main (574ad3fb9) vs feat/nnx-post-train-fixes (574ad3fb9)| Metric | main 574ad3fb9 | feat/nnx-post-train-fixes 574ad3fb9 | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1790 | 8.1790 | 0 |
| TFLOP/s | 92.234 | 92.209 | -0.025 |
| Tok/s | 13902.5 | 13898.8 | -3.68 |
| Avg s/step | 3.071 | 2.837 | -0.234 |
| Memory % | 1.38 | 1.38 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Sat Apr 18 18:01:13 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-18 18:01:38.448324: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0418 18:01:38.627807 136016632297280 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 18:01:47.667797 136016632297280 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 18:01:47.670194 136016632297280 distributed.py:172] Connecting to JAX distributed service on mt-01-base-2sag8-slice-job-0-0.mt-01-base-2sag8:8482 I0418 18:01:49.137647 136016632297280 max_utils.py:284] Jax distributed system initialized! I0418 18:01:56.197314 136016632297280 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 18:01:56.197413 136016632297280 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 18:01:56.197469 136016632297280 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0418 18:01:56.197514 136016632297280 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 18:01:56.911277 136016632297280 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 18:01:56.911563 136016632297280 checkpointing.py:677] Setting up checkpoint logger... I0418 18:01:56.911619 136016632297280 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 18:01:56.911664 136016632297280 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 18:01:56.912001 136016632297280 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 0x7bb436918260>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 18:02:01.482331 136016632297280 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0418 18:02:01.482582 136016632297280 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 0x7b9f743869c0>}, handler_registry=None I0418 18:02:01.482824 136016632297280 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b9f743869c0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0418 18:02:01.482872 136016632297280 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b9f74388290>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0418 18:02:01.482908 136016632297280 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 0x7b9f743869c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b9f743869c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b9f74388290>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b9f74388290>}). I0418 18:02:01.483229 136016632297280 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 18:02:01.483299 136016632297280 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b9f546afba0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 18:02:02.509403 136016632297280 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints I0418 18:02:03.389142 136016632297280 checkpoint_manager.py:929] [process=5][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7bb315bdffb0> I0418 18:02:03.389307 136016632297280 checkpointing.py:301] Checkpoint manager created! I0418 18:02:04.318331 136016632297280 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 18:02:04.318439 136016632297280 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:02:04.697327 136016632297280 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 18:02:04.697414 136016632297280 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:02:04.713699 136016632297280 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 18:02:04.713760 136016632297280 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:02:04.737531 136016632297280 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:02:04.737595 136016632297280 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:02:04.753716 136016632297280 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:02:04.753776 136016632297280 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:02:04.769909 136016632297280 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:02:04.769973 136016632297280 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:02:04.794766 136016632297280 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 18:02:04.794831 136016632297280 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:02:04.815628 136016632297280 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 18:02:04.815692 136016632297280 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 18:02:05.031522 136016632297280 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0418 18:02:05.031626 136016632297280 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 18:02:06.457116 136016632297280 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 18:02:06.457235 136016632297280 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) I0418 18:02:06.457286 136016632297280 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) I0418 18:02:06.457343 136016632297280 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') I0418 18:02:06.457395 136016632297280 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) I0418 18:02:06.457432 136016632297280 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) I0418 18:02:06.457497 136016632297280 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) I0418 18:02:06.457551 136016632297280 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') I0418 18:02:06.457590 136016632297280 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) I0418 18:02:06.457626 136016632297280 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) I0418 18:02:06.457673 136016632297280 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 18:02:06.457719 136016632297280 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 18:02:06.945318 136016632297280 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 18:02:06.945410 136016632297280 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 18:02:06.960833 136016632297280 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 18:02:06.960896 136016632297280 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 18:02:17.925596 136016632297280 