main (147168c3a) vs feat/nnx-trainstate-and-training-loop (190761531)| Metric | main 147168c3a | feat/nnx-trainstate-and-training-loop 190761531 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1790 | 8.1790 | 0 |
| TFLOP/s | 92.292 | 92.283 | -0.009 |
| Tok/s | 13911.2 | 13909.9 | -1.323 |
| Avg s/step | 3.010 | 2.799 | -0.211 |
| 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 00:04:43 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 00:05:08.699021: 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 00:05:09.233546 134760609109824 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 00:05:18.274761 134760609109824 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 00:05:18.277187 134760609109824 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-jsi18-slice-job-0-0.mt-02-synthetic-jsi18:8482 I0418 00:05:19.871809 134760609109824 max_utils.py:284] Jax distributed system initialized! I0418 00:05:26.016683 134760609109824 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 00:05:26.016785 134760609109824 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 00:05:26.016824 134760609109824 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 00:05:26.016858 134760609109824 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 00:05:26.708018 134760609109824 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 00:05:26.708298 134760609109824 checkpointing.py:677] Setting up checkpoint logger... I0418 00:05:26.708350 134760609109824 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 00:05:26.708395 134760609109824 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 00:05:26.708805 134760609109824 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 0x7a8fcfb1b4d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 00:05:30.410940 134760609109824 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0418 00:05:30.411177 134760609109824 checkpoint_manager.py:708] [process=1][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a7ba02d5b50>}, handler_registry=None I0418 00:05:30.411437 134760609109824 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a7ba02d5b50>` 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 00:05:30.411498 134760609109824 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a7b88708740>` 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 00:05:30.411535 134760609109824 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 0x7a7ba02d5b50>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a7ba02d5b50>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a7b88708740>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a7b88708740>}). I0418 00:05:30.411863 134760609109824 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 00:05:30.411934 134760609109824 async_checkpointer.py:192] [process=1][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7a7b8849bc40> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 00:05:31.997223 134760609109824 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints I0418 00:05:32.835627 134760609109824 checkpoint_manager.py:929] [process=1][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_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7a7b88701790> I0418 00:05:32.835787 134760609109824 checkpointing.py:301] Checkpoint manager created! I0418 00:05:33.756928 134760609109824 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 00:05:33.757051 134760609109824 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 00:05:34.136436 134760609109824 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 00:05:34.136548 134760609109824 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 00:05:34.152654 134760609109824 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 00:05:34.152714 134760609109824 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 00:05:34.176204 134760609109824 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 00:05:34.176272 134760609109824 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 00:05:34.192477 134760609109824 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 00:05:34.192537 134760609109824 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 00:05:34.208591 134760609109824 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 00:05:34.208649 134760609109824 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 00:05:34.233908 134760609109824 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 00:05:34.233979 134760609109824 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 00:05:34.254607 134760609109824 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 00:05:34.254673 134760609109824 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 00:05:34.470966 134760609109824 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0418 00:05:34.471077 134760609109824 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 00:05:35.895006 134760609109824 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 00:05:35.895130 134760609109824 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 00:05:35.895182 134760609109824 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 00:05:35.895238 134760609109824 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 00:05:35.895290 134760609109824 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 00:05:35.895329 134760609109824 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 00:05:35.895383 134760609109824 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 00:05:35.895438 134760609109824 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 00:05:35.895496 134760609109824 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 00:05:35.895536 134760609109824 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 00:05:35.895584 134760609109824 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 00:05:35.895631 134760609109824 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 00:05:36.379543 134760609109824 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 00:05:36.379639 134760609109824 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 00:05:36.394860 134760609109824 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 00:05:36.394920 134760609109824 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 00:05:47.364952 134760609109824 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 00:05:47.365742 134760609109824 metric_logger.py:301] number parameters: 1.104 billion I0418 