main (586e69205) vs test/pipeline-scan-nnx (5b21b66fc)| Metric | main 586e69205 | test/pipeline-scan-nnx 5b21b66fc | Diff (test/pipeline-scan-nnx − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 7.1810 | 7.1810 | 0 |
| TFLOP/s | 92.223 | 92.201 | -0.022 |
| Tok/s | 13900.9 | 13897.6 | -3.301 |
| Avg s/step | 13.809 | 13.763 | -0.046 |
| 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: Thu Apr 23 08:08:51 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. 2026-04-23 08:09:16.009448: 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) I0423 08:09:16.222002 132090224158528 max_utils.py:273] Attempting to initialize the jax distributed system... I0423 08:09:25.261249 132090224158528 distributed.py:149] Starting JAX distributed service on [::]:8482 I0423 08:09:25.263519 132090224158528 distributed.py:172] Connecting to JAX distributed service on mt-14-async-ckpt-false--99l1e-slice-job-0-0.mt-14-async-ckpt-false--99l1e:8482 I0423 08:09:26.351478 132090224158528 max_utils.py:284] Jax distributed system initialized! I0423 08:09:32.722122 132090224158528 max_utils.py:800] System Information: Jax Version: 0.9.2 I0423 08:09:32.722229 132090224158528 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0423 08:09:32.722268 132090224158528 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 I0423 08:09:32.722308 132090224158528 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 08:09:33.429997 132090224158528 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 08:09:33.430296 132090224158528 checkpointing.py:677] Setting up checkpoint logger... I0423 08:09:33.430356 132090224158528 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0423 08:09:33.430401 132090224158528 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0423 08:09:33.430735 132090224158528 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 0x7822353c26c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 08:09:36.503528 132090224158528 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0423 08:09:36.503765 132090224158528 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x780de877f3e0>}, handler_registry=None I0423 08:09:36.504008 132090224158528 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x780de877f3e0>` 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`. I0423 08:09:36.504058 132090224158528 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x780de8782f00>` 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`. I0423 08:09:36.504107 132090224158528 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 0x780de877f3e0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x780de877f3e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x780de8782f00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x780de8782f00>}). I0423 08:09:36.504437 132090224158528 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0423 08:09:38.508623 132090224158528 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints I0423 08:09:38.511401 132090224158528 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=False, 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=5), 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_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x780de877fa10> I0423 08:09:38.511523 132090224158528 checkpointing.py:301] Checkpoint manager created! I0423 08:09:39.445586 132090224158528 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 08:09:39.445703 132090224158528 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:09:39.825119 132090224158528 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0423 08:09:39.825212 132090224158528 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:09:39.842002 132090224158528 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0423 08:09:39.842063 132090224158528 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:09:39.866126 132090224158528 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:09:39.866195 132090224158528 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:09:39.882915 132090224158528 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:09:39.882975 132090224158528 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:09:39.899611 132090224158528 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:09:39.899670 132090224158528 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:09:39.924766 132090224158528 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0423 08:09:39.924846 132090224158528 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:09:39.946339 132090224158528 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 08:09:39.946402 132090224158528 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0423 08:09:40.159889 132090224158528 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0423 08:09:40.160003 132090224158528 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0423 08:09:41.592607 132090224158528 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0423 08:09:41.592740 132090224158528 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 08:09:41.592794 132090224158528 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 08:09:41.592853 132090224158528 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0423 08:09:41.593056 132090224158528 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 08:09:41.593138 132090224158528 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 08:09:41.593207 132090224158528 maxtext_utils.py:1707] 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) I0423 08:09:41.593273 132090224158528 maxtext_utils.py:1707] 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') I0423 08:09:41.593315 132090224158528 maxtext_utils.py:1707] 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) I0423 08:09:41.593351 132090224158528 maxtext_utils.py:1707] 