpointing.py:301] Checkpoint manager created! I0419 18:15:29.697951 133482703050560 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0419 18:15:29.698065 133482703050560 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0419 18:15:30.078252 133482703050560 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 18:15:30.078343 133482703050560 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0419 18:15:30.094433 133482703050560 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 18:15:30.094491 133482703050560 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0419 18:15:30.118172 133482703050560 attentions.py:1154] attentions/query Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 18:15:30.118241 133482703050560 attentions.py:1154] attentions/query Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0419 18:15:30.134423 133482703050560 attentions.py:1155] attentions/key Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 18:15:30.134484 133482703050560 attentions.py:1155] attentions/key Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0419 18:15:30.150620 133482703050560 attentions.py:1156] attentions/value Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 18:15:30.150679 133482703050560 attentions.py:1156] attentions/value Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0419 18:15:30.175965 133482703050560 attentions.py:1197] attentions/out Logical: bfloat16[32,128,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0419 18:15:30.176038 133482703050560 attentions.py:1197] attentions/out Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0419 18:15:30.202764 133482703050560 linears.py:525] linears/x Logical: bfloat16[32,128,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0419 18:15:30.202833 133482703050560 linears.py:525] linears/x Physical: bfloat16[32,128,7168]....................................... ('fsdp', None, None). I0419 18:15:30.436529 133482703050560 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0419 18:15:30.436636 133482703050560 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0419 18:15:31.871581 133482703050560 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0419 18:15:31.871706 133482703050560 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, None) I0419 18:15:31.871760 133482703050560 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, None) I0419 18:15:31.871817 133482703050560 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0419 18:15:31.871869 133482703050560 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 18:15:31.871913 133482703050560 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 18:15:31.871968 133482703050560 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, None, None) I0419 18:15:31.872043 133482703050560 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: (None, None, None, 'fsdp') I0419 18:15:31.872088 133482703050560 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, None, None) I0419 18:15:31.872127 133482703050560 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, None, None) I0419 18:15:31.872177 133482703050560 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0419 18:15:31.872226 133482703050560 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0419 18:15:32.396288 133482703050560 train.py:155] train/xent Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0419 18:15:32.396385 133482703050560 train.py:155] train/xent Physical: float32[32,128]............................................. ('fsdp', None). I0419 18:15:32.411733 133482703050560 train.py:162] train/z_loss Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0419 18:15:32.411795 133482703050560 train.py:162] train/z_loss Physical: float32[32,128]............................................. ('fsdp', None). I0419 18:15:35.959335 133482703050560 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. I0419 18:15:35.960190 133482703050560 metric_logger.py:301] number parameters: 1.104 billion I0419 18:15:40.163894 133482703050560 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0419 18:15:40.550569 133482703050560 checkpointing.py:776] Waited 0.38665318489074707 seconds for step 0 to finish before starting checkpointing. I0419 18:15:40.553098 133482703050560 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 18:15:40.555039 133482703050560 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0419 18:15:40.556483 133482703050560 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints/0. I0419 18:15:41.276967 133482703050560 signaling_client.py:364] Using JaxDistributedSignalingClient I0419 18:15:41.277831 133482703050560 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 18:15:41.277888 133482703050560 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 18:15:41.561460 133482703050560 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.284565s I0419 18:15:41.561643 133482703050560 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.318 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2900667190551758 s) (per-host) I0419 18:15:41.561697 133482703050560 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.290132s (batch_requests_ready=0.002347s, total_serialization_initiated=0.287710s, others=0.000075s) I0419 18:15:41.561788 133482703050560 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.294329s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.294310) I0419 18:15:41.562536 133482703050560 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.01 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints/0. I0419 18:15:41.562859 133354730096384 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 18:35:41.562821 I0419 18:15:41.567186 133482703050560 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0419 18:15:41.567503 133353663616768 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0419 18:15:41.567664 133482703050560 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776622540.5530803, 'wait_for_prev_duration_secs': 5.9604644775390625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776622540.5550764, 'checkpointer_blocking_duration_secs': 1.0079340934753418, 'get_old_steps_start_time': 1776622541.563034, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776622540.5510917, 'checkpoint_manager_blocking_duration_secs': 1.016530990600586} I0419 18:15:41.567774 133482703050560 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0419 18:15:41.567824 133482703050560 max_utils.py:750] Memstats: After params initialized: I0419 18:15:41.567873 133482703050560 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_24(process=6,(0,6,0,0)) I0419 18:15:41.567904 133482703050560 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_25(process=6,(1,6,0,0)) I0419 18:15:41.567931 133482703050560 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_28(process=6,(0,7,0,0)) I0419 18:15:41.567955 133482703050560 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_29(process=6,(1,7,0,0)) I0419 18:15:41.885023 133482703050560 metric_logger.py:196] completed step: 0, seconds: 4.203, TFLOP/s/device: 0.191, Tokens/s/device: 30.451, total_weights: 4096, loss: 10.889, lm_loss: 10.889, perplexity: 53560.699 I0419 18:15:41.974500 133482703050560 metric_logger.py:196] completed step: 1, seconds: 1.720, TFLOP/s/device: 0.466, Tokens/s/device: 74.434, total_weights: 4096, loss: 10.902, lm_loss: 10.902, perplexity: 54298.523 I0419 18:15:42.387519 133482703050560 metric_logger.py:196] completed step: 2, seconds: 0.019, TFLOP/s/device: 42.616, Tokens/s/device: 6811.047, total_weights: 4096, loss: 9.956, lm_loss: 9.956, perplexity: 21082.051 I0419 18:15:42.458229 133482703050560 metric_logger.py:196] completed step: 3, seconds: 0.413, TFLOP/s/device: 1.937, Tokens/s/device: 309.654, total_weights: 4096, loss: 9.139, lm_loss: 9.139, perplexity: 9314.691 I0419 18:15:42.600160 133482703050560 metric_logger.py:196] completed step: 4, seconds: 0.076, TFLOP/s/device: 10.518, Tokens/s/device: 1680.981, total_weights: 4096, loss: 8.459, lm_loss: 8.459, perplexity: 4715.493 I0419 18:15:42.605970 133482703050560 metric_logger.py:196] completed step: 5, seconds: 0.071, TFLOP/s/device: 11.326, Tokens/s/device: 1810.185, total_weights: 4096, loss: 7.924, lm_loss: 7.924, perplexity: 2762.373 I0419 18:15:44.841188 2818 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0419 18:15:47.298632 133353672009472 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints/0/items/array_metadatas/process_6 I0419 18:15:52.034139 133482703050560 metric_logger.py:196] completed step: 6, seconds: 0.142, TFLOP/s/device: 5.622, Tokens/s/device: 898.605, total_weights: 4096, loss: 7.553, lm_loss: 7.553, perplexity: 1906.617 I0419 18:15:52.105050 133482703050560 metric_logger.py:196] completed step: 7, seconds: 9.359, TFLOP/s/device: 0.086, Tokens/s/device: 13.677, total_weights: 4096, loss: 7.267, lm_loss: 7.267, perplexity: 1431.653 I0419 18:15:52.177129 133482703050560 metric_logger.py:196] completed step: 8, seconds: 0.075, TFLOP/s/device: 10.683, Tokens/s/device: 1707.486, total_weights: 4096, loss: 7.111, lm_loss: 7.111, perplexity: 1225.983 I0419 18:15:52.247215 133482703050560 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0419 18:15:52.247913 133482703050560 checkpointing.py:776] Waited 0.0007162094116210938 seconds for step 9 to finish before starting checkpointing. I0419 18:15:52.249906 133482703050560 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 18:16:13.814811 133354730096384 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 48.538 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 32.54321074485779 s) (per-host) I0419 18:16:13.814917 133354730096384 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.251946s. I0419 18:16:22.795621 133354730096384 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 41.232633s. I0419 18:16:22.795893 133353663616768 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 18:16:22.796015 133353663616768 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0419 18:16:22.796067 133353663616768 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0419 18:16:22.797733 133353663616768 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0419 18:16:22.797898 133482703050560 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. W0419 