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 | 8.1750 | 8.1750 | 0 |
| TFLOP/s | 59.244 | 58.965 | -0.279 |
| Tok/s | 8930.0 | 8887.8 | -42.164 |
| Avg s/step | 4.040 | 3.812 | -0.228 |
| 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 00:12:52 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 00:13:17.239422: 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 00:13:17.777427 134309745391424 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 00:13:26.818732 134309745391424 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 00:13:26.821142 134309745391424 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-lwpyu-slice-job-0-0.mt-04-int8-lwpyu:8482 I0418 00:13:27.851605 134309745391424 max_utils.py:284] Jax distributed system initialized! I0418 00:13:34.073078 134309745391424 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 00:13:34.073183 134309745391424 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 00:13:34.073223 134309745391424 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 00:13:34.073257 134309745391424 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 00:13:34.750223 134309745391424 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 00:13:34.750607 134309745391424 checkpointing.py:677] Setting up checkpoint logger... I0418 00:13:34.750664 134309745391424 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 00:13:34.750708 134309745391424 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 00:13:34.751099 134309745391424 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 0x7a26ced1f230>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 00:13:38.529014 134309745391424 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0418 00:13:38.529255 134309745391424 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a125c3bb8c0>}, handler_registry=None I0418 00:13:38.529531 134309745391424 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a125c3bb8c0>` 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 00:13:38.529601 134309745391424 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a125c3c69c0>` 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 00:13:38.529655 134309745391424 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 0x7a125c3bb8c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a125c3bb8c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a125c3c69c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a125c3c69c0>}). I0418 00:13:38.530039 134309745391424 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 00:13:38.530132 134309745391424 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7a125c133ba0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 00:13:39.570201 134309745391424 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints I0418 00:13:40.437277 134309745391424 checkpoint_manager.py:929] [process=0][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_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7a125c3bb0e0> I0418 00:13:40.437443 134309745391424 checkpointing.py:301] Checkpoint manager created! I0418 00:13:41.557769 134309745391424 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 00:13:41.557911 134309745391424 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 00:13:41.945113 134309745391424 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 00:13:41.945204 134309745391424 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 00:13:41.961582 134309745391424 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 00:13:41.961648 134309745391424 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 00:13:42.027798 134309745391424 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 00:13:42.027930 134309745391424 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 00:13:42.044468 134309745391424 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 00:13:42.044528 134309745391424 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 00:13:42.060865 134309745391424 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 00:13:42.060932 134309745391424 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 00:13:42.085526 134309745391424 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 00:13:42.085597 134309745391424 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 00:13:42.152023 134309745391424 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 00:13:42.152106 134309745391424 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 00:13:42.678521 134309745391424 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0418 00:13:42.678635 134309745391424 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 00:13:44.714485 134309745391424 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 00:13:44.714609 134309745391424 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 00:13:44.714661 134309745391424 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 00:13:44.714717 134309745391424 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 00:13:44.714770 134309745391424 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 00:13:44.714809 134309745391424 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 00:13:44.714891 134309745391424 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 00:13:44.714950 134309745391424 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 00:13:44.714993 134309745391424 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 00:13:44.715087 134309745391424 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 00:13:44.715149 134309745391424 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 00:13:44.715214 134309745391424 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 00:13:46.543640 134309745391424 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 00:13:46.543737 134309745391424 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 00:13:46.559165 134309745391424 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 00:13:46.559221 134309745391424 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 00:13:59.931931 134309745391424 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. I0418 00:13:59.940666 134309745391424 metric_logger.py:301] number parameters: 1.104 billion I0418 00:14:15.414241 134309745391424 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0418 00:14:15.652921 134309745391424 checkpointing.py:776] Waited 0.23865485191345215 seconds for step 0 to finish before starting checkpointing. I0418 00:14:15.655769 134309745391424 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 00:14:15.657651 134309745391424 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0 I0418 00:14:15.659153 134309745391424 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0. I0418 00:14:16.398636 134309745391424 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 00:14:16.412874 134309745391424 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 00:14:16.412944 134309745391424 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 