main (574ad3fb9) vs feat/nnx-trainstate-and-training-loop (574ad3fb9)| Metric | main 574ad3fb9 | feat/nnx-trainstate-and-training-loop 574ad3fb9 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1750 | 8.1750 | 0 |
| TFLOP/s | 59.290 | 59.195 | -0.095 |
| Tok/s | 8936.9 | 8922.5 | -14.445 |
| Avg s/step | 3.606 | 3.554 | -0.052 |
| 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: Mon Apr 20 06:10:29 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-20 06:10:53.179083: 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) I0420 06:10:53.711882 139341872637760 max_utils.py:273] Attempting to initialize the jax distributed system... I0420 06:11:02.752271 139341872637760 distributed.py:149] Starting JAX distributed service on [::]:8482 I0420 06:11:02.754640 139341872637760 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-s3hmc-slice-job-0-0.mt-04-int8-s3hmc:8482 I0420 06:11:04.212276 139341872637760 max_utils.py:284] Jax distributed system initialized! I0420 06:11:10.382898 139341872637760 max_utils.py:800] System Information: Jax Version: 0.9.2 I0420 06:11:10.383000 139341872637760 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0420 06:11:10.383040 139341872637760 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 I0420 06:11:10.383074 139341872637760 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0420 06:11:11.068904 139341872637760 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 06:11:11.069288 139341872637760 checkpointing.py:677] Setting up checkpoint logger... I0420 06:11:11.069355 139341872637760 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0420 06:11:11.069400 139341872637760 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0420 06:11:11.069736 139341872637760 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 0x7eba7afea6c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0420 06:11:14.406631 139341872637760 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0420 06:11:14.406860 139341872637760 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ea68413ee70>}, handler_registry=None I0420 06:11:14.407094 139341872637760 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ea68413ee70>` 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`. I0420 06:11:14.407141 139341872637760 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ea6442292b0>` 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`. I0420 06:11:14.407176 139341872637760 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 0x7ea68413ee70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ea68413ee70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ea6442292b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ea6442292b0>}). I0420 06:11:14.407507 139341872637760 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0420 06:11:14.407582 139341872637760 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7ea5245837e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0420 06:11:15.771504 139341872637760 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints I0420 06:11:15.773730 139341872637760 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ea644477440> I0420 06:11:15.773840 139341872637760 checkpointing.py:301] Checkpoint manager created! I0420 06:11:16.911753 139341872637760 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0420 06:11:16.911866 139341872637760 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 06:11:17.288491 139341872637760 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 06:11:17.288582 139341872637760 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 06:11:17.304770 139341872637760 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 06:11:17.304830 139341872637760 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 06:11:17.370537 139341872637760 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 06:11:17.370622 139341872637760 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 06:11:17.387021 139341872637760 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 06:11:17.387081 139341872637760 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 06:11:17.403186 139341872637760 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 06:11:17.403241 139341872637760 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 06:11:17.427483 139341872637760 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0420 06:11:17.427546 139341872637760 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 06:11:17.493270 139341872637760 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0420 06:11:17.493358 139341872637760 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0420 06:11:18.003211 139341872637760 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0420 06:11:18.003318 139341872637760 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0420 06:11:20.016225 139341872637760 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0420 06:11:20.016365 139341872637760 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) I0420 06:11:20.016420 139341872637760 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) I0420 06:11:20.016477 139341872637760 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') I0420 06:11:20.016534 139341872637760 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) I0420 06:11:20.016578 139341872637760 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) I0420 06:11:20.016632 139341872637760 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) I0420 06:11:20.016684 139341872637760 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') I0420 06:11:20.016725 139341872637760 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) I0420 06:11:20.016762 139341872637760 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) I0420 06:11:20.016808 139341872637760 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0420 06:11:20.016853 139341872637760 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0420 06:11:21.820974 139341872637760 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0420 06:11:21.821072 139341872637760 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0420 06:11:21.836486 139341872637760 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0420 06:11:21.836545 139341872637760 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0420 06:11:35.197810 139341872637760 max_utils.py:791] Total memory size: 1.8 GB, Output size: 0.4 GB, Temp size: 1.