00:05:58.808921 134760609109824 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0418 00:05:58.961389 134760609109824 checkpointing.py:776] Waited 0.15244674682617188 seconds for step 0 to finish before starting checkpointing. I0418 00:05:58.964130 134760609109824 checkpoint_manager.py:2009] [process=1][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 00:05:58.965722 134760609109824 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 0 I0418 00:05:58.967255 134760609109824 event_tracking.py:70] [process=1] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints/0. I0418 00:05:59.729075 134760609109824 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 00:05:59.730046 134760609109824 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 00:05:59.730105 134760609109824 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 00:06:00.006512 134760609109824 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.277511s I0418 00:06:00.006682 134760609109824 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 5.453 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28287315368652344 s) (per-host) I0418 00:06:00.006739 134760609109824 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.282940s (batch_requests_ready=0.002197s, total_serialization_initiated=0.280669s, others=0.000074s) I0418 00:06:00.006834 134760609109824 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287574s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.287555) I0418 00:06:00.007558 134760609109824 event_tracking.py:125] [process=1] [async] Finished blocking save in 1.04 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints/0. I0418 00:06:00.007841 134630598481664 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 00:26:00.007800 I0418 00:06:00.018283 134760609109824 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 00:06:00.018570 134632231216896 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save. I0418 00:06:00.018725 134760609109824 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776470758.9641116, 'wait_for_prev_duration_secs': 6.198883056640625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776470758.965762, 'checkpointer_blocking_duration_secs': 1.0422389507293701, 'get_old_steps_start_time': 1776470760.0080237, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1776470758.961883, 'checkpoint_manager_blocking_duration_secs': 1.0568056106567383} I0418 00:06:00.018832 134760609109824 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0418 00:06:00.018892 134760609109824 max_utils.py:750] Memstats: After params initialized: I0418 00:06:00.018943 134760609109824 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_2(process=1,(2,0,0,0)) I0418 00:06:00.018978 134760609109824 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_3(process=1,(3,0,0,0)) I0418 00:06:00.019003 134760609109824 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_6(process=1,(2,1,0,0)) I0418 00:06:00.019028 134760609109824 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_7(process=1,(3,1,0,0)) I0418 00:06:00.328201 134760609109824 metric_logger.py:196] completed step: 0, seconds: 11.443, TFLOP/s/device: 1.187, Tokens/s/device: 178.973, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 00:06:00.504634 134760609109824 metric_logger.py:196] completed step: 1, seconds: 1.518, TFLOP/s/device: 8.952, Tokens/s/device: 1349.365, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 00:06:00.906298 134760609109824 metric_logger.py:196] completed step: 2, seconds: 0.029, TFLOP/s/device: 466.511, Tokens/s/device: 70317.597, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0418 00:06:01.053775 134760609109824 metric_logger.py:196] completed step: 3, seconds: 0.402, TFLOP/s/device: 33.778, Tokens/s/device: 5091.399, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0418 00:06:01.348857 134760609109824 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.972, Tokens/s/device: 13410.777, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0418 00:06:01.354846 134760609109824 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.199, Tokens/s/device: 13897.277, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0418 00:06:05.120651 2816 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 00:06:06.981716 134630618089216 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints/0/items/array_metadatas/process_1 I0418 00:06:25.744389 134760609109824 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.955, Tokens/s/device: 6926.852, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0418 00:06:25.891682 134760609109824 metric_logger.py:196] completed step: 7, seconds: 24.243, TFLOP/s/device: 0.560, Tokens/s/device: 84.479, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0418 00:06:26.039178 134760609109824 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.279, Tokens/s/device: 13457.128, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0418 00:06:26.185644 134760609109824 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0418 00:06:26.186290 134760609109824 checkpointing.py:776] Waited 0.0006680488586425781 seconds for step 9 to finish before starting checkpointing. I0418 00:06:26.188531 134760609109824 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 00:06:35.835058 134630598481664 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 43.743 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.11120963096619 s) (per-host) I0418 00:06:35.835177 134630598481664 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 35.827222s. I0418 00:06:47.946554 134630598481664 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 47.938583s. I0418 00:06:47.946844 134632231216896 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 00:06:47.946973 134632231216896 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save. I0418 00:06:47.947023 134632231216896 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 00:06:47.951328 134632231216896 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 00:06:47.951544 134760609109824 checkpoint_manager.py:2032] [process=1][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0418 00:06:47.951675 134760609109824 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.763145 seconds. If this number is high, consider checkpointing less frequently. I0418 00:06:47.953551 134760609109824 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 9 I0418 