00:14:16.695826 134309745391424 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.283609s I0418 00:14:16.696575 134309745391424 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.318 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2900881767272949 s) (per-host) I0418 00:14:16.696650 134309745391424 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.290184s (batch_requests_ready=0.002238s, total_serialization_initiated=0.287283s, others=0.000662s) I0418 00:14:16.696730 134309745391424 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.294085s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.294067) I0418 00:14:16.697533 134309745391424 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.04 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0. I0418 00:14:16.697891 134177085171456 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 00:34:16.697854 I0418 00:14:16.700187 134309745391424 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 00:14:16.700477 134174950287104 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0418 00:14:16.700631 134309745391424 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776471255.655751, 'wait_for_prev_duration_secs': 6.151199340820312e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776471255.6576898, 'checkpointer_blocking_duration_secs': 1.0403549671173096, 'get_old_steps_start_time': 1776471256.6980693, 'get_old_steps_duration_secs': 2.7418136596679688e-05, 'checkpoint_manager_blocking_start_time': 1776471255.6534493, 'checkpoint_manager_blocking_duration_secs': 1.047145128250122} I0418 00:14:16.700741 134309745391424 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0418 00:14:16.700790 134309745391424 max_utils.py:750] Memstats: After params initialized: I0418 00:14:16.700850 134309745391424 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0)) I0418 00:14:16.700891 134309745391424 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0)) I0418 00:14:16.700921 134309745391424 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0)) I0418 00:14:16.700946 134309745391424 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0)) I0418 00:14:16.933204 134177076778752 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0 I0418 00:14:17.046174 134309745391424 metric_logger.py:196] completed step: 0, seconds: 15.151, TFLOP/s/device: 0.897, Tokens/s/device: 135.170, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0418 00:14:17.049201 134309745391424 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/tensorboard/' I0418 00:14:17.431531 134309745391424 metric_logger.py:196] completed step: 1, seconds: 1.630, TFLOP/s/device: 8.336, Tokens/s/device: 1256.435, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0418 00:14:17.659081 134309745391424 metric_logger.py:196] completed step: 2, seconds: 0.386, TFLOP/s/device: 35.189, Tokens/s/device: 5304.037, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820 I0418 00:14:17.888492 134309745391424 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1432.033, Tokens/s/device: 215851.602, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199 I0418 00:14:20.624186 134174405023488 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776471260159670302, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776471260159670302, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0/_CHECKPOINT_METADATA I0418 00:14:21.280026 134176581871360 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0/items I0418 00:14:22.904384 2972 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 00:14:25.485218 134184106456832 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0/items/array_metadatas/process_0 I0418 00:14:51.303341 134309745391424 metric_logger.py:196] completed step: 4, seconds: 0.229, TFLOP/s/device: 59.439, Tokens/s/device: 8959.350, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531 I0418 00:14:51.312146 134309745391424 metric_logger.py:196] completed step: 5, seconds: 0.229, TFLOP/s/device: 59.254, Tokens/s/device: 8931.453, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941 I0418 00:14:51.539773 134309745391424 metric_logger.py:196] completed step: 6, seconds: 33.413, TFLOP/s/device: 0.407, Tokens/s/device: 61.293, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468 I0418 00:14:51.769160 134309745391424 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 2028.535, Tokens/s/device: 305762.914, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779 I0418 00:14:51.998356 134309745391424 metric_logger.py:196] completed step: 8, seconds: 0.228, TFLOP/s/device: 59.469, Tokens/s/device: 8963.860, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453 I0418 00:14:52.226996 134309745391424 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0418 00:14:52.227606 134309745391424 checkpointing.py:776] Waited 0.0006258487701416016 seconds for step 9 to finish before starting checkpointing. I0418 00:14:52.229694 134309745391424 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 00:14:56.919316 134175999584000 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 34.020766s (commit=31.564913s, array_metadata_write=2.455853s) I0418 00:14:56.920439 134177085171456 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 38.989 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.513925075531006 s) (per-host) I0418 00:14:56.920565 134177085171456 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.222554s. I0418 00:14:58.518923 134177085171456 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776471260159670302, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0/_CHECKPOINT_METADATA I0418 00:14:59.243038 134174405023488 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776471260159670302, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0/_CHECKPOINT_METADATA I0418 00:14:59.617274 134177085171456 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000258s. I0418 00:14:59.889112 134177085171456 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0418 00:15:00.800883 134177085171456 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.161507s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0/items I0418 00:15:00.802592 134177085171456 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0/items I0418 00:15:02.209267 134177085171456 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0 I0418 00:15:04.427681 134177085171456 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0`. I0418 00:15:04.428554 134177085171456 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 48.77 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/0 I0418 00:15:04.430071 134177085171456 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 47.732056s. I0418 00:15:04.430275 134174950287104 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 00:15:04.430389 134174950287104 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0418 00:15:04.430452 134174950287104 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 