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0418 18:02:17.926374 136016632297280 metric_logger.py:301] number parameters: 1.104 billion I0418 18:02:28.987720 136016632297280 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0418 18:02:29.319164 136016632297280 checkpointing.py:776] Waited 0.3314321041107178 seconds for step 0 to finish before starting checkpointing. I0418 18:02:29.321613 136016632297280 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 18:02:29.323310 136016632297280 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0418 18:02:29.324758 136016632297280 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints/0. I0418 18:02:30.509867 136016632297280 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 18:02:30.510792 136016632297280 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 18:02:30.510847 136016632297280 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 18:02:30.788037 136016632297280 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278241s I0418 18:02:30.788214 136016632297280 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.429 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28411173820495605 s) (per-host) I0418 18:02:30.788267 136016632297280 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.284174s (batch_requests_ready=0.002627s, total_serialization_initiated=0.281476s, others=0.000071s) I0418 18:02:30.788356 136016632297280 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288152s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.288136) I0418 18:02:30.789228 136016632297280 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.47 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints/0. I0418 18:02:30.789592 135888973575936 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 18:22:30.789554 I0418 18:02:30.796347 136016632297280 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:02:30.796626 135887344273152 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:02:30.796753 136016632297280 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776535349.3215866, 'wait_for_prev_duration_secs': 8.726119995117188e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776535349.3233628, 'checkpointer_blocking_duration_secs': 1.466383457183838, 'get_old_steps_start_time': 1776535350.7897708, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776535349.319707, 'checkpoint_manager_blocking_duration_secs': 1.4770071506500244} I0418 18:02:30.796863 136016632297280 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0418 18:02:30.796917 136016632297280 max_utils.py:750] Memstats: After params initialized: I0418 18:02:30.796967 136016632297280 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0)) I0418 18:02:30.797000 136016632297280 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0418 18:02:30.797027 136016632297280 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0418 18:02:30.797050 136016632297280 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0418 18:02:31.105650 136016632297280 metric_logger.py:196] completed step: 0, seconds: 11.061, TFLOP/s/device: 1.228, Tokens/s/device: 185.152, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 18:02:31.296370 136016632297280 metric_logger.py:196] completed step: 1, seconds: 2.116, TFLOP/s/device: 6.420, Tokens/s/device: 967.643, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 18:02:31.692446 136016632297280 metric_logger.py:196] completed step: 2, seconds: 0.044, TFLOP/s/device: 312.262, Tokens/s/device: 47067.476, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0418 18:02:31.839858 136016632297280 metric_logger.py:196] completed step: 3, seconds: 0.397, TFLOP/s/device: 34.233, Tokens/s/device: 5159.990, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0418 18:02:32.135124 136016632297280 metric_logger.py:196] completed step: 4, seconds: 0.152, TFLOP/s/device: 89.128, Tokens/s/device: 13434.353, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0418 18:02:32.140931 136016632297280 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.207, Tokens/s/device: 13898.409, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0418 18:02:36.769223 2828 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 18:02:40.319857 135887393478400 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints/0/items/array_metadatas/process_5 I0418 18:02:56.479434 136016632297280 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.968, Tokens/s/device: 6928.844, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0418 18:02:56.626775 136016632297280 metric_logger.py:196] completed step: 7, seconds: 24.192, TFLOP/s/device: 0.562, Tokens/s/device: 84.658, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0418 18:02:56.774048 136016632297280 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.207, Tokens/s/device: 13446.261, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0418 18:02:56.920902 136016632297280 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0418 18:02:56.921597 136016632297280 checkpointing.py:776] Waited 0.0007135868072509766 seconds for step 9 to finish before starting checkpointing. I0418 18:02:56.923789 136016632297280 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:03:09.217653 135888973575936 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 40.802 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.713520526885986 s) (per-host) I0418 18:03:09.217759 135888973575936 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.428054s. I0418 18:03:21.081743 135888973575936 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.292021s. I0418 18:03:21.082024 135887344273152 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:03:21.082161 135887344273152 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:03:21.082236 135887344273152 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:03:21.084198 135887344273152 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 18:03:21.084340 136016632297280 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0418 18:03:21.084431 136016632297280 checkpoint_manager.py:1452] Waiting for previous save to complete took 24.160645 seconds. If this number is high, consider checkpointing less