main (574ad3fb9) vs feat/nnx-linen-converter-and-sharding-tools (574ad3fb9)| Metric | main 574ad3fb9 | feat/nnx-linen-converter-and-sharding-tools 574ad3fb9 | Diff (feat/nnx-linen-converter-and-sharding-tools − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1750 | 8.1750 | 0 |
| TFLOP/s | 59.228 | 59.183 | -0.045 |
| Tok/s | 8927.5 | 8920.8 | -6.728 |
| Avg s/step | 3.756 | 3.770 | +0.014 |
| Memory % | 1.38 | 1.38 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Sat Apr 18 18:12:38 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-18 18:13:02.627725: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0418 18:13:03.159810 132270230222656 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 18:13:12.201794 132270230222656 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 18:13:12.204128 132270230222656 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-y1cga-slice-job-0-0.mt-04-int8-y1cga:8482 I0418 18:13:13.681090 132270230222656 max_utils.py:284] Jax distributed system initialized! I0418 18:13:18.775402 132270230222656 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 18:13:18.775501 132270230222656 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 18:13:18.775539 132270230222656 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0418 18:13:18.775572 132270230222656 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 18:13:19.465929 132270230222656 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 18:13:19.466346 132270230222656 checkpointing.py:677] Setting up checkpoint logger... I0418 18:13:19.466416 132270230222656 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 18:13:19.466462 132270230222656 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 18:13:19.466780 132270230222656 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 0x784bef11f650>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 18:13:22.341234 132270230222656 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0418 18:13:22.341469 132270230222656 checkpoint_manager.py:708] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7838880ff860>}, handler_registry=None I0418 18:13:22.341770 132270230222656 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7838880ff860>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0418 18:13:22.341827 132270230222656 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x783888101610>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0418 18:13:22.341864 132270230222656 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 0x7838880ff860>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7838880ff860>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x783888101610>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x783888101610>}). I0418 18:13:22.342198 132270230222656 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 18:13:22.342268 132270230222656 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7838584c37e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 18:13:23.898544 132270230222656 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints I0418 18:13:24.337405 132270230222656 checkpoint_manager.py:929] [process=7][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x783888291d00> I0418 18:13:24.337574 132270230222656 checkpointing.py:301] Checkpoint manager created! I0418 18:13:25.486507 132270230222656 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 18:13:25.486622 132270230222656 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:13:25.873187 132270230222656 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 18:13:25.873278 132270230222656 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:13:25.889510 132270230222656 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 18:13:25.889569 132270230222656 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:13:25.954802 132270230222656 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:13:25.954892 132270230222656 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:13:25.971244 132270230222656 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:13:25.971302 132270230222656 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:13:25.987523 132270230222656 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:13:25.987581 132270230222656 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:13:26.011954 132270230222656 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 18:13:26.012025 132270230222656 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:13:26.077721 132270230222656 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 18:13:26.077803 132270230222656 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 18:13:26.591408 132270230222656 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0418 18:13:26.591516 132270230222656 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 18:13:28.618637 132270230222656 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 18:13:28.618762 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 18:13:28.618814 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 18:13:28.618869 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0418 18:13:28.618925 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 18:13:28.618963 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 18:13:28.619014 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0418 18:13:28.619068 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0418 18:13:28.619126 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0418 18:13:28.619169 132270230222656 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0418 18:13:28.619215 132270230222656 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 18:13:28.619262 132270230222656 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 18:13:30.461583 132270230222656 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 18:13:30.461678 132270230222656 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 18:13:30.476938 132270230222656 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 18:13:30.476998 132270230222656 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 18:13:43.841150 132270230222656 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 18:13:43.841910 132270230222656 metric_logger.py:301] number parameters: 1.104 billion I0418 18:13:59.014615 132270230222656 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0418 18:13:59.255299 132270230222656 checkpointing.py:776] Waited 0.24066615104675293 seconds for step 0 to finish before starting checkpointing. I0418 18:13:59.257932 132270230222656 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 18:13:59.259819 132270230222656 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0418 18:13:59.261306 132270230222656 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints/0. I0418 18:14:00.451537 132270230222656 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 18:14:00.452484 