main (8a17c3d19) vs feat/nnx-trainstate-and-training-loop (29bbe47ed)| Metric | main 8a17c3d19 | feat/nnx-trainstate-and-training-loop 29bbe47ed | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1790 | 8.1790 | 0 |
| TFLOP/s | 92.270 | 92.010 | -0.26 |
| Tok/s | 13907.9 | 13868.8 | -39.18 |
| Avg s/step | 2.744 | 2.761 | +0.017 |
| 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: Wed Apr 22 07:48:25 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-22 07:48:50.252376: 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) I0422 07:48:50.463694 135546370250560 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 07:48:59.506284 135546370250560 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 07:48:59.508655 135546370250560 distributed.py:172] Connecting to JAX distributed service on mt-10-shardy-true-o1w1q-slice-job-0-0.mt-10-shardy-true-o1w1q:8482 I0422 07:49:00.982299 135546370250560 max_utils.py:284] Jax distributed system initialized! I0422 07:49:06.949256 135546370250560 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 07:49:06.949362 135546370250560 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 07:49:06.949402 135546370250560 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 I0422 07:49:06.949439 135546370250560 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 07:49:07.648813 135546370250560 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 07:49:07.649089 135546370250560 checkpointing.py:677] Setting up checkpoint logger... I0422 07:49:07.649143 135546370250560 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 07:49:07.649193 135546370250560 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 07:49:07.649544 135546370250560 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 0x7b46bb51c0b0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 07:49:10.537778 135546370250560 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 07:49:10.538017 135546370250560 checkpoint_manager.py:708] [process=2][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b45954f7ec0>}, handler_registry=None I0422 07:49:10.538260 135546370250560 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b45954f7ec0>` 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`. I0422 07:49:10.538308 135546370250560 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b33207d1ca0>` 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`. I0422 07:49:10.538345 135546370250560 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 0x7b45954f7ec0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b45954f7ec0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b33207d1ca0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b33207d1ca0>}). I0422 07:49:10.538673 135546370250560 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 07:49:10.538754 135546370250560 async_checkpointer.py:192] [process=2][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b33205d5800> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 07:49:11.663927 135546370250560 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints I0422 07:49:12.097850 135546370250560 checkpoint_manager.py:929] [process=2][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b33207ce180> I0422 07:49:12.098022 135546370250560 checkpointing.py:301] Checkpoint manager created! I0422 07:49:13.032258 135546370250560 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 07:49:13.032375 135546370250560 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:49:13.414421 135546370250560 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:49:13.414514 135546370250560 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:49:13.430889 135546370250560 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:49:13.430951 135546370250560 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:49:13.454884 135546370250560 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:49:13.454950 135546370250560 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:49:13.471373 135546370250560 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:49:13.471440 135546370250560 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:49:13.487758 135546370250560 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:49:13.487822 135546370250560 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:49:13.512492 135546370250560 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 07:49:13.512563 135546370250560 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:49:13.533736 135546370250560 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 07:49:13.533803 135546370250560 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 07:49:13.745870 135546370250560 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 07:49:13.745981 135546370250560 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0422 07:49:15.172584 135546370250560 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 07:49:15.172743 135546370250560 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 07:49:15.172803 135546370250560 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 07:49:15.172862 135546370250560 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 07:49:15.172920 135546370250560 maxtext_utils.py:1668] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 07:49:15.172960 135546370250560 maxtext_utils.py:1668] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 07:49:15.173011 135546370250560 maxtext_utils.py:1668] 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) I0422 07:49:15.173062 135546370250560 maxtext_utils.py:1668] 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') I0422 07:49:15.173102 135546370250560 maxtext_utils.py:1668] 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) I0422 07:49:15.173138 135546370250560 maxtext_utils.py:1668] 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) I0422 07:49:15.173191 135546370250560 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 07:49:15.173240 135546370250560 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 07:49:15.670477 135546370250560 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:49:15.670574 135546370250560 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 07:49:15.686125 135546370250560 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:49:15.686193 135546370250560 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 07:49:26.590001 