18:16:22.798028 133482703050560 checkpoint_manager.py:1452] Waiting for previous save to complete took 30.548126 seconds. If this number is high, consider checkpointing less frequently. I0419 18:16:22.799839 133482703050560 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0419 18:16:22.801812 133482703050560 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints/9. I0419 18:16:23.507758 133482703050560 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 18:16:23.507853 133482703050560 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 18:16:23.544453 133482703050560 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037736s I0419 18:16:23.544617 133482703050560 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 37.402 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04124307632446289 s) (per-host) I0419 18:16:23.544665 133482703050560 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.041301s (batch_requests_ready=0.001889s, total_serialization_initiated=0.039347s, others=0.000065s) I0419 18:16:23.544752 133482703050560 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.045302s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.045287) I0419 18:16:23.545494 133482703050560 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints/9. I0419 18:16:23.545752 133353663616768 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 18:36:23.545731 I0419 18:16:23.547609 133482703050560 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0419 18:16:23.547825 133354209621760 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0419 18:16:23.547921 133482703050560 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776622552.2498765, 'wait_for_prev_duration_secs': 30.548125982284546, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776622582.7998781, 'checkpointer_blocking_duration_secs': 0.7459628582000732, 'get_old_steps_start_time': 1776622583.5458658, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1776622552.2481365, 'checkpoint_manager_blocking_duration_secs': 31.299751043319702} I0419 18:16:23.548085 133482703050560 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0419 18:16:23.548132 133482703050560 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 18:16:29.107107 133348822611712 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_03_dropout/checkpoints/9/items/array_metadatas/process_6 I0419 18:17:04.628642 133353663616768 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.409 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.125226736068726 s) (per-host) I0419 18:17:04.628774 133353663616768 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.082956s. I0419 18:17:12.887315 133353663616768 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.341481s. I0419 18:17:12.887562 133354209621760 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 18:17:12.887635 133354209621760 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0419 18:17:12.887695 133354209621760 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0419 18:17:12.889267 133354209621760 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0419 18:17:12.889442 133482703050560 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. I0419 18:17:12.889594 133482703050560 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 18:17:12.890653 133482703050560 metric_logger.py:196] completed step: 9, seconds: 0.071, TFLOP/s/device: 11.299, Tokens/s/device: 1805.895, 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: Sun Apr 19 18:17:21 UTC 2026 EXIT_CODE=0
574ad3fb9
· feat_nnx_trainstate_and_training_loop_20260419_214525
· full log
XPK Start: Sun Apr 19 21:58:54 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 21:59:19.068567: 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 21:59:19.247097 140459403499328 max_utils.py:273] Attempting to initialize the jax distributed system... I0419 21:59:28.287106 140459403499328 distributed.py:149] Starting JAX distributed service on [::]:8482 I0419 21:59:28.289421 140459403499328 distributed.py:172] Connecting to JAX distributed service on mt-03-dropout-w041y-slice-job-0-0.mt-03-dropout-w041y:8482 I0419 21:59:31.043092 140459403499328 max_utils.py:284] Jax distributed system initialized! I0419 21:59:37.158908 140459403499328 max_utils.py:800] System Information: Jax Version: 0.9.2 I0419 21:59:37.159008 140459403499328 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0419 21:59:37.159046 140459403499328 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 21:59:37.159079 140459403499328 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0419 21:59:38.212983 140459403499328 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 21:59:38.213578 140459403499328 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 21:59:38.213754 140459403499328 checkpointing.py:688] Setting up checkpoint logger... I0419 21:59:38.213803 140459403499328 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0419 21:59:38.213844 140459403499328 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0419 21:59:38.214172 140459403499328 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 