00:15:04.432012 134174950287104 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 00:15:04.432201 134309745391424 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0418 00:15:04.432340 134309745391424 checkpoint_manager.py:1452] Waiting for previous save to complete took 12.202648 seconds. If this number is high, consider checkpointing less frequently. I0418 00:15:04.434080 134309745391424 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9 I0418 00:15:04.436710 134309745391424 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9. I0418 00:15:05.151497 134309745391424 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 00:15:05.151586 134309745391424 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 00:15:05.177691 134309745391424 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026853s I0418 00:15:05.178568 134309745391424 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 49.249 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03132224082946777 s) (per-host) I0418 00:15:05.178643 134309745391424 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031415s (batch_requests_ready=0.001797s, total_serialization_initiated=0.028842s, others=0.000776s) I0418 00:15:05.178724 134309745391424 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035549s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.035533) I0418 00:15:05.179518 134309745391424 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9. I0418 00:15:05.179792 134174950287104 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 00:35:05.179770 I0418 00:15:05.189641 134309745391424 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 00:15:05.189893 134176028948224 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0418 00:15:05.190018 134309745391424 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776471292.2296576, 'wait_for_prev_duration_secs': 12.202647924423218, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776471304.4341178, 'checkpointer_blocking_duration_secs': 0.7457630634307861, 'get_old_steps_start_time': 1776471305.1799047, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776471292.2278934, 'checkpoint_manager_blocking_duration_secs': 12.962089538574219} I0418 00:15:05.190130 134309745391424 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0418 00:15:05.190176 134309745391424 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 00:15:05.385396 134177085171456 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9 I0418 00:15:07.089907 134176548300544 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9/items I0418 00:15:12.882918 134176565085952 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9/items/array_metadatas/process_0 I0418 00:15:52.972553 134176556693248 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 43.051305s (commit=40.995324s, array_metadata_write=2.055981s) I0418 00:15:52.973641 134174950287104 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 33.028 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 47.826374769210815 s) (per-host) I0418 00:15:52.973752 134174950287104 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 47.793358s. I0418 00:15:55.602020 134174950287104 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000292s. I0418 00:15:55.988516 134174950287104 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0418 00:15:56.931182 134174950287104 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.630550s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9/items I0418 00:15:56.933676 134174950287104 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9/items I0418 00:15:57.457729 134174950287104 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9 I0418 00:15:59.248335 134174950287104 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9`. I0418 00:15:59.249188 134174950287104 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 54.82 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_235819/linen_xpk_main_20260417_235819_04_int8/checkpoints/9 I0418 00:15:59.250672 134174950287104 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 54.070275s. I0418 00:15:59.250885 134176028948224 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 00:15:59.250996 134176028948224 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0418 00:15:59.251056 134176028948224 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 00:15:59.255549 134176028948224 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 00:15:59.255694 134309745391424 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0418 00:15:59.255793 134309745391424 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 00:15:59.256787 134309745391424 metric_logger.py:196] completed step: 9, seconds: 0.229, TFLOP/s/device: 59.244, Tokens/s/device: 8929.973, 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: Sat Apr 18 00:16:12 UTC 2026 EXIT_CODE=0
190761531
· feat_nnx_trainstate_and_training_loop_20260418_060141
· full log
XPK Start: Sat Apr 18 06:15:12 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 06:15:37.312469: 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 06:15:37.493368 139655138232128 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 06:15:46.535242 139655138232128 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 06:15:46.537690 139655138232128 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-djpul-slice-job-0-0.mt-04-int8-djpul:8482 I0418 06:15:48.134695 139655138232128 max_utils.py:284] Jax distributed system initialized! I0418 06:15:53.381285 139655138232128 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 06:15:53.381388 139655138232128 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 06:15:53.381428 139655138232128 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 06:15:53.381464 139655138232128 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 06:15:54.433964 139655138232128 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 06:15:54.434542 139655138232128 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 06:15:54.434857 139655138232128 checkpointing.py:688] Setting up checkpoint logger... I0418 06:15:54.434914 139655138232128 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 06:15:54.434956 139655138232128 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 06:15:54.435278 139655138232128 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 0x7f03475481d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 06:15:57.588247 139655138232128 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0418 06:15:57.588593 139655138232128 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f023ac95700>}, handler_registry=None I0418 06:15:57.588845 139655138232128 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f023ac95700>` 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 06:15:57.588894 139655138232128 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f023add7200>` 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 