132270230222656 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 18:14:00.452566 132270230222656 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 18:14:00.735319 132270230222656 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.283854s I0418 18:14:00.735493 132270230222656 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 5.332 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28933119773864746 s) (per-host) I0418 18:14:00.735543 132270230222656 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.289391s (batch_requests_ready=0.002228s, total_serialization_initiated=0.287095s, others=0.000068s) I0418 18:14:00.735634 132270230222656 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.293585s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.293568) I0418 18:14:00.736411 132270230222656 event_tracking.py:125] [process=7] [async] Finished blocking save in 1.48 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints/0. I0418 18:14:00.736680 132140876412672 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 18:34:00.736657 I0418 18:14:00.742790 132270230222656 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:14:00.743037 132138179479296 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:14:00.743153 132270230222656 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776536039.2579155, 'wait_for_prev_duration_secs': 5.91278076171875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776536039.2598581, 'checkpointer_blocking_duration_secs': 1.476926326751709, 'get_old_steps_start_time': 1776536040.736807, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776536039.2558239, 'checkpoint_manager_blocking_duration_secs': 1.487290859222412} I0418 18:14:00.743311 132270230222656 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0418 18:14:00.743361 132270230222656 max_utils.py:750] Memstats: After params initialized: I0418 18:14:00.743410 132270230222656 max_utils.py:756] Using (GB) 0.44 / 31.25 (1.408000%) on TPU_26(process=7,(2,6,0,0)) I0418 18:14:00.743442 132270230222656 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0)) I0418 18:14:00.743469 132270230222656 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0)) I0418 18:14:00.743493 132270230222656 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0)) I0418 18:14:01.065436 132270230222656 metric_logger.py:196] completed step: 0, seconds: 15.173, TFLOP/s/device: 0.896, Tokens/s/device: 134.980, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0418 18:14:01.325167 132270230222656 metric_logger.py:196] completed step: 1, seconds: 2.049, TFLOP/s/device: 6.631, Tokens/s/device: 999.423, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0418 18:14:01.746206 132270230222656 metric_logger.py:196] completed step: 2, seconds: 0.031, TFLOP/s/device: 435.331, Tokens/s/device: 65617.891, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820 I0418 18:14:01.975643 132270230222656 metric_logger.py:196] completed step: 3, seconds: 0.421, TFLOP/s/device: 32.271, Tokens/s/device: 4864.250, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199 I0418 18:14:02.434826 132270230222656 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.803, Tokens/s/device: 8712.743, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531 I0418 18:14:02.440926 132270230222656 metric_logger.py:196] completed step: 5, seconds: 0.229, TFLOP/s/device: 59.281, Tokens/s/device: 8935.545, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941 I0418 18:14:04.155244 2878 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 18:14:07.005270 132138733135616 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints/0/items/array_metadatas/process_7 I0418 18:14:32.582834 132270230222656 metric_logger.py:196] completed step: 6, seconds: 0.460, TFLOP/s/device: 29.560, Tokens/s/device: 4455.554, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468 I0418 18:14:32.812278 132270230222656 metric_logger.py:196] completed step: 7, seconds: 29.914, TFLOP/s/device: 0.454, Tokens/s/device: 68.463, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779 I0418 18:14:33.042235 132270230222656 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.097, Tokens/s/device: 8757.039, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453 I0418 18:14:33.271201 132270230222656 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0418 18:14:33.271802 132270230222656 checkpointing.py:776] Waited 0.0006241798400878906 seconds for step 9 to finish before starting checkpointing. I0418 18:14:33.274027 132270230222656 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:14:34.176393 132140876412672 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 46.830 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 33.73019456863403 s) (per-host) I0418 18:14:34.176515 132140876412672 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.439760s. I0418 18:14:43.286806 132140876412672 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 42.550034s. I0418 18:14:43.287055 132138179479296 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:14:43.287137 132138179479296 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:14:43.287188 132138179479296 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:14:43.289611 132138179479296 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 18:14:43.289793 132270230222656 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0418 18:14:43.289946 132270230222656 checkpoint_manager.py:1452] Waiting for previous save to complete took 10.015918 seconds. If this number is high, consider checkpointing less frequently. I0418 18:14:43.291633 132270230222656 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 9 I0418 18:14:43.293642 132270230222656 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints/9. I0418 18:14:44.429162 132270230222656 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 18:14:44.429256 132270230222656 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 18:14:44.464814 132270230222656 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037524s I0418 18:14:44.464951 132270230222656 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 37.676 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04094266891479492 s) (per-host) I0418 18:14:44.464998 132270230222656 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040997s (batch_requests_ready=0.001766s, total_serialization_initiated=0.039171s, others=0.000060s) I0418 18:14:44.465120 132270230222656 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.045118s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.045104) I0418 18:14:44.465821 132270230222656 event_tracking.py:125] [process=7] [async] Finished blocking save in 1.17 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints/9. I0418 18:14:44.466175 132138179479296 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 18:34:44.466136 I0418 18:14:44.468362 132270230222656 