135546370250560 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 07:49:26.590782 135546370250560 metric_logger.py:301] number parameters: 1.104 billion I0422 07:49:38.320556 135546370250560 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 07:49:38.473835 135546370250560 checkpointing.py:776] Waited 0.15326261520385742 seconds for step 0 to finish before starting checkpointing. I0422 07:49:38.476034 135546370250560 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:49:38.477779 135546370250560 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 0 I0422 07:49:38.479184 135546370250560 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints/0. I0422 07:49:39.219989 135546370250560 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 07:49:39.220960 135546370250560 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:49:39.221021 135546370250560 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 I0422 07:49:39.500040 135546370250560 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.280126s I0422 07:49:39.500221 135546370250560 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 5.399 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2856898307800293 s) (per-host) I0422 07:49:39.500274 135546370250560 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.285754s (batch_requests_ready=0.002271s, total_serialization_initiated=0.283411s, others=0.000072s) I0422 07:49:39.500370 135546370250560 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.289903s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.289885) I0422 07:49:39.501202 135546370250560 event_tracking.py:125] [process=2] [async] Finished blocking save in 1.02 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints/0. I0422 07:49:39.501509 135421518407424 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 08:09:39.501475 I0422 07:49:39.504146 135546370250560 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:49:39.504381 135417721005824 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:49:39.504504 135546370250560 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776844178.476015, 'wait_for_prev_duration_secs': 6.270408630371094e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776844178.477818, 'checkpointer_blocking_duration_secs': 1.0238392353057861, 'get_old_steps_start_time': 1776844179.5016828, 'get_old_steps_duration_secs': 4.506111145019531e-05, 'checkpoint_manager_blocking_start_time': 1776844178.474246, 'checkpoint_manager_blocking_duration_secs': 1.0302197933197021} I0422 07:49:39.504604 135546370250560 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 07:49:39.504652 135546370250560 max_utils.py:750] Memstats: After params initialized: I0422 07:49:39.504699 135546370250560 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_8(process=2,(0,2,0,0)) I0422 07:49:39.504745 135546370250560 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_9(process=2,(1,2,0,0)) I0422 07:49:39.504772 135546370250560 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_12(process=2,(0,3,0,0)) I0422 07:49:39.504797 135546370250560 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_13(process=2,(1,3,0,0)) I0422 07:49:39.817209 135546370250560 metric_logger.py:196] completed step: 0, seconds: 11.730, TFLOP/s/device: 1.158, Tokens/s/device: 174.600, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0422 07:49:39.997887 135546370250560 metric_logger.py:196] completed step: 1, seconds: 1.495, TFLOP/s/device: 9.088, Tokens/s/device: 1369.798, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0422 07:49:40.423774 135546370250560 metric_logger.py:196] completed step: 2, seconds: 0.034, TFLOP/s/device: 404.355, Tokens/s/device: 60948.753, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0422 07:49:40.571357 135546370250560 metric_logger.py:196] completed step: 3, seconds: 0.426, TFLOP/s/device: 31.892, Tokens/s/device: 4807.083, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0422 07:49:40.866319 135546370250560 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.938, Tokens/s/device: 13405.773, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0422 07:49:40.871760 135546370250560 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.212, Tokens/s/device: 13899.163, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0422 07:49:43.767298 2774 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 07:49:46.319660 135417768703744 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints/0/items/array_metadatas/process_2 I0422 07:50:02.863008 135546370250560 metric_logger.py:196] completed step: 6, seconds: 0.295, TFLOP/s/device: 46.002, Tokens/s/device: 6933.911, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0422 07:50:03.010366 135546370250560 metric_logger.py:196] completed step: 7, seconds: 21.844, TFLOP/s/device: 0.622, Tokens/s/device: 93.755, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0422 07:50:03.157574 135546370250560 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.168, Tokens/s/device: 13440.437, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0422 07:50:03.304202 135546370250560 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 07:50:03.304815 135546370250560 checkpointing.py:776] Waited 0.0006384849548339844 seconds for step 9 to finish before starting checkpointing. I0422 07:50:03.306825 135546370250560 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:50:17.362434 135421518407424 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 41.407 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.14785194396973 s) (per-host) I0422 07:50:17.362571 135421518407424 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.860951s. I0422 07:50:26.602834 135421518407424 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 47.101198s. I0422 07:50:26.603135 135417721005824 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:50:26.603265 135417721005824 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:50:26.603316 135417721005824 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:50:26.604886 135417721005824 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 07:50:26.605086 135546370250560 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0422 07:50:26.605237 135546370250560 checkpoint_manager.py:1452] Waiting for previous save to complete took 23.298413 seconds. If this number is high, consider