06:15:57.588929 139655138232128 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 0x7f023ac95700>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f023ac95700>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f023add7200>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f023add7200>}). I0418 06:15:57.589246 139655138232128 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 06:15:57.589315 139655138232128 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7eeec84ef740> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 06:16:00.001181 139655138232128 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_04_int8/checkpoints I0418 06:16:00.912124 139655138232128 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f023ac94860> I0418 06:16:00.912299 139655138232128 checkpointing.py:302] Checkpoint manager created! I0418 06:16:02.081090 139655138232128 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 06:16:02.081190 139655138232128 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 06:16:02.467809 139655138232128 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 06:16:02.467912 139655138232128 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 06:16:02.484341 139655138232128 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 06:16:02.484398 139655138232128 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 06:16:02.550425 139655138232128 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 06:16:02.550511 139655138232128 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 06:16:02.567021 139655138232128 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 06:16:02.567081 139655138232128 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 06:16:02.583773 139655138232128 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 06:16:02.583853 139655138232128 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 06:16:02.608445 139655138232128 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 06:16:02.608515 139655138232128 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 06:16:02.674335 139655138232128 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 06:16:02.674414 139655138232128 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 06:16:03.193494 139655138232128 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0418 06:16:03.193607 139655138232128 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 06:16:05.237080 139655138232128 maxtext_utils.py:1821] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 06:16:05.237203 139655138232128 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 06:16:05.237256 139655138232128 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 06:16:05.237313 139655138232128 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 06:16:05.237365 139655138232128 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 06:16:05.237403 139655138232128 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 06:16:05.237455 139655138232128 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 06:16:05.237508 139655138232128 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 06:16:05.237548 139655138232128 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 06:16:05.237584 139655138232128 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 06:16:05.237631 139655138232128 maxtext_utils.py:1821] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 06:16:05.237676 139655138232128 maxtext_utils.py:1821] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 06:16:07.088831 139655138232128 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 06:16:07.088926 139655138232128 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 06:16:07.104424 139655138232128 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 06:16:07.104483 139655138232128 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 06:16:20.281158 139655138232128 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. I0418 06:16:20.281982 139655138232128 metric_logger.py:301] number parameters: 1.104 billion I0418 06:16:35.350025 139655138232128 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0418 06:16:35.594563 139655138232128 checkpointing.py:798] Waited 0.2445213794708252 seconds for step 0 to finish before starting checkpointing. I0418 06:16:35.597553 139655138232128 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 06:16:35.599435 139655138232128 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0418 06:16:35.600867 139655138232128 event_tracking.py:70] [process=6] [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_04_int8/checkpoints/0. I0418 06:16:36.758458 139655138232128 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 06:16:36.759403 139655138232128 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 06:16:36.759461 139655138232128 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 06:16:37.064249 139655138232128 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.305864s I0418 06:16:37.064414 139655138232128 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 4.956 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.3112499713897705 s) (per-host) I0418 06:16:37.064465 139655138232128 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.311313s (batch_requests_ready=0.002209s, total_serialization_initiated=0.309033s, others=0.000071s) I0418 06:16:37.064554 139655138232128 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.317778s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.317760) I0418 06:16:37.065324 139655138232128 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.47 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_04_int8/checkpoints/0. I0418 06:16:37.065663 139524478850816 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 06:36:37.065624 I0418 06:16:37.067689 139655138232128 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 06:16:37.068005 139523911067392 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0418 06:16:37.068137 139655138232128 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_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776492995.5975342, 'wait_for_prev_duration_secs': 6.437301635742188e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776492995.5994744, 'checkpointer_blocking_duration_secs': 1.4663572311401367, 'get_old_steps_start_time': 1776492997.0658572, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776492995.5951161, 'checkpoint_manager_blocking_duration_secs': 1.472982406616211} I0418 06:16:37.068291 139655138232128 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0418 06:16:37.068342 139655138232128 max_utils.py:750] Memstats: After params initialized: I0418 06:16:37.068391 139655138232128 max_utils.py:756] Using (GB) 0.44 / 31.25 (1.408000%) on TPU_24(process=6,(0,6,0,0)) I0418 06:16:37.068422 139655138232128 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0418 06:16:37.068460 139655138232128 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0418 