0x7fbe891530b0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0419 21:59:42.065934 140459403499328 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0419 21:59:42.066116 140459403499328 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fbd7cede180>}, handler_registry=None I0419 21:59:42.066362 140459403499328 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fbd7cede180>` 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 21:59:42.066411 140459403499328 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fbd7cf210a0>` 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 21:59:42.066447 140459403499328 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 0x7fbd7cede180>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fbd7cede180>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fbd7cf210a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fbd7cf210a0>}). I0419 21:59:42.066763 140459403499328 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0419 21:59:42.066835 140459403499328 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7faa202bb740> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0419 21:59:45.008913 140459403499328 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints I0419 21:59:45.011185 140459403499328 checkpoint_manager.py:929] [process=4][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_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7fbd7cf23fe0> I0419 21:59:45.011294 140459403499328 checkpointing.py:302] Checkpoint manager created! I0419 21:59:45.360044 140459403499328 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0419 21:59:45.360144 140459403499328 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0419 21:59:45.735981 140459403499328 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 21:59:45.736068 140459403499328 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0419 21:59:45.752318 140459403499328 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 21:59:45.752392 140459403499328 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None). I0419 21:59:45.775968 140459403499328 attentions.py:1154] attentions/query Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 21:59:45.776032 140459403499328 attentions.py:1154] attentions/query Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0419 21:59:45.792112 140459403499328 attentions.py:1155] attentions/key Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 21:59:45.792166 140459403499328 attentions.py:1155] attentions/key Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0419 21:59:45.808098 140459403499328 attentions.py:1156] attentions/value Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 21:59:45.808155 140459403499328 attentions.py:1156] attentions/value Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0419 21:59:45.833486 140459403499328 attentions.py:1197] attentions/out Logical: bfloat16[32,128,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0419 21:59:45.833562 140459403499328 attentions.py:1197] attentions/out Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None). I0419 21:59:45.859417 140459403499328 linears.py:525] linears/x Logical: bfloat16[32,128,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0419 21:59:45.859480 140459403499328 linears.py:525] linears/x Physical: bfloat16[32,128,7168]....................................... ('fsdp', None, None). I0419 21:59:46.079460 140459403499328 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0419 21:59:46.079565 140459403499328 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0419 21:59:47.505699 140459403499328 maxtext_utils.py:1821] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0419 21:59:47.505826 140459403499328 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 21:59:47.505886 140459403499328 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 21:59:47.505944 140459403499328 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0419 21:59:47.505998 140459403499328 maxtext_utils.py:1821] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 21:59:47.506071 140459403499328 maxtext_utils.py:1821] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 21:59:47.506139 140459403499328 maxtext_utils.py:1821] 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) I0419 21:59:47.506193 140459403499328 maxtext_utils.py:1821] 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') I0419 21:59:47.506234 140459403499328 maxtext_utils.py:1821] 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) I0419 21:59:47.506269 140459403499328 maxtext_utils.py:1821] 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) I0419 21:59:47.506316 140459403499328 maxtext_utils.py:1821] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0419 21:59:47.506392 140459403499328 maxtext_utils.py:1821] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0419 21:59:48.028398 140459403499328 train.py:157] train/xent Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0419 21:59:48.028490 140459403499328 train.py:157] train/xent Physical: float32[32,128]............................................. ('fsdp', None). I0419 21:59:48.043816 