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) I0423 08:09:41.593398 132090224158528 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 08:09:41.593447 132090224158528 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 08:09:42.089645 132090224158528 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 08:09:42.089747 132090224158528 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0423 08:09:42.105577 132090224158528 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 08:09:42.105642 132090224158528 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0423 08:09:53.022798 132090224158528 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. I0423 08:09:53.023551 132090224158528 metric_logger.py:301] number parameters: 1.104 billion I0423 08:10:04.931851 132090224158528 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0423 08:10:05.084162 132090224158528 checkpointing.py:776] Waited 0.15229201316833496 seconds for step 0 to finish before starting checkpointing. I0423 08:10:05.086827 132090224158528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:10:05.088452 132090224158528 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0423 08:10:05.089913 132090224158528 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/0. I0423 08:10:08.274255 132090224158528 signaling_client.py:364] Using JaxDistributedSignalingClient I0423 08:10:08.275666 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:10:10.145296 132090224158528 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:10:10.145400 132090224158528 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 I0423 08:10:10.427692 132090224158528 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.283530s I0423 08:10:10.427877 132090224158528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.331 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2893669605255127 s) (per-host) I0423 08:10:10.427931 132090224158528 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.289432s (batch_requests_ready=0.002378s, total_serialization_initiated=0.286981s, others=0.000072s) I0423 08:10:10.428032 132090224158528 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.153864s (all_items=0.000039s, per_item={'items': '0.00003910'}, temp_paths=2.153825) I0423 08:10:10.428079 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:10:10.428149 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:10:10.440849 2774 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 08:10:13.930608 131961607665408 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/0/items/array_metadatas/process_6 I0423 08:10:44.177163 132090224158528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 46.406 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.038597106933594 s) (per-host) I0423 08:10:49.324490 132090224158528 event_tracking.py:125] [process=6] [sync] Finished blocking save in 44.24 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/0. I0423 08:10:53.703381 132090224158528 event_tracking.py:138] [process=6] [sync] Finished save in 48.61 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/0 I0423 08:10:53.705996 132090224158528 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:10:53.707779 132090224158528 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=0] CheckpointManager Save Finalize is done on all hosts. I0423 08:10:53.707831 132090224158528 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=0] Finished synchronous save. I0423 08:10:53.707888 132090224158528 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776931805.0868075, 'wait_for_prev_duration_secs': 6.365776062011719e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776931805.088491, 'checkpointer_blocking_duration_secs': 48.61500692367554, 'get_old_steps_start_time': 1776931853.7035213, 'get_old_steps_duration_secs': 2.4557113647460938e-05, 'checkpoint_manager_blocking_start_time': 1776931805.0847125, 'checkpoint_manager_blocking_duration_secs': 48.623146295547485} I0423 08:10:53.707978 132090224158528 checkpointing.py:410] Saved a checkpoint at step 0. I0423 08:10:53.708023 132090224158528 max_utils.py:750] Memstats: After params initialized: I0423 08:10:53.708070 132090224158528 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0423 08:10:53.708113 132090224158528 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0423 08:10:53.708142 132090224158528 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0423 08:10:53.708166 132090224158528 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0423 08:10:54.033793 132090224158528 metric_logger.py:196] completed step: 0, seconds: 11.908, TFLOP/s/device: 1.141, Tokens/s/device: 171.983, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 08:10:54.211807 132090224158528 metric_logger.py:196] completed step: 1, seconds: 49.087, TFLOP/s/device: 0.277, Tokens/s/device: 41.722, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 08:10:54.623366 132090224158528 metric_logger.py:196] completed step: 2, seconds: 0.044, TFLOP/s/device: 305.562, Tokens/s/device: 46057.662, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0423 08:10:54.770668 132090224158528 metric_logger.py:196] completed step: 3, seconds: 0.412, TFLOP/s/device: 32.972, Tokens/s/device: 4969.945, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0423 08:10:54.775447 132090224158528 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0423 08:10:55.064792 132090224158528 checkpointing.py:776] Waited 0.28931164741516113 seconds for step 5 to finish before starting checkpointing. I0423 08:10:55.067753 132090224158528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:10:55.069578 132090224158528 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5 I0423 08:10:55.071292 132090224158528 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/5. I0423 08:10:57.566881 