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:14:44.468656 132138733135616 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:14:44.468817 132270230222656 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776536073.273995, 'wait_for_prev_duration_secs': 10.015918493270874, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776536083.2916718, 'checkpointer_blocking_duration_secs': 1.174652099609375, 'get_old_steps_start_time': 1776536084.466348, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776536073.272038, 'checkpoint_manager_blocking_duration_secs': 11.196742534637451} I0418 18:14:44.468935 132270230222656 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0418 18:14:44.468981 132270230222656 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:14:50.672130 132114515191552 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_04_int8/checkpoints/9/items/array_metadatas/process_7 I0418 18:15:25.734281 132138179479296 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 38.237 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.31024765968323 s) (per-host) I0418 18:15:25.734389 132138179479296 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.268100s. I0418 18:15:34.576668 132138179479296 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 50.110364s. I0418 18:15:34.577056 132138733135616 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:15:34.577201 132138733135616 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:15:34.577249 132138733135616 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:15:34.578672 132138733135616 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 18:15:34.578869 132270230222656 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0418 18:15:34.579021 132270230222656 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 18:15:34.580116 132270230222656 metric_logger.py:196] completed step: 9, seconds: 0.229, TFLOP/s/device: 59.228, Tokens/s/device: 8927.482, 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 18:15:43 UTC 2026 EXIT_CODE=0
574ad3fb9
· feat_nnx_linen_converter_and_sharding_tools_20260418_204152
· full log
XPK Start: Sat Apr 18 20:50:44 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 20:51:07.840401: 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 20:51:08.369836 135824736319296 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 20:51:17.412226 135824736319296 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 20:51:17.414432 135824736319296 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-y9moy-slice-job-0-0.mt-04-int8-y9moy:8482 I0418 20:51:19.277647 135824736319296 max_utils.py:284] Jax distributed system initialized! I0418 20:51:25.042555 135824736319296 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 20:51:25.042655 135824736319296 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 20:51:25.042696 135824736319296 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 20:51:25.042729 135824736319296 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 20:51:25.728288 135824736319296 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 20:51:25.728878 135824736319296 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 20:51:25.729176 135824736319296 checkpointing.py:688] Setting up checkpoint logger... I0418 20:51:25.729228 135824736319296 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 20:51:25.729271 135824736319296 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 20:51:25.729590 135824736319296 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 0x7b878ab1c5f0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 20:51:29.106481 135824736319296 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0418 20:51:29.106654 135824736319296 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 0x7b73b82f2480>}, handler_registry=None I0418 20:51:29.106899 135824736319296 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b73b82f2480>` 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 20:51:29.106948 135824736319296 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b73b82f8d10>` 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 20:51:29.106984 135824736319296 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 0x7b73b82f2480>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b73b82f2480>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b73b82f8d10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b73b82f8d10>}). I0418 20:51:29.107389 135824736319296 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 20:51:29.107485 135824736319296 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b72f467b4c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 20:51:30.433160 135824736319296 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints I0418 20:51:31.288527 135824736319296 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_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b73b8451be0> I0418 20:51:31.288690 135824736319296 checkpointing.py:302] Checkpoint manager created! I0418 20:51:32.435352 135824736319296 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 20:51:32.435455 135824736319296 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 20:51:32.821596 135824736319296 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 20:51:32.821686 135824736319296 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 20:51:32.838040 135824736319296 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 20:51:32.838102 135824736319296 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 20:51:32.904039 135824736319296 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 20:51:32.904124 135824736319296 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 20:51:32.920246 135824736319296 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 20:51:32.920307 135824736319296 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 20:51:32.936378 135824736319296 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 20:51:32.936436 135824736319296 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 20:51:32.960880 135824736319296 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 20:51:32.960945 135824736319296 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 20:51:33.026275 135824736319296 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 20:51:33.026361 135824736319296 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 20:51:33.535906 135824736319296 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0418 20:51:33.536009 135824736319296 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 20:51:35.567472 135824736319296 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 20:51:35.567600 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 20:51:35.567654 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 20:51:35.567710 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0418 20:51:35.567761 