checkpointing less frequently. I0422 07:50:26.607015 135546370250560 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 9 I0422 07:50:26.609121 135546370250560 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints/9. I0422 07:50:26.905403 135546370250560 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:50:26.905492 135546370250560 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 I0422 07:50:26.945222 135546370250560 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041052s I0422 07:50:26.945378 135546370250560 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 34.640 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04453158378601074 s) (per-host) I0422 07:50:26.945426 135546370250560 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044589s (batch_requests_ready=0.001796s, total_serialization_initiated=0.042730s, others=0.000062s) I0422 07:50:26.945514 135546370250560 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048964s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.048948) I0422 07:50:26.946175 135546370250560 event_tracking.py:125] [process=2] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints/9. I0422 07:50:26.946454 135417721005824 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 08:10:26.946426 I0422 07:50:26.948345 135546370250560 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:50:26.948632 135393064216320 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:50:26.948786 135546370250560 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776844203.3067932, 'wait_for_prev_duration_secs': 23.298413038253784, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776844226.6070583, 'checkpointer_blocking_duration_secs': 0.339493989944458, 'get_old_steps_start_time': 1776844226.9465752, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1776844203.3050456, 'checkpoint_manager_blocking_duration_secs': 23.64370822906494} I0422 07:50:26.948943 135546370250560 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 07:50:26.948989 135546370250560 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:50:32.931490 135417768703744 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_10_shardy_true/checkpoints/9/items/array_metadatas/process_2 I0422 07:51:09.094206 135417721005824 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 37.437 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.19331979751587 s) (per-host) I0422 07:51:09.094328 135417721005824 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.147797s. I0422 07:51:16.139063 135417721005824 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 49.192518s. I0422 07:51:16.139356 135393064216320 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:51:16.139478 135393064216320 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:51:16.139528 135393064216320 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:51:16.141067 135393064216320 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 07:51:16.141254 135546370250560 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 07:51:16.141402 135546370250560 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:51:16.142358 135546370250560 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.270, Tokens/s/device: 13907.941, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 07:51:25 UTC 2026 EXIT_CODE=0
29bbe47ed
· feat_nnx_trainstate_and_training_loop_20260422_093106
· full log
XPK Start: Wed Apr 22 10:05:35 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-22 10:06:00.424183: 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) I0422 10:06:00.635079 138944337086272 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 10:06:09.676084 138944337086272 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 10:06:09.678515 138944337086272 distributed.py:172] Connecting to JAX distributed service on mt-10-shardy-true-d2zkb-slice-job-0-0.mt-10-shardy-true-d2zkb:8482 I0422 10:06:11.114175 138944337086272 max_utils.py:284] Jax distributed system initialized! I0422 10:06:16.322038 138944337086272 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 10:06:16.322144 138944337086272 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 10:06:16.322185 138944337086272 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 I0422 10:06:16.322220 138944337086272 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 10:06:17.009933 138944337086272 maxtext_utils.py:1732] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 10:06:17.010526 138944337086272 maxtext_utils.py:1732] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 10:06:17.010705 138944337086272 checkpointing.py:688] Setting up checkpoint logger... I0422 10:06:17.010756 138944337086272 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 10:06:17.010799 138944337086272 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 10:06:17.011155 138944337086272 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 0x7e5dded14830>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 10:06:19.994951 138944337086272 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0422 10:06:19.995226 138944337086272 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e49ac2667b0>}, handler_registry=None I0422 10:06:19.995472 138944337086272 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e49ac2667b0>` 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`. I0422 10:06:19.995521 138944337086272 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e49ac2682f0>` 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`. I0422 10:06:19.995556 138944337086272 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 0x7e49ac2667b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e49ac2667b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e49ac2682f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e49ac2682f0>}). I0422 10:06:19.995889 138944337086272 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 10:06:19.995961 138944337086272 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e488c4fdd00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 10:06:21.518809 138944337086272 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints I0422 10:06:21.525148 138944337086272 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e5cbccffa10> I0422 10:06:21.525262 138944337086272 checkpointing.py:302] Checkpoint manager created! I0422 10:06:22.447122 