00:06:47.955726 134760609109824 event_tracking.py:70] [process=1] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints/9. I0418 00:06:48.672383 134760609109824 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 00:06:48.672488 134760609109824 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 00:06:48.712367 134760609109824 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041084s I0418 00:06:48.712543 134760609109824 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 34.693 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.044463396072387695 s) (per-host) I0418 00:06:48.712596 134760609109824 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044527s (batch_requests_ready=0.001728s, total_serialization_initiated=0.042727s, others=0.000072s) I0418 00:06:48.712686 134760609109824 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048767s (all_items=0.000015s, per_item={'items': '0.00001550'}, temp_paths=0.048752) I0418 00:06:48.713367 134760609109824 event_tracking.py:125] [process=1] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints/9. I0418 00:06:48.713727 134632231216896 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 00:26:48.713691 I0418 00:06:48.715674 134760609109824 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 00:06:48.715971 134631722579712 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save. I0418 00:06:48.716114 134760609109824 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776470786.188502, 'wait_for_prev_duration_secs': 21.763144731521606, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776470807.9535913, 'checkpointer_blocking_duration_secs': 0.7602841854095459, 'get_old_steps_start_time': 1776470808.7138996, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776470786.186538, 'checkpoint_manager_blocking_duration_secs': 22.529542207717896} I0418 00:06:48.716280 134760609109824 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0418 00:06:48.716331 134760609109824 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 00:06:56.345995 134630618089216 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_02_synthetic/checkpoints/9/items/array_metadatas/process_1 I0418 00:07:30.568964 134632231216896 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 37.698 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.90084767341614 s) (per-host) I0418 00:07:30.569087 134632231216896 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.855247s. I0418 00:07:40.653762 134632231216896 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 51.939905s. I0418 00:07:40.654027 134631722579712 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 00:07:40.654149 134631722579712 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save. I0418 00:07:40.654198 134631722579712 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 00:07:40.656504 134631722579712 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 00:07:40.656684 134760609109824 checkpoint_manager.py:2032] [process=1][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0418 00:07:40.656843 134760609109824 checkpoint_manager.py:2009] [process=1][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 00:07:40.657910 134760609109824 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.292, Tokens/s/device: 13911.248, 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 00:07:49 UTC 2026 EXIT_CODE=0
190761531
· feat_nnx_trainstate_and_training_loop_20260418_060141
· full log
XPK Start: Sat Apr 18 06:10:05 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 06:10:29.672043: 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 06:10:29.852503 134008598599488 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 06:10:38.894081 134008598599488 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 06:10:38.896386 134008598599488 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-vb5jr-slice-job-0-0.mt-02-synthetic-vb5jr:8482 I0418 06:10:40.908177 134008598599488 max_utils.py:284] Jax distributed system initialized! I0418 06:10:46.936250 134008598599488 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 06:10:46.936354 134008598599488 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 06:10:46.936395 134008598599488 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 06:10:46.936430 134008598599488 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 06:10:48.008181 134008598599488 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 06:10:48.008781 134008598599488 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 06:10:48.008957 134008598599488 checkpointing.py:688] Setting up checkpoint logger... I0418 06:10:48.009005 134008598599488 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 06:10:48.009047 134008598599488 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 06:10:48.009405 134008598599488 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 0x79e097350d10>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 06:10:52.279821 134008598599488 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0418 06:10:52.280009 134008598599488 checkpoint_manager.py:708] [process=2][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79df8b4c27e0>}, handler_registry=None I0418 06:10:52.280262 134008598599488 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79df8b4c27e0>` 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 06:10:52.280311 134008598599488 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79df8b5385c0>` 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 06:10:52.280346 134008598599488 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 0x79df8b4c27e0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79df8b4c27e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79df8b5385c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79df8b5385c0>}). I0418 06:10:52.280667 134008598599488 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 06:10:52.280734 134008598599488 async_checkpointer.py:192] [process=2][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x79df8b648360> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 06:10:53.330866 134008598599488 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints I0418 06:10:54.209665 134008598599488 checkpoint_manager.py:929] [process=2][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_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x79cc880f5100> I0418 06:10:54.209839 134008598599488 