frequently. I0418 18:03:21.086312 136016632297280 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0418 18:03:21.088404 136016632297280 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints/9. I0418 18:03:21.805513 136016632297280 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 18:03:21.805606 136016632297280 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 18:03:21.851402 136016632297280 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.047068s I0418 18:03:21.851588 136016632297280 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 30.451 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.05065727233886719 s) (per-host) I0418 18:03:21.851641 136016632297280 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.050721s (batch_requests_ready=0.001908s, total_serialization_initiated=0.048741s, others=0.000072s) I0418 18:03:21.851732 136016632297280 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.054832s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=0.054816) I0418 18:03:21.852479 136016632297280 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints/9. I0418 18:03:21.852841 135887344273152 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 18:23:21.852802 I0418 18:03:21.854827 136016632297280 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:03:21.855110 135887393478400 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:03:21.855244 136016632297280 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776535376.9237592, 'wait_for_prev_duration_secs': 24.160645008087158, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776535401.086351, 'checkpointer_blocking_duration_secs': 0.766655445098877, 'get_old_steps_start_time': 1776535401.8530304, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776535376.9218338, 'checkpoint_manager_blocking_duration_secs': 24.933377981185913} I0418 18:03:21.855412 136016632297280 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0418 18:03:21.855471 136016632297280 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:03:29.938111 135865812625152 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_01_base/checkpoints/9/items/array_metadatas/process_5 I0418 18:04:05.668946 135887344273152 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 36.008 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.86797738075256 s) (per-host) I0418 18:04:05.669079 135887344273152 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.816113s. I0418 18:04:14.249446 135887344273152 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 52.396466s. I0418 18:04:14.249764 135887393478400 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:04:14.249889 135887393478400 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:04:14.249945 135887393478400 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:04:14.252395 135887393478400 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 18:04:14.252585 136016632297280 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0418 18:04:14.252735 136016632297280 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 18:04:14.253656 136016632297280 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.234, Tokens/s/device: 13902.466, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sat Apr 18 18:04:23 UTC 2026 EXIT_CODE=0
XPK Start: Sat Apr 18 23:51:17 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-18 23:51:42.057267: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0418 23:51:42.234272 139108507039552 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 23:51:51.275598 139108507039552 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 23:51:51.277922 139108507039552 distributed.py:172] Connecting to JAX distributed service on mt-01-base-5ovt2-slice-job-0-0.mt-01-base-5ovt2:8482 I0418 23:51:55.207773 139108507039552 max_utils.py:284] Jax distributed system initialized! I0418 23:52:01.472810 139108507039552 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 23:52:01.472913 139108507039552 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 23:52:01.472953 139108507039552 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0418 23:52:01.472986 139108507039552 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 23:52:02.549135 139108507039552 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 23:52:02.549827 139108507039552 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 23:52:02.550009 139108507039552 checkpointing.py:688] Setting up checkpoint logger... I0418 23:52:02.550060 139108507039552 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 23:52:02.550101 139108507039552 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 23:52:02.550452 139108507039552 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 0x7e842cf0a600>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 23:52:06.825747 139108507039552 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0418 23:52:06.825983 139108507039552 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 0x7e82f5ae3170>}, handler_registry=None I0418 23:52:06.826218 139108507039552 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e82f5ae3170>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0418 23:52:06.826297 139108507039552 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e82f5ae2bd0>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0418 23:52:06.826345 139108507039552 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 0x7e82f5ae3170>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e82f5ae3170>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e82f5ae2bd0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e82f5ae2bd0>}). I0418 23:52:06.826740 139108507039552 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 23:52:06.826812 139108507039552 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e6ec43cec00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 23:52:08.283997 139108507039552 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints I0418 23:52:08.729521 139108507039552 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_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e82f59ea270> I0418 23:52:08.729692 139108507039552 checkpointing.py:302] Checkpoint