140459403499328 train.py:164] train/z_loss Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0419 21:59:48.043873 140459403499328 train.py:164] train/z_loss Physical: float32[32,128]............................................. ('fsdp', None). I0419 21:59:51.573023 140459403499328 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. I0419 21:59:51.573889 140459403499328 metric_logger.py:301] number parameters: 1.104 billion I0419 21:59:55.825962 140459403499328 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0419 21:59:55.901834 140459403499328 checkpointing.py:798] Waited 0.07585287094116211 seconds for step 0 to finish before starting checkpointing. I0419 21:59:55.904158 140459403499328 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 21:59:55.906237 140459403499328 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0419 21:59:55.907706 140459403499328 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints/0. I0419 21:59:57.043075 140459403499328 signaling_client.py:364] Using JaxDistributedSignalingClient I0419 21:59:57.044009 140459403499328 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 21:59:57.044064 140459403499328 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 21:59:57.322092 140459403499328 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279089s I0419 21:59:57.322272 140459403499328 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.423 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2844374179840088 s) (per-host) I0419 21:59:57.322326 140459403499328 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.284503s (batch_requests_ready=0.002155s, total_serialization_initiated=0.282275s, others=0.000073s) I0419 21:59:57.322444 140459403499328 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288810s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.288794) I0419 21:59:57.323279 140459403499328 event_tracking.py:125] [process=4] [async] Finished blocking save in 1.42 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints/0. I0419 21:59:57.323626 140330338875136 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 22:19:57.323589 I0419 21:59:57.336109 140459403499328 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0419 21:59:57.336403 140330284332800 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0419 21:59:57.336555 140459403499328 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776635995.9041398, 'wait_for_prev_duration_secs': 6.389617919921875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776635995.906277, 'checkpointer_blocking_duration_secs': 1.4175059795379639, 'get_old_steps_start_time': 1776635997.3238082, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776635995.9022923, 'checkpoint_manager_blocking_duration_secs': 1.4342217445373535} I0419 21:59:57.336665 140459403499328 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0419 21:59:57.336717 140459403499328 max_utils.py:750] Memstats: After params initialized: I0419 21:59:57.336768 140459403499328 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_16(process=4,(0,4,0,0)) I0419 21:59:57.336803 140459403499328 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_17(process=4,(1,4,0,0)) I0419 21:59:57.336831 140459403499328 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_20(process=4,(0,5,0,0)) I0419 21:59:57.336857 140459403499328 max_utils.py:756] Using (GB) 0.41 / 31.25 (1.312000%) on TPU_21(process=4,(1,5,0,0)) I0419 21:59:57.650826 140459403499328 metric_logger.py:196] completed step: 0, seconds: 4.252, TFLOP/s/device: 0.188, Tokens/s/device: 30.104, total_weights: 4096, loss: 10.889, lm_loss: 10.889, perplexity: 53560.699 I0419 21:59:57.747386 140459403499328 metric_logger.py:196] completed step: 1, seconds: 1.823, TFLOP/s/device: 0.439, Tokens/s/device: 70.201, total_weights: 4096, loss: 10.902, lm_loss: 10.902, perplexity: 54298.523 I0419 21:59:58.577263 140459403499328 metric_logger.py:196] completed step: 2, seconds: 0.026, TFLOP/s/device: 30.326, Tokens/s/device: 4846.833, total_weights: 4096, loss: 9.956, lm_loss: 9.956, perplexity: 21082.051 I0419 21:59:58.647988 140459403499328 metric_logger.py:196] completed step: 3, seconds: 0.830, TFLOP/s/device: 0.964, Tokens/s/device: 154.151, total_weights: 4096, loss: 9.139, lm_loss: 9.139, perplexity: 9314.691 I0419 21:59:58.790224 140459403499328 metric_logger.py:196] completed step: 4, seconds: 0.076, TFLOP/s/device: 10.516, Tokens/s/device: 1680.783, total_weights: 4096, loss: 8.459, lm_loss: 8.459, perplexity: 4715.493 I0419 21:59:58.796494 140459403499328 metric_logger.py:196] completed step: 5, seconds: 0.071, TFLOP/s/device: 11.312, Tokens/s/device: 1808.012, total_weights: 4096, loss: 7.924, lm_loss: 7.924, perplexity: 2762.373 I0419 22:00:01.244379 2803 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0419 22:00:04.989454 140330292725504 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints/0/items/array_metadatas/process_4 I0419 22:00:10.957105 140459403499328 metric_logger.py:196] completed step: 6, seconds: 0.143, TFLOP/s/device: 5.620, Tokens/s/device: 898.208, total_weights: 4096, loss: 7.553, lm_loss: 