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 20:51:35.567798 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 20:51:35.567868 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0418 20:51:35.567930 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0418 20:51:35.567972 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0418 20:51:35.568011 135824736319296 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0418 20:51:35.568058 135824736319296 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 20:51:35.568119 135824736319296 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 20:51:37.399652 135824736319296 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 20:51:37.399747 135824736319296 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 20:51:37.415183 135824736319296 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 20:51:37.415245 135824736319296 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 20:51:50.760930 135824736319296 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 20:51:50.768545 135824736319296 metric_logger.py:301] number parameters: 1.104 billion I0418 20:52:06.498206 135824736319296 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0418 20:52:06.737752 135824736319296 checkpointing.py:798] Waited 0.23952746391296387 seconds for step 0 to finish before starting checkpointing. I0418 20:52:06.744849 135824736319296 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 20:52:06.746742 135824736319296 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0 I0418 20:52:06.748710 135824736319296 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0. I0418 20:52:07.465300 135824736319296 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 20:52:07.480181 135824736319296 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 20:52:07.480250 135824736319296 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 20:52:07.754629 135824736319296 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.275385s I0418 20:52:07.755462 135824736319296 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.478 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28157758712768555 s) (per-host) I0418 20:52:07.755537 135824736319296 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.281672s (batch_requests_ready=0.002038s, total_serialization_initiated=0.278881s, others=0.000753s) I0418 20:52:07.755620 135824736319296 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.286202s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.286184) I0418 20:52:07.756631 135695733618432 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0 I0418 20:52:07.757307 135824736319296 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.01 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0. I0418 20:52:07.757715 135696217540352 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 21:12:07.757676 I0418 20:52:07.767036 135824736319296 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 20:52:07.767336 135694073739008 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0418 20:52:07.767495 135824736319296 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776545526.74483, 'wait_for_prev_duration_secs': 6.198883056640625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776545526.7467802, 'checkpointer_blocking_duration_secs': 1.011049509048462, 'get_old_steps_start_time': 1776545527.7578542, 'get_old_steps_duration_secs': 2.6941299438476562e-05, 'checkpoint_manager_blocking_start_time': 1776545526.7382803, 'checkpoint_manager_blocking_duration_secs': 1.029175043106079} I0418 20:52:07.767610 135824736319296 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0418 20:52:07.767656 135824736319296 max_utils.py:750] Memstats: After params initialized: I0418 20:52:07.767704 135824736319296 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0)) I0418 20:52:07.767735 135824736319296 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0)) I0418 20:52:07.767762 135824736319296 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0)) I0418 20:52:07.767787 135824736319296 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0)) I0418 20:52:08.123194 135824736319296 metric_logger.py:196] completed step: 0, seconds: 15.383, TFLOP/s/device: 0.883, Tokens/s/device: 133.132, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0418 20:52:08.126251 135824736319296 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/tensorboard/' I0418 20:52:08.539852 135824736319296 metric_logger.py:196] completed step: 1, seconds: 1.623, TFLOP/s/device: 8.371, Tokens/s/device: 1261.779, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0418 20:52:08.767287 135824736319296 metric_logger.py:196] completed step: 2, seconds: 0.417, TFLOP/s/device: 32.556, Tokens/s/device: 4907.211, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820 I0418 20:52:08.996920 135824736319296 metric_logger.py:196] completed step: 3, seconds: 0.010, TFLOP/s/device: 1393.981, Tokens/s/device: 210115.933, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199 I0418 20:52:14.962013 135695716833024 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0/items I0418 20:52:15.116389 135693576967936 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776545534703281083, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776545534703281083, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0/_CHECKPOINT_METADATA I0418 20:52:16.512358 2922 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 20:52:19.108901 135695679895296 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0/items/array_metadatas/process_0 I0418 20:52:39.952073 135824736319296 metric_logger.py:196] completed step: 4, seconds: 0.228, TFLOP/s/device: 59.493, Tokens/s/device: 8967.392, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531 I0418 20:52:39.961169 135824736319296 metric_logger.py:196] completed step: 5, seconds: 0.230, TFLOP/s/device: 59.158, Tokens/s/device: 8916.909, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941 I0418 20:52:40.188757 135824736319296 metric_logger.py:196] completed step: 6, seconds: 30.954, TFLOP/s/device: 0.439, Tokens/s/device: 66.163, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468 I0418 20:52:40.418439 135824736319296 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 1976.597, Tokens/s/device: 297934.245, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779 I0418 20:52:40.648931 135824736319296 metric_logger.py:196] completed step: 8, seconds: 0.229, TFLOP/s/device: 59.408, Tokens/s/device: 8954.650, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453 I0418 20:52:40.878728 135824736319296 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0418 20:52:40.879393 135824736319296 checkpointing.py:798] Waited 0.0006854534149169922 seconds for step 9 to finish before starting checkpointing. I0418 20:52:40.881514 135824736319296 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 