checkpointing.py:302] Checkpoint manager created! I0418 06:10:55.143819 134008598599488 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 06:10:55.143924 134008598599488 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 06:10:55.524585 134008598599488 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 06:10:55.524675 134008598599488 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 06:10:55.540878 134008598599488 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 06:10:55.540938 134008598599488 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 06:10:55.564335 134008598599488 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 06:10:55.564400 134008598599488 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 06:10:55.580656 134008598599488 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 06:10:55.580717 134008598599488 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 06:10:55.596824 134008598599488 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 06:10:55.596883 134008598599488 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 06:10:55.621178 134008598599488 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 06:10:55.621254 134008598599488 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 06:10:55.641971 134008598599488 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 06:10:55.642036 134008598599488 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 06:10:55.855521 134008598599488 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0418 06:10:55.855629 134008598599488 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 06:10:57.299191 134008598599488 maxtext_utils.py:1821] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 06:10:57.299340 134008598599488 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) I0418 06:10:57.299395 134008598599488 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) I0418 06:10:57.299454 134008598599488 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') I0418 06:10:57.299513 134008598599488 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) I0418 06:10:57.299554 134008598599488 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) I0418 06:10:57.299607 134008598599488 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) I0418 06:10:57.299669 134008598599488 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') I0418 06:10:57.299709 134008598599488 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) I0418 06:10:57.299746 134008598599488 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) I0418 06:10:57.299795 134008598599488 maxtext_utils.py:1821] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 06:10:57.299843 134008598599488 maxtext_utils.py:1821] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 06:10:57.797126 134008598599488 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 06:10:57.797240 134008598599488 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 06:10:57.812573 134008598599488 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 06:10:57.812636 134008598599488 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 06:11:08.793683 134008598599488 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 06:11:08.794445 134008598599488 metric_logger.py:301] number parameters: 1.104 billion I0418 06:11:20.031253 134008598599488 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0418 06:11:20.185589 134008598599488 checkpointing.py:798] Waited 0.15430998802185059 seconds for step 0 to finish before starting checkpointing. I0418 06:11:20.187880 134008598599488 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 06:11:20.189589 134008598599488 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 0 I0418 06:11:20.191043 134008598599488 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints/0. I0418 06:11:20.956854 134008598599488 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 06:11:20.957833 134008598599488 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 06:11:20.957888 134008598599488 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 06:11:21.242630 134008598599488 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.285846s I0418 06:11:21.242800 134008598599488 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 5.289 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.29167723655700684 s) (per-host) I0418 06:11:21.242853 134008598599488 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.291742s (batch_requests_ready=0.002627s, total_serialization_initiated=0.289043s, others=0.000071s) I0418 06:11:21.242939 134008598599488 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.295552s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.295534) I0418 06:11:21.243793 134008598599488 event_tracking.py:125] [process=2] [async] Finished blocking save in 1.05 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints/0. I0418 06:11:21.244138 133880451770112 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 06:31:21.244101 I0418 06:11:21.246126 134008598599488 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 06:11:21.246465 133878322996992 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0418 06:11:21.246601 134008598599488 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776492680.1878622, 'wait_for_prev_duration_secs': 5.984306335449219e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776492680.189627, 'checkpointer_blocking_duration_secs': 1.0546317100524902, 'get_old_steps_start_time': 1776492681.244285, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1776492680.1859763, 'checkpoint_manager_blocking_duration_secs': 1.0605874061584473} I0418 06:11:21.246762 134008598599488 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0418 06:11:21.246814 134008598599488 max_utils.py:750] Memstats: After params initialized: I0418 06:11:21.246863 134008598599488 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_8(process=2,(0,2,0,0)) I0418 06:11:21.246895 134008598599488 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_9(process=2,(1,2,0,0)) I0418 06:11:21.246922 134008598599488 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_12(process=2,(0,3,0,0)) I0418 06:11:21.246945 134008598599488 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_13(process=2,(1,3,0,0)) I0418 06:11:21.560156 134008598599488 metric_logger.py:196] completed step: 0, seconds: 11.237, TFLOP/s/device: 1.209, Tokens/s/device: 