7.553, perplexity: 1906.617 I0419 22:00:11.029131 140459403499328 metric_logger.py:196] completed step: 7, seconds: 12.091, TFLOP/s/device: 0.066, Tokens/s/device: 10.586, total_weights: 4096, loss: 7.267, lm_loss: 7.267, perplexity: 1431.653 I0419 22:00:11.099808 140459403499328 metric_logger.py:196] completed step: 8, seconds: 0.075, TFLOP/s/device: 10.626, Tokens/s/device: 1698.311, total_weights: 4096, loss: 7.111, lm_loss: 7.111, perplexity: 1225.983 I0419 22:00:11.169790 140459403499328 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0419 22:00:11.170450 140459403499328 checkpointing.py:798] Waited 0.0006768703460693359 seconds for step 9 to finish before starting checkpointing. I0419 22:00:11.172713 140459403499328 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 22:00:31.059661 140330338875136 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 46.429 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.02179002761841 s) (per-host) I0419 22:00:31.059788 140330338875136 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.736050s. I0419 22:00:40.090674 140330338875136 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 42.766918s. I0419 22:00:40.090943 140330284332800 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 22:00:40.091063 140330284332800 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0419 22:00:40.091122 140330284332800 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0419 22:00:40.093317 140330284332800 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0419 22:00:40.093518 140459403499328 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0419 22:00:40.093658 140459403499328 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.920947 seconds. If this number is high, consider checkpointing less frequently. I0419 22:00:40.095648 140459403499328 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0419 22:00:40.097756 140459403499328 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints/9. I0419 22:00:40.823173 140459403499328 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 22:00:40.823260 140459403499328 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 22:00:40.862549 140459403499328 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.040505s I0419 22:00:40.862717 140459403499328 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 35.164 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.043868064880371094 s) (per-host) I0419 22:00:40.862773 140459403499328 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043931s (batch_requests_ready=0.001731s, total_serialization_initiated=0.042130s, others=0.000069s) I0419 22:00:40.862895 140459403499328 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048257s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.048241) I0419 22:00:40.863562 140459403499328 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints/9. I0419 22:00:40.863898 140329796728576 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 22:20:40.863859 I0419 22:00:40.865848 140459403499328 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0419 22:00:40.866114 140330284332800 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0419 22:00:40.866270 140459403499328 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776636011.1726825, 'wait_for_prev_duration_secs': 28.920947313308716, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776636040.095687, 'checkpointer_blocking_duration_secs': 0.768359899520874, 'get_old_steps_start_time': 1776636040.8640697, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776636011.1706715, 'checkpoint_manager_blocking_duration_secs': 29.69556498527527} I0419 22:00:40.866392 140459403499328 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0419 22:00:40.866438 140459403499328 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 22:00:48.878462 140306838042368 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_03_dropout/checkpoints/9/items/array_metadatas/process_4 I0419 22:01:23.678103 140329796728576 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 36.855 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.859214067459106 s) (per-host) I0419 22:01:23.678228 140329796728576 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.814216s. I0419 22:01:33.484940 140329796728576 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 52.620912s. I0419 22:01:33.485243 140330284332800 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 22:01:33.485367 140330284332800 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0419 22:01:33.485413 140330284332800 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0419 22:01:33.487994 140330284332800 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0419 22:01:33.488176 140459403499328 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0419 22:01:33.488360 140459403499328 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 22:01:33.489350 140459403499328 metric_logger.py:196] completed step: 9, seconds: 0.072, TFLOP/s/device: 11.111, Tokens/s/device: 1775.854, 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: Sun Apr 19 22:01:45 UTC 2026 EXIT_CODE=0