138944337086272 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 10:06:22.447222 138944337086272 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 10:06:22.828021 138944337086272 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 10:06:22.828111 138944337086272 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 10:06:22.849789 138944337086272 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 10:06:22.849893 138944337086272 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 10:06:22.873590 138944337086272 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 10:06:22.873656 138944337086272 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 10:06:22.889948 138944337086272 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 10:06:22.890010 138944337086272 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 10:06:22.906212 138944337086272 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 10:06:22.906279 138944337086272 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 10:06:22.930804 138944337086272 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 10:06:22.930886 138944337086272 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 10:06:22.951372 138944337086272 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 10:06:22.951432 138944337086272 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 10:06:23.164724 138944337086272 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0422 10:06:23.164834 138944337086272 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0422 10:06:24.595957 138944337086272 maxtext_utils.py:1835] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 10:06:24.596086 138944337086272 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 10:06:24.596140 138944337086272 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 10:06:24.596196 138944337086272 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 10:06:24.596251 138944337086272 maxtext_utils.py:1835] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 10:06:24.596290 138944337086272 maxtext_utils.py:1835] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 10:06:24.596340 138944337086272 maxtext_utils.py:1835] 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) I0422 10:06:24.596391 138944337086272 maxtext_utils.py:1835] 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') I0422 10:06:24.596429 138944337086272 maxtext_utils.py:1835] 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) I0422 10:06:24.596465 138944337086272 maxtext_utils.py:1835] 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) I0422 10:06:24.596512 138944337086272 maxtext_utils.py:1835] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 10:06:24.596562 138944337086272 maxtext_utils.py:1835] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 10:06:25.100739 138944337086272 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 10:06:25.100839 138944337086272 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 10:06:25.116234 138944337086272 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 10:06:25.116292 138944337086272 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 10:06:36.038739 138944337086272 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 10:06:36.039533 138944337086272 metric_logger.py:301] number parameters: 1.104 billion I0422 10:06:47.508570 138944337086272 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0422 10:06:47.661662 138944337086272 checkpointing.py:798] Waited 0.15307235717773438 seconds for step 0 to finish before starting checkpointing. I0422 10:06:47.664363 138944337086272 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 10:06:47.666025 138944337086272 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0422 10:06:47.667928 138944337086272 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints/0. I0422 10:06:48.167534 138944337086272 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 10:06:48.168474 138944337086272 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 10:06:48.168534 138944337086272 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 I0422 10:06:48.441547 138944337086272 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274086s I0422 10:06:48.441718 138944337086272 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.514 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2797563076019287 s) (per-host) I0422 10:06:48.441770 138944337086272 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279819s (batch_requests_ready=0.002465s, total_serialization_initiated=0.277283s, others=0.000070s) I0422 10:06:48.441861 138944337086272 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283839s (all_items=0.000019s, per_item={'items': '0.00001907'}, temp_paths=0.283820) I0422 10:06:48.442694 138944337086272 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints/0. I0422 10:06:48.443034 138815887030016 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 10:26:48.442996 I0422 10:06:48.453321 138944337086272 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 10:06:48.453619 138814764398336 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0422 10:06:48.453776 138944337086272 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776852407.6643462, 'wait_for_prev_duration_secs': 6.461143493652344e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776852407.6660647, 'checkpointer_blocking_duration_secs': 0.7771172523498535, 'get_old_steps_start_time': 1776852408.4432075, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776852407.6622043, 'checkpoint_manager_blocking_duration_secs': 0.7915322780609131} I0422 10:06:48.453894 138944337086272 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0422 10:06:48.453945 138944337086272 max_utils.py:750] Memstats: After params initialized: I0422 10:06:48.453996 138944337086272 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0)) I0422 10:06:48.454030 138944337086272 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0422 10:06:48.454062 138944337086272 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0422 10:06:48.454087 138944337086272 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0422 10:06:48.769019 138944337086272 metric_logger.py:196] completed step: 0, seconds: 11.469, TFLOP/s/device: 1.185, Tokens/s/device: 178.569, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0422 10:06:48.937238 