06:16:37.068485 139655138232128 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0418 06:16:37.389843 139655138232128 metric_logger.py:196] completed step: 0, seconds: 15.068, TFLOP/s/device: 0.902, Tokens/s/device: 135.918, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0418 06:16:37.641008 139655138232128 metric_logger.py:196] completed step: 1, seconds: 2.038, TFLOP/s/device: 6.666, Tokens/s/device: 1004.788, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0418 06:16:38.066019 139655138232128 metric_logger.py:196] completed step: 2, seconds: 0.022, TFLOP/s/device: 618.271, Tokens/s/device: 93192.574, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820 I0418 06:16:38.296772 139655138232128 metric_logger.py:196] completed step: 3, seconds: 0.425, TFLOP/s/device: 31.941, Tokens/s/device: 4814.507, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199 I0418 06:16:38.757309 139655138232128 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.864, Tokens/s/device: 8721.871, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531 I0418 06:16:38.763423 139655138232128 metric_logger.py:196] completed step: 5, seconds: 0.231, TFLOP/s/device: 58.787, Tokens/s/device: 8861.083, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941 I0418 06:16:41.415647 2880 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 06:16:43.877037 139524455331584 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_04_int8/checkpoints/0/items/array_metadatas/process_6 I0418 06:17:09.419642 139655138232128 metric_logger.py:196] completed step: 6, seconds: 0.461, TFLOP/s/device: 29.481, Tokens/s/device: 4443.769, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468 I0418 06:17:09.650327 139655138232128 metric_logger.py:196] completed step: 7, seconds: 30.428, TFLOP/s/device: 0.447, Tokens/s/device: 67.306, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779 I0418 06:17:09.881041 139655138232128 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.037, Tokens/s/device: 8747.912, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453 I0418 06:17:09.906788 139524478850816 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 47.645 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 33.15358805656433 s) (per-host) I0418 06:17:09.906922 139524478850816 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.841147s. I0418 06:17:10.109930 139655138232128 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0418 06:17:10.110698 139655138232128 checkpointing.py:798] Waited 0.0007851123809814453 seconds for step 9 to finish before starting checkpointing. I0418 06:17:10.112875 139655138232128 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 06:17:21.709805 139524478850816 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 44.644014s. I0418 06:17:21.710108 139523911067392 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 06:17:21.710241 139523911067392 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0418 06:17:21.710292 139523911067392 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 06:17:21.712769 139523911067392 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 06:17:21.712972 139655138232128 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0418 06:17:21.713109 139655138232128 checkpoint_manager.py:1452] Waiting for previous save to complete took 11.600235 seconds. If this number is high, consider checkpointing less frequently. I0418 06:17:21.715085 139655138232128 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0418 06:17:21.717232 139655138232128 event_tracking.py:70] [process=6] [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_04_int8/checkpoints/9. I0418 06:17:22.431860 139655138232128 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 06:17:22.431952 139655138232128 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 06:17:22.470129 139655138232128 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039418s I0418 06:17:22.470266 139655138232128 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 36.026 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04281806945800781 s) (per-host) I0418 06:17:22.470311 139655138232128 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.042873s (batch_requests_ready=0.001793s, total_serialization_initiated=0.041019s, others=0.000061s) I0418 06:17:22.470393 139655138232128 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.046895s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.046880) I0418 06:17:22.471057 139655138232128 event_tracking.py:125] [process=6] [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_04_int8/checkpoints/9. I0418 06:17:22.471423 139523911067392 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 06:37:22.471386 I0418 06:17:22.473427 139655138232128 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 06:17:22.473721 139499321386752 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0418 06:17:22.473894 139655138232128 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_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776493030.1128435, 'wait_for_prev_duration_secs': 11.600234985351562, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776493041.7151227, 'checkpointer_blocking_duration_secs': 0.7564740180969238, 'get_old_steps_start_time': 1776493042.4716215, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776493030.1109943, 'checkpoint_manager_blocking_duration_secs': 12.362865209579468} I0418 06:17:22.474003 139655138232128 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0418 06:17:22.474045 139655138232128 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 06:17:31.590474 139524455331584 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141/linen_xpk_feat_nnx_trainstate_and_training_loop_20260418_060141_04_int8/checkpoints/9/items/array_metadatas/process_6 I0418 06:18:06.746337 139523911067392 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 35.642 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.31883454322815 s) (per-host) I0418 06:18:06.746500 139523911067392 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.274958s. I0418 06:18:17.245983 139523911067392 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 54.774429s. I0418 06:18:17.246256 139499321386752 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 06:18:17.246378 139499321386752 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0418 06:18:17.246429 139499321386752 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 06:18:17.248119 139499321386752 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 06:18:17.248301 139655138232128 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0418 06:18:17.248456 139655138232128 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 06:18:17.249387 139655138232128 metric_logger.py:196] completed step: 9, seconds: 0.230, TFLOP/s/device: 58.965, Tokens/s/device: 8887.809, 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: Sat Apr 18 06:18:28 UTC 2026 EXIT_CODE=0