manager created! I0418 23:52:09.657986 139108507039552 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 23:52:09.658091 139108507039552 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 23:52:10.042268 139108507039552 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 23:52:10.042361 139108507039552 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 23:52:10.058639 139108507039552 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 23:52:10.058702 139108507039552 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 23:52:10.082466 139108507039552 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 23:52:10.082547 139108507039552 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 23:52:10.098946 139108507039552 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 23:52:10.099013 139108507039552 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 23:52:10.115288 139108507039552 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 23:52:10.115349 139108507039552 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 23:52:10.139869 139108507039552 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 23:52:10.139943 139108507039552 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 23:52:10.161647 139108507039552 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 23:52:10.161714 139108507039552 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 23:52:10.376447 139108507039552 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0418 23:52:10.376556 139108507039552 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 23:52:11.884388 139108507039552 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 23:52:11.884511 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 23:52:11.884563 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 23:52:11.884620 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0418 23:52:11.884673 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 23:52:11.884724 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 23:52:11.884781 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0418 23:52:11.884834 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0418 23:52:11.884874 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0418 23:52:11.884909 139108507039552 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0418 23:52:11.884955 139108507039552 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 23:52:11.885003 139108507039552 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 23:52:12.384192 139108507039552 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 23:52:12.384299 139108507039552 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 23:52:12.399750 139108507039552 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 23:52:12.399809 139108507039552 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 23:52:22.870283 139108507039552 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0418 23:52:22.871152 139108507039552 metric_logger.py:301] number parameters: 1.104 billion I0418 23:52:34.262472 139108507039552 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0418 23:52:34.658338 139108507039552 checkpointing.py:798] Waited 0.3958439826965332 seconds for step 0 to finish before starting checkpointing. I0418 23:52:34.660838 139108507039552 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 23:52:34.662690 139108507039552 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0418 23:52:34.664049 139108507039552 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints/0. I0418 23:52:35.416614 139108507039552 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 23:52:35.417567 139108507039552 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 23:52:35.417627 139108507039552 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 23:52:35.693761 139108507039552 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.277223s I0418 23:52:35.693931 139108507039552 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.455 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28279662132263184 s) (per-host) I0418 23:52:35.693984 139108507039552 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.282859s (batch_requests_ready=0.002367s, total_serialization_initiated=0.280421s, others=0.000070s) I0418 23:52:35.694073 139108507039552 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287230s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.287213) I0418 23:52:35.694903 139108507039552 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.03 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints/0. I0418 23:52:35.695279 138979013400320 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:12:35.695221 I0418 23:52:35.702432 139108507039552 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 23:52:35.702751 138978988222208 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0418 23:52:35.702881 139108507039552 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776556354.6608195, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776556354.6627269, 'checkpointer_blocking_duration_secs': 1.0327222347259521, 'get_old_steps_start_time': 1776556355.6954741, 'get_old_steps_duration_secs': 3.3855438232421875e-05, 'checkpoint_manager_blocking_start_time': 1776556354.6588771, 'checkpoint_manager_blocking_duration_secs': 1.0439646244049072} I0418 23:52:35.702986 139108507039552 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0418 23:52:35.703037 139108507039552 max_utils.py:750] Memstats: After params initialized: I0418 23:52:35.703084 139108507039552 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0418 23:52:35.703116 139108507039552 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0418 23:52:35.703141 139108507039552 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0418 23:52:35.703164 139108507039552 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0418 23:52:36.016721 139108507039552 metric_logger.py:196] completed step: 0, seconds: 11.391, TFLOP/s/device: 1.193, Tokens/s/device: 179.788, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 23:52:36.194932 139108507039552 metric_logger.py:196] completed step: 1, seconds: 1.753, TFLOP/s/device: 7.752, Tokens/s/device: 