20:52:49.833887 135694659876608 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 33.322909s (commit=31.679276s, array_metadata_write=1.643633s) I0418 20:52:49.835059 135696217540352 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 37.289 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.36114478111267 s) (per-host) I0418 20:52:49.835194 135696217540352 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.076851s. I0418 20:52:51.011340 135696217540352 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776545534703281083, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0/_CHECKPOINT_METADATA I0418 20:52:51.738537 135693576967936 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776545534703281083, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0/_CHECKPOINT_METADATA I0418 20:52:52.111455 135696217540352 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000260s. I0418 20:52:52.381599 135696217540352 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0418 20:52:53.317944 135696217540352 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.183360s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0/items I0418 20:52:53.319500 135696217540352 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0/items I0418 20:52:53.916990 135696217540352 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0 I0418 20:52:55.747762 135696217540352 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0`. I0418 20:52:55.748502 135696217540352 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 49.00 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/0 I0418 20:52:55.749829 135696217540352 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 47.991485s. I0418 20:52:55.750027 135694073739008 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 20:52:55.750155 135694073739008 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0418 20:52:55.750244 135694073739008 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 20:52:55.753079 135694073739008 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 20:52:55.753267 135824736319296 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 20:52:55.753406 135824736319296 checkpoint_manager.py:1452] Waiting for previous save to complete took 14.871890 seconds. If this number is high, consider checkpointing less frequently. I0418 20:52:55.755362 135824736319296 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9 I0418 20:52:55.757425 135824736319296 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9. I0418 20:52:56.471309 135824736319296 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 20:52:56.471406 135824736319296 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 20:52:56.497777 135824736319296 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027187s I0418 20:52:56.498651 135824736319296 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 48.652 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03170609474182129 s) (per-host) I0418 20:52:56.498729 135824736319296 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031803s (batch_requests_ready=0.001817s, total_serialization_initiated=0.029201s, others=0.000784s) I0418 20:52:56.498826 135824736319296 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035912s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=0.035897) I0418 20:52:56.499598 135824736319296 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9. I0418 20:52:56.499938 135695154784000 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 21:12:56.499908 I0418 20:52:56.513844 135824736319296 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0418 20:52:56.514112 135694073739008 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0418 20:52:56.514255 135824736319296 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776545560.881482, 'wait_for_prev_duration_secs': 14.871889591217041, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776545575.7554038, 'checkpointer_blocking_duration_secs': 0.7446515560150146, 'get_old_steps_start_time': 1776545576.5000784, 'get_old_steps_duration_secs': 3.24249267578125e-05, 'checkpoint_manager_blocking_start_time': 1776545560.87962, 'checkpoint_manager_blocking_duration_secs': 15.634599685668945} I0418 20:52:56.514389 135824736319296 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0418 20:52:56.514434 135824736319296 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 20:52:56.703112 135696217540352 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9 I0418 20:52:59.302385 135695679895296 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9/items I0418 20:53:02.591929 135695188354816 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9/items/array_metadatas/process_0 I0418 20:53:42.512887 135695179962112 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.655753s (commit=40.470963s, array_metadata_write=1.184790s) I0418 20:53:42.513963 135695154784000 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.304 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 46.04698967933655 s) (per-host) I0418 20:53:42.514068 135695154784000 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 46.014035s. I0418 20:53:44.385084 135695154784000 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0418 20:53:44.439207 135695154784000 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000263s. I0418 20:53:45.459987 135695154784000 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.067649s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9/items I0418 20:53:45.465233 135695154784000 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9/items I0418 20:53:46.010450 135695154784000 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9 I0418 20:53:47.799144 135695154784000 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9`. I0418 20:53:47.799887 135695154784000 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 52.04 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152/linen_xpk_feat_nnx_linen_converter_and_sharding_tools_20260418_204152_04_int8/checkpoints/9 I0418 20:53:47.800979 135695154784000 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 51.300943s. I0418 20:53:47.801190 135694073739008 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 20:53:47.801305 135694073739008 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0418 20:53:47.801368 135694073739008 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 20:53:47.803511 135694073739008 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 20:53:47.805157 135824736319296 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 20:53:47.805308 135824736319296 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 20:53:47.806199 135824736319296 metric_logger.py:196] completed step: 9, seconds: 0.230, TFLOP/s/device: 59.183, Tokens/s/device: 8920.754, 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 20:53:59 UTC 2026 EXIT_CODE=0