138944337086272 metric_logger.py:196] completed step: 1, seconds: 1.259, TFLOP/s/device: 10.792, Tokens/s/device: 1626.715, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0422 10:06:49.350239 138944337086272 metric_logger.py:196] completed step: 2, seconds: 0.021, TFLOP/s/device: 649.325, Tokens/s/device: 97873.357, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0422 10:06:49.497650 138944337086272 metric_logger.py:196] completed step: 3, seconds: 0.413, TFLOP/s/device: 32.872, Tokens/s/device: 4954.879, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0422 10:06:49.792681 138944337086272 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.786, Tokens/s/device: 13382.734, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0422 10:06:49.798399 138944337086272 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.252, Tokens/s/device: 13905.297, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0422 10:06:51.754943 2843 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 10:06:53.965936 138814811543296 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints/0/items/array_metadatas/process_4 I0422 10:07:12.206036 138944337086272 metric_logger.py:196] completed step: 6, seconds: 0.295, TFLOP/s/device: 46.013, Tokens/s/device: 6935.531, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0422 10:07:12.353398 138944337086272 metric_logger.py:196] completed step: 7, seconds: 22.261, TFLOP/s/device: 0.610, Tokens/s/device: 92.000, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0422 10:07:12.500902 138944337086272 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.150, Tokens/s/device: 13437.703, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0422 10:07:12.647563 138944337086272 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0422 10:07:12.648233 138944337086272 checkpointing.py:798] Waited 0.0006902217864990234 seconds for step 9 to finish before starting checkpointing. I0422 10:07:12.650354 138944337086272 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 10:07:26.223618 138815887030016 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 41.501 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.06163311004639 s) (per-host) I0422 10:07:26.223724 138815887030016 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.780581s. I0422 10:07:35.280353 138815887030016 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 46.837191s. I0422 10:07:35.280622 138814764398336 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 10:07:35.280745 138814764398336 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0422 10:07:35.280797 138814764398336 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 10:07:35.282378 138814764398336 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 10:07:35.282552 138944337086272 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0422 10:07:35.282699 138944337086272 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.632344 seconds. If this number is high, consider checkpointing less frequently. I0422 10:07:35.284485 138944337086272 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0422 10:07:35.286654 138944337086272 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints/9. I0422 10:07:35.960545 138944337086272 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 10:07:35.960638 138944337086272 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 I0422 10:07:36.001733 138944337086272 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041972s I0422 10:07:36.001892 138944337086272 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 33.898 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04550600051879883 s) (per-host) I0422 10:07:36.001942 138944337086272 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045574s (batch_requests_ready=0.001801s, total_serialization_initiated=0.043697s, others=0.000077s) I0422 10:07:36.002024 138944337086272 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049748s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.049732) I0422 10:07:36.002671 138944337086272 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.72 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints/9. I0422 10:07:36.003010 138814764398336 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 10:27:36.002972 I0422 10:07:36.004985 138944337086272 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 10:07:36.005260 138790869894912 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0422 10:07:36.005393 138944337086272 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776852432.6503263, 'wait_for_prev_duration_secs': 22.632344007492065, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776852455.2845235, 'checkpointer_blocking_duration_secs': 0.7186336517333984, 'get_old_steps_start_time': 1776852456.0031824, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1776852432.6484573, 'checkpoint_manager_blocking_duration_secs': 23.356902837753296} I0422 10:07:36.005561 138944337086272 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0422 10:07:36.005613 138944337086272 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 10:07:41.608776 138814811543296 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_10_shardy_true/checkpoints/9/items/array_metadatas/process_4 I0422 10:08:18.107824 138814764398336 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.474 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.151406049728394 s) (per-host) I0422 10:08:18.107970 138814764398336 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.104846s. I0422 10:08:27.523939 138814764398336 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 51.520799s. I0422 10:08:27.524230 138790869894912 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 10:08:27.524378 138790869894912 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0422 10:08:27.524438 138790869894912 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 10:08:27.526299 138790869894912 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 10:08:27.526473 138944337086272 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 10:08:27.526624 138944337086272 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 10:08:27.527509 138944337086272 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.010, Tokens/s/device: 13868.761, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 10:08:35 UTC 2026 EXIT_CODE=0