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0420 06:11:35.198616 139341872637760 metric_logger.py:301] number parameters: 1.104 billion I0420 06:11:50.319031 139341872637760 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0420 06:11:50.558239 139341872637760 checkpointing.py:776] Waited 0.23919248580932617 seconds for step 0 to finish before starting checkpointing. I0420 06:11:50.564233 139341872637760 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 06:11:50.565783 139341872637760 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0420 06:11:50.569496 139341872637760 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints/0. I0420 06:11:50.915031 139341872637760 signaling_client.py:364] Using JaxDistributedSignalingClient I0420 06:11:50.916085 139341872637760 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 06:11:50.916141 139341872637760 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 I0420 06:11:51.203214 139341872637760 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.288258s I0420 06:11:51.203402 139341872637760 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.252 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2936890125274658 s) (per-host) I0420 06:11:51.203458 139341872637760 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.293756s (batch_requests_ready=0.002203s, total_serialization_initiated=0.291478s, others=0.000075s) I0420 06:11:51.203553 139341872637760 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.298063s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.298045) I0420 06:11:51.204382 139341872637760 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.64 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints/0. I0420 06:11:51.204645 139208162293504 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 06:31:51.204616 I0420 06:11:51.206622 139341872637760 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0420 06:11:51.206840 139209282246400 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0420 06:11:51.206934 139341872637760 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776665510.564214, 'wait_for_prev_duration_secs': 6.437301635742188e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776665510.5658221, 'checkpointer_blocking_duration_secs': 0.6389296054840088, 'get_old_steps_start_time': 1776665511.2047763, 'get_old_steps_duration_secs': 3.528594970703125e-05, 'checkpoint_manager_blocking_start_time': 1776665510.5588, 'checkpoint_manager_blocking_duration_secs': 0.6480984687805176} I0420 06:11:51.207090 139341872637760 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0420 06:11:51.207140 139341872637760 max_utils.py:750] Memstats: After params initialized: I0420 06:11:51.207187 139341872637760 max_utils.py:756] Using (GB) 0.44 / 31.25 (1.408000%) on TPU_16(process=4,(0,4,0,0)) I0420 06:11:51.207219 139341872637760 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0420 06:11:51.207245 139341872637760 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0420 06:11:51.207269 139341872637760 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0420 06:11:51.536281 139341872637760 metric_logger.py:196] completed step: 0, seconds: 15.120, TFLOP/s/device: 0.899, Tokens/s/device: 135.447, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0420 06:11:51.774803 139341872637760 metric_logger.py:196] completed step: 1, seconds: 1.216, TFLOP/s/device: 11.176, Tokens/s/device: 1684.528, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0420 06:11:52.189761 139341872637760 metric_logger.py:196] completed step: 2, seconds: 0.009, TFLOP/s/device: 1504.499, Tokens/s/device: 226774.444, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820 I0420 06:11:52.419369 139341872637760 metric_logger.py:196] completed step: 3, seconds: 0.416, TFLOP/s/device: 32.661, Tokens/s/device: 4922.947, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199 I0420 06:11:52.879920 139341872637760 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.875, Tokens/s/device: 8723.580, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531 I0420 06:11:52.885873 139341872637760 metric_logger.py:196] completed step: 5, seconds: 0.229, TFLOP/s/device: 59.212, Tokens/s/device: 8925.070, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941 I0420 06:11:54.362723 2950 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0420 06:11:56.871751 139208693069568 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints/0/items/array_metadatas/process_4 I0420 06:12:22.537898 139341872637760 metric_logger.py:196] completed step: 6, seconds: 0.461, TFLOP/s/device: 29.483, Tokens/s/device: 4443.953, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468 I0420 06:12:22.767249 139341872637760 metric_logger.py:196] completed step: 7, seconds: 29.424, TFLOP/s/device: 0.462, Tokens/s/device: 69.603, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779 I0420 06:12:22.996757 139341872637760 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.115, Tokens/s/device: 8759.661, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453 I0420 06:12:23.225599 139341872637760 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0420 06:12:23.226294 139341872637760 checkpointing.py:776] Waited 0.0007171630859375 seconds for step 9 to finish before starting checkpointing. I0420 06:12:23.228499 139341872637760 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 06:12:24.565229 139208162293504 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 46.934 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 33.655495405197144 s) (per-host) I0420 06:12:24.565348 139208162293504 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.360614s. I0420 06:12:33.172482 139208162293504 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 41.967742s. I0420 06:12:33.172706 139209282246400 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 06:12:33.172773 139209282246400 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0420 06:12:33.172822 139209282246400 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0420 06:12:33.174700 139209282246400 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0420 06:12:33.174857 139341872637760 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0420 06:12:33.174991 139341872637760 checkpoint_manager.py:1452] Waiting for previous save to complete took 9.946491 seconds. If this number is high, consider checkpointing less frequently. I0420 06:12:33.176856 139341872637760 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0420 