182.260, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 06:11:21.734909 134008598599488 metric_logger.py:196] completed step: 1, seconds: 1.527, TFLOP/s/device: 8.897, Tokens/s/device: 1341.089, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 06:11:22.155299 134008598599488 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 489.115, Tokens/s/device: 73724.756, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0418 06:11:22.302526 134008598599488 metric_logger.py:196] completed step: 3, seconds: 0.421, TFLOP/s/device: 32.293, Tokens/s/device: 4867.603, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0418 06:11:22.597944 134008598599488 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.785, Tokens/s/device: 13382.647, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0418 06:11:22.603867 134008598599488 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.421, Tokens/s/device: 13930.646, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0418 06:11:26.227402 2806 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 06:11:30.048967 133878865979136 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints/0/items/array_metadatas/process_2 I0418 06:11:45.068910 134008598599488 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.915, Tokens/s/device: 6920.836, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0418 06:11:45.216323 134008598599488 metric_logger.py:196] completed step: 7, seconds: 22.318, TFLOP/s/device: 0.609, Tokens/s/device: 91.763, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0418 06:11:45.363694 134008598599488 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.220, Tokens/s/device: 13448.203, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0418 06:11:45.510374 134008598599488 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0418 06:11:45.510970 134008598599488 checkpointing.py:798] Waited 0.0006196498870849609 seconds for step 9 to finish before starting checkpointing. I0418 06:11:45.513170 134008598599488 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 06:11:59.551069 133880451770112 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 40.922 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.59992599487305 s) (per-host) I0418 06:11:59.551178 133880451770112 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.306404s. I0418 06:12:11.185171 133880451770112 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 49.940380s. I0418 06:12:11.185526 133878322996992 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 06:12:11.185682 133878322996992 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0418 06:12:11.185741 133878322996992 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 06:12:11.187468 133878322996992 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 06:12:11.187636 134008598599488 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0418 06:12:11.187774 134008598599488 checkpoint_manager.py:1452] Waiting for previous save to complete took 25.674600 seconds. If this number is high, consider checkpointing less frequently. I0418 06:12:11.190639 134008598599488 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 9 I0418 06:12:11.192792 134008598599488 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints/9. I0418 06:12:11.910575 134008598599488 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 06:12:11.910671 134008598599488 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 06:12:11.948357 134008598599488 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038806s I0418 06:12:11.948523 134008598599488 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 36.380 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.042402029037475586 s) (per-host) I0418 06:12:11.948574 134008598599488 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.042462s (batch_requests_ready=0.001911s, total_serialization_initiated=0.040484s, others=0.000067s) I0418 06:12:11.948662 134008598599488 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.046523s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=0.046508) I0418 06:12:11.950095 134008598599488 event_tracking.py:125] [process=2] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints/9. I0418 06:12:11.950444 133878322996992 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 06:32:11.950407 I0418 06:12:11.952486 134008598599488 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 06:12:11.952733 133878865979136 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0418 06:12:11.952836 134008598599488 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776492705.5131397, 'wait_for_prev_duration_secs': 25.67460036277771, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776492731.1906784, 'checkpointer_blocking_duration_secs': 0.7599115371704102, 'get_old_steps_start_time': 1776492731.950614, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1776492705.5112207, 'checkpoint_manager_blocking_duration_secs': 26.441579341888428} I0418 06:12:11.952988 134008598599488 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0418 06:12:11.953035 134008598599488 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 06:12:19.096733 133857680549632 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_02_synthetic/checkpoints/9/items/array_metadatas/process_2 I0418 06:12:54.666024 133878322996992 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 36.941 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.75986170768738 s) (per-host) I0418 06:12:54.666150 133878322996992 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.715589s. I0418 06:13:05.274932 133878322996992 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 53.324356s. I0418 06:13:05.275248 133878865979136 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 06:13:05.275386 133878865979136 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0418 06:13:05.275437 133878865979136 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 06:13:05.277073 133878865979136 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 06:13:05.277261 134008598599488 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0418 06:13:05.277422 134008598599488 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 06:13:05.278346 134008598599488 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.283, Tokens/s/device: 13909.925, 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 06:13:15 UTC 2026 EXIT_CODE=0