1168.432, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 23:52:36.587102 139108507039552 metric_logger.py:196] completed step: 2, seconds: 0.031, TFLOP/s/device: 439.855, Tokens/s/device: 66299.773, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0418 23:52:36.734630 139108507039552 metric_logger.py:196] completed step: 3, seconds: 0.393, TFLOP/s/device: 34.609, Tokens/s/device: 5216.572, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0418 23:52:37.029587 139108507039552 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.942, Tokens/s/device: 13406.300, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0418 23:52:37.035388 139108507039552 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.197, Tokens/s/device: 13896.900, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0418 23:52:40.246731 2847 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 23:52:43.114006 138978996614912 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints/0/items/array_metadatas/process_6 I0418 23:52:59.643822 139108507039552 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.976, Tokens/s/device: 6930.040, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0418 23:52:59.791174 139108507039552 metric_logger.py:196] completed step: 7, seconds: 22.461, TFLOP/s/device: 0.605, Tokens/s/device: 91.178, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0418 23:52:59.938665 139108507039552 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.166, Tokens/s/device: 13439.996, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0418 23:53:00.085189 139108507039552 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0418 23:53:00.085804 139108507039552 checkpointing.py:798] Waited 0.0006372928619384766 seconds for step 9 to finish before starting checkpointing. I0418 23:53:00.087976 139108507039552 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 23:53:14.416257 138979013400320 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 40.497 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 39.00505614280701 s) (per-host) I0418 23:53:14.416381 138979013400320 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.720975s. I0418 23:53:24.189558 138979013400320 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.494135s. I0418 23:53:24.189840 138978988222208 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 23:53:24.189964 138978988222208 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0418 23:53:24.190019 138978988222208 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 23:53:24.192598 138978988222208 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 23:53:24.192777 139108507039552 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. W0418 23:53:24.192915 139108507039552 checkpoint_manager.py:1452] Waiting for previous save to complete took 24.104943 seconds. If this number is high, consider checkpointing less frequently. I0418 23:53:24.194827 139108507039552 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0418 23:53:24.197014 139108507039552 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints/9. I0418 23:53:24.896249 139108507039552 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 23:53:24.896343 139108507039552 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 23:53:24.935354 139108507039552 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.040211s I0418 23:53:24.935535 139108507039552 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 35.230 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.043785810470581055 s) (per-host) I0418 23:53:24.935585 139108507039552 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043845s (batch_requests_ready=0.001903s, total_serialization_initiated=0.041875s, others=0.000067s) I0418 23:53:24.935676 139108507039552 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048057s (all_items=0.000014s, per_item={'items': '0.00001383'}, temp_paths=0.048043) I0418 23:53:24.936396 139108507039552 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints/9. I0418 23:53:24.936692 138978988222208 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:13:24.936660 I0418 23:53:24.938726 139108507039552 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 23:53:24.939009 138978996614912 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0418 23:53:24.939168 139108507039552 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776556380.0879447, 'wait_for_prev_duration_secs': 24.104942798614502, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776556404.1948636, 'checkpointer_blocking_duration_secs': 0.7419703006744385, 'get_old_steps_start_time': 1776556404.9368553, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776556380.0860353, 'checkpoint_manager_blocking_duration_secs': 24.85309886932373} I0418 23:53:24.939286 139108507039552 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0418 23:53:24.939331 139108507039552 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 23:53:32.205334 138954984638208 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_01_base/checkpoints/9/items/array_metadatas/process_6 I0418 23:54:08.769095 138978988222208 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.000 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.8773090839386 s) (per-host) I0418 23:54:08.769205 138978988222208 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.832404s. I0418 23:54:15.919665 138978988222208 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.982848s. I0418 23:54:15.919974 138978996614912 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 23:54:15.920107 138978996614912 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0418 23:54:15.920157 138978996614912 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 23:54:15.921958 138978996614912 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 23:54:15.922123 139108507039552 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. I0418 23:54:15.922286 139108507039552 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 23:54:15.923219 139108507039552 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.209, Tokens/s/device: 13898.786, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sat Apr 18 23:54:27 UTC 2026 EXIT_CODE=0