06:12:33.179001 139341872637760 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints/9. I0420 06:12:33.462944 139341872637760 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 06:12:33.463035 139341872637760 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 I0420 06:12:33.494695 139341872637760 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.033660s I0420 06:12:33.494866 139341872637760 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 41.604 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03707766532897949 s) (per-host) I0420 06:12:33.494919 139341872637760 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.037141s (batch_requests_ready=0.001764s, total_serialization_initiated=0.035306s, others=0.000071s) I0420 06:12:33.495010 139341872637760 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.041377s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=0.041361) I0420 06:12:33.495620 139341872637760 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints/9. I0420 06:12:33.495931 139209260123904 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 06:32:33.495892 I0420 06:12:33.499048 139341872637760 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0420 06:12:33.499313 139209282246400 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0420 06:12:33.499481 139341872637760 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776665543.2284691, 'wait_for_prev_duration_secs': 9.946491479873657, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776665553.176895, 'checkpointer_blocking_duration_secs': 0.31917881965637207, 'get_old_steps_start_time': 1776665553.4960935, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1776665543.2265308, 'checkpoint_manager_blocking_duration_secs': 10.2729172706604} I0420 06:12:33.499598 139341872637760 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0420 06:12:33.499642 139341872637760 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 06:12:38.542168 139185784551168 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_04_int8/checkpoints/9/items/array_metadatas/process_4 I0420 06:13:15.516994 139209260123904 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.557 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.05916357040405 s) (per-host) I0420 06:13:15.517115 139209260123904 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.021070s. I0420 06:13:24.296513 139209260123904 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 50.800446s. I0420 06:13:24.296830 139209282246400 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 06:13:24.296957 139209282246400 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0420 06:13:24.297005 139209282246400 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0420 06:13:24.298881 139209282246400 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0420 06:13:24.299046 139341872637760 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0420 06:13:24.299190 139341872637760 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 06:13:24.300184 139341872637760 metric_logger.py:196] completed step: 9, seconds: 0.229, TFLOP/s/device: 59.290, Tokens/s/device: 8936.909, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3552.333 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Mon Apr 20 06:13:36 UTC 2026 EXIT_CODE=0
574ad3fb9
· feat_nnx_trainstate_and_training_loop_20260420_081533
· full log
XPK Start: Mon Apr 20 08:22:04 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-20 08:22:28.015320: 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) I0420 08:22:28.194730 139225435412288 max_utils.py:273] Attempting to initialize the jax distributed system... I0420 08:22:37.235675 139225435412288 distributed.py:149] Starting JAX distributed service on [::]:8482 I0420 08:22:37.238139 139225435412288 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-4dfws-slice-job-0-0.mt-04-int8-4dfws:8482 I0420 08:22:38.689145 139225435412288 max_utils.py:284] Jax distributed system initialized! I0420 08:22:44.898000 139225435412288 max_utils.py:800] System Information: Jax Version: 0.9.2 I0420 08:22:44.898103 139225435412288 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0420 08:22:44.898144 139225435412288 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 I0420 08:22:44.898178 139225435412288 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0420 08:22:45.954192 139225435412288 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 08:22:45.954809 139225435412288 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 08:22:45.955107 139225435412288 checkpointing.py:688] Setting up checkpoint logger... I0420 08:22:45.955161 139225435412288 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0420 08:22:45.955205 139225435412288 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0420 08:22:45.955558 139225435412288 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 0x7e9f3ac4c6b0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0420 08:22:48.507765 139225435412288 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0420 08:22:48.507995 139225435412288 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e9e2e8d1f70>}, handler_registry=None I0420 08:22:48.508231 139225435412288 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e9e2e8d1f70>` 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`. I0420 08:22:48.508279 139225435412288 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e9e2e91faa0>` 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`. I0420 08:22:48.508314 139225435412288 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 0x7e9e2e8d1f70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e9e2e8d1f70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e9e2e91faa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e9e2e91faa0>}). I0420 08:22:48.508647 139225435412288 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0420 08:22:48.508719 139225435412288 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e8ae43b3740> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0420 08:22:49.631399 139225435412288 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints I0420 08:22:50.064033 139225435412288 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e9e2e8d0f80> I0420 08:22:50.064205 139225435412288 checkpointing.py:302] Checkpoint manager created! I0420 08:22:51.202841 139225435412288 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0420 08:22:51.202944 139225435412288 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 08:22:51.591292 139225435412288 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 08:22:51.591394 139225435412288 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 08:22:51.607878 139225435412288 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 08:22:51.607939 139225435412288 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 08:22:51.674952 139225435412288 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 08:22:51.675040 139225435412288 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 08:22:51.697617 139225435412288 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 08:22:51.697723 139225435412288 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 08:22:51.714207 139225435412288 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 08:22:51.714266 139225435412288 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 08:22:51.739145 139225435412288 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0420 08:22:51.739216 139225435412288 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 08:22:51.806309 139225435412288 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0420 08:22:51.806404 139225435412288 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0420 08:22:52.325625 139225435412288 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0420 08:22:52.325734 139225435412288 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0420 08:22:54.369509 139225435412288 maxtext_utils.py:1821] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0420 08:22:54.369639 139225435412288 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) I0420 08:22:54.369715 139225435412288 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) I0420 08:22:54.369775 139225435412288 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') I0420 08:22:54.369829 139225435412288 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) I0420 08:22:54.369867 139225435412288 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) I0420 08:22:54.369920 139225435412288 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) I0420 08:22:54.369974 139225435412288 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') I0420 08:22:54.370015 139225435412288 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) I0420 08:22:54.370051 139225435412288 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) I0420 08:22:54.370098 139225435412288 maxtext_utils.py:1821] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0420 08:22:54.370145 139225435412288 maxtext_utils.py:1821] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0420 08:22:56.225814 139225435412288 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0420 08:22:56.225910 139225435412288 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0420 08:22:56.241329 139225435412288 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0420 08:22:56.241399 139225435412288 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0420 08:23:09.825434 139225435412288 max_utils.py:791] Total memory size: 1.8 GB, Output size: 0.4 GB, Temp size: 1.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0420 08:23:09.826205 139225435412288 metric_logger.py:301] number parameters: 1.104 billion I0420 08:23:25.082369 139225435412288 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0420 08:23:25.319796 139225435412288 checkpointing.py:798] Waited 0.2374269962310791 seconds for step 0 to finish before starting checkpointing. I0420 08:23:25.322360 139225435412288 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 08:23:25.324577 139225435412288 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0420 08:23:25.326042 139225435412288 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints/0. I0420 08:23:25.663937 139225435412288 signaling_client.py:364] Using JaxDistributedSignalingClient I0420 08:23:25.664861 139225435412288 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 08:23:25.664917 139225435412288 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 I0420 08:23:25.937600 139225435412288 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273740s I0420 08:23:25.937765 139225435412288 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.522 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27936601638793945 s) (per-host) I0420 08:23:25.937822 139225435412288 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279431s (batch_requests_ready=0.002279s, total_serialization_initiated=0.277080s, others=0.000073s) I0420 08:23:25.937913 139225435412288 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283675s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.283657) I0420 08:23:25.938685 139225435412288 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints/0. I0420 08:23:25.938951 139093092357888 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 08:43:25.938927 I0420 08:23:25.950163 139225435412288 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0420 08:23:25.950433 139091479881472 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0420 08:23:25.950557 139225435412288 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776673405.3223276, 'wait_for_prev_duration_secs': 7.510185241699219e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776673405.3246155, 'checkpointer_blocking_duration_secs': 0.6144421100616455, 'get_old_steps_start_time': 1776673405.9390829, 'get_old_steps_duration_secs': 3.2901763916015625e-05, 'checkpoint_manager_blocking_start_time': 1776673405.320316, 'checkpoint_manager_blocking_duration_secs': 0.6302011013031006} I0420 08:23:25.950664 139225435412288 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0420 08:23:25.950714 139225435412288 max_utils.py:750] Memstats: After params initialized: I0420 08:23:25.950762 139225435412288 max_utils.py:756] Using (GB) 0.44 / 31.25 (1.408000%) on TPU_16(process=4,(0,4,0,0)) I0420 08:23:25.950796 139225435412288 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0420 08:23:25.950824 139225435412288 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0420 08:23:25.950850 139225435412288 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0420 08:23:26.274942 139225435412288 metric_logger.py:196] completed step: 0, seconds: 15.256, TFLOP/s/device: 0.891, Tokens/s/device: 134.242, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0420 08:23:26.516820 139225435412288 metric_logger.py:196] completed step: 1, seconds: 1.191, TFLOP/s/device: 11.408, Tokens/s/device: 1719.498, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0420 08:23:26.926359 139225435412288 metric_logger.py:196] completed step: 2, seconds: 0.013, TFLOP/s/device: 1053.021, Tokens/s/device: 158722.778, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820 I0420 08:23:27.157041 139225435412288 metric_logger.py:196] completed step: 3, seconds: 0.410, TFLOP/s/device: 33.151, Tokens/s/device: 4996.816, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199 I0420 08:23:27.618108 139225435412288 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.848, Tokens/s/device: 8719.532, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531 I0420 08:23:27.624417 139225435412288 metric_logger.py:196] completed step: 5, seconds: 0.231, TFLOP/s/device: 58.892, Tokens/s/device: 8876.907, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941 I0420 08:23:29.406123 2965 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0420 08:23:31.793913 139092046116608 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints/0/items/array_metadatas/process_4 I0420 08:23:56.835649 139225435412288 metric_logger.py:196] completed step: 6, seconds: 0.462, TFLOP/s/device: 29.428, Tokens/s/device: 4435.771, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468 I0420 08:23:57.065282 139225435412288 metric_logger.py:196] completed step: 7, seconds: 28.983, TFLOP/s/device: 0.469, Tokens/s/device: 70.662, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779 I0420 08:23:57.294891 139225435412288 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.108, Tokens/s/device: 8758.687, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453 I0420 08:23:57.523627 139225435412288 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0420 08:23:57.524324 139225435412288 checkpointing.py:798] Waited 0.0007162094116210938 seconds for step 9 to finish before starting checkpointing. I0420 08:23:57.526415 139225435412288 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 08:23:59.299492 139093092357888 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 46.954 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 33.64105415344238 s) (per-host) I0420 08:23:59.299616 139093092357888 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.360588s. I0420 08:24:16.870148 139093092357888 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 50.931103s. I0420 08:24:16.870403 139091479881472 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 08:24:16.870462 139091479881472 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0420 08:24:16.870510 139091479881472 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0420 08:24:16.873185 139091479881472 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0420 08:24:16.873363 139225435412288 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0420 08:24:16.873504 139225435412288 checkpoint_manager.py:1452] Waiting for previous save to complete took 19.347089 seconds. If this number is high, consider checkpointing less frequently. I0420 08:24:16.875378 139225435412288 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0420 08:24:16.877396 139225435412288 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints/9. I0420 08:24:17.238685 139225435412288 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 08:24:17.238779 139225435412288 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 I0420 08:24:17.270835 139225435412288 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.033225s I0420 08:24:17.270980 139225435412288 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 41.956 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03676629066467285 s) (per-host) I0420 08:24:17.271029 139225435412288 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.036824s (batch_requests_ready=0.001876s, total_serialization_initiated=0.034884s, others=0.000064s) I0420 08:24:17.271118 139225435412288 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.041313s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.041298) I0420 08:24:17.271814 139225435412288 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.40 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints/9. I0420 08:24:17.272129 139091479881472 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 08:44:17.272094 I0420 08:24:17.275795 139225435412288 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0420 08:24:17.276099 139092582987520 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0420 08:24:17.276261 139225435412288 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776673437.526384, 'wait_for_prev_duration_secs': 19.34708857536316, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776673456.8754165, 'checkpointer_blocking_duration_secs': 0.39685654640197754, 'get_old_steps_start_time': 1776673457.272294, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776673437.5245724, 'checkpoint_manager_blocking_duration_secs': 19.75165367126465} I0420 08:24:17.276384 139225435412288 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0420 08:24:17.276431 139225435412288 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 08:24:22.755663 139073540830976 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_04_int8/checkpoints/9/items/array_metadatas/process_4 I0420 08:24:58.879175 139091479881472 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.930 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.6449191570282 s) (per-host) I0420 08:24:58.879302 139091479881472 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.607061s. I0420 08:25:18.398878 139091479881472 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 61.126620s. I0420 08:25:18.399181 139092582987520 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 08:25:18.399313 139092582987520 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0420 08:25:18.399383 139092582987520 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0420 08:25:18.401857 139092582987520 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0420 08:25:18.402037 139225435412288 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0420 08:25:18.402185 139225435412288 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 08:25:18.403097 139225435412288 metric_logger.py:196] completed step: 9, seconds: 0.230, TFLOP/s/device: 59.195, Tokens/s/device: 8922.464, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3552.333 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Mon Apr 20 08:25:30 UTC 2026 EXIT_CODE=0