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:10:58.973841 132090224158528 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:10:58.973942 132090224158528 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 I0423 08:10:59.018708 132090224158528 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.046237s I0423 08:10:59.018896 132090224158528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 30.881 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04995274543762207 s) (per-host) I0423 08:10:59.018953 132090224158528 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.050021s (batch_requests_ready=0.001949s, total_serialization_initiated=0.047996s, others=0.000077s) I0423 08:10:59.019055 132090224158528 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.453460s (all_items=0.000029s, per_item={'items': '0.00002861'}, temp_paths=1.453432) I0423 08:10:59.019110 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:10:59.019150 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:11:02.027965 131961607665408 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/5/items/array_metadatas/process_6 I0423 08:11:38.516503 132090224158528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 39.942 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 39.547524213790894 s) (per-host) I0423 08:11:42.223091 132090224158528 event_tracking.py:125] [process=6] [sync] Finished blocking save in 47.15 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/5. I0423 08:11:46.232196 132090224158528 event_tracking.py:138] [process=6] [sync] Finished save in 51.16 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/5 I0423 08:11:46.235321 132090224158528 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:11:46.237131 132090224158528 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=5] CheckpointManager Save Finalize is done on all hosts. I0423 08:11:46.237182 132090224158528 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=5] Finished synchronous save. I0423 08:11:46.237235 132090224158528 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776931855.0677316, 'wait_for_prev_duration_secs': 6.413459777832031e-05, 'time_between_consecutive_saves_sec': 1.3599121570587158, 'checkpointer_blocking_start_time': 1776931855.0696182, 'checkpointer_blocking_duration_secs': 51.162700176239014, 'get_old_steps_start_time': 1776931906.2323508, 'get_old_steps_duration_secs': 2.288818359375e-05, 'checkpoint_manager_blocking_start_time': 1776931855.065446, 'checkpoint_manager_blocking_duration_secs': 51.171761989593506} I0423 08:11:46.237324 132090224158528 checkpointing.py:410] Saved a checkpoint at step 5. I0423 08:11:46.238126 132090224158528 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 89.024, Tokens/s/device: 13418.597, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0423 08:11:46.260696 132090224158528 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.133, Tokens/s/device: 13887.288, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0423 08:12:09.063855 132090224158528 metric_logger.py:196] completed step: 6, seconds: 51.467, TFLOP/s/device: 0.264, Tokens/s/device: 39.792, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0423 08:12:09.211359 132090224158528 metric_logger.py:196] completed step: 7, seconds: 22.674, TFLOP/s/device: 0.599, Tokens/s/device: 90.323, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0423 08:12:09.358974 132090224158528 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.487, Tokens/s/device: 13488.504, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0423 08:12:09.505381 132090224158528 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0423 08:12:09.506073 132090224158528 checkpointing.py:776] Waited 0.0007143020629882812 seconds for step 9 to finish before starting checkpointing. I0423 08:12:09.508010 132090224158528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:12:09.510213 132090224158528 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0423 08:12:09.511732 132090224158528 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/9. I0423 08:12:12.143566 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:12:13.595447 132090224158528 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:12:13.595547 132090224158528 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 I0423 08:12:13.636990 132090224158528 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042694s I0423 08:12:13.637187 132090224158528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 33.245 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.046399593353271484 s) (per-host) I0423 08:12:13.637240 132090224158528 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046463s (batch_requests_ready=0.001787s, total_serialization_initiated=0.044607s, others=0.000068s) I0423 08:12:13.637341 132090224158528 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.495047s (all_items=0.000021s, per_item={'items': '0.00002098'}, temp_paths=1.495026) I0423 08:12:13.637384 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:12:13.637421 132090224158528 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 08:12:16.175303 131967005738752 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/9/items/array_metadatas/process_6 I0423 08:12:52.453687 132090224158528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 40.645 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.86285734176636 s) (per-host) I0423 08:12:56.093533 132090224158528 event_tracking.py:125] [process=6] [sync] Finished blocking save in 46.58 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/9. I0423 08:13:00.160883 132090224158528 event_tracking.py:138] [process=6] [sync] Finished save in 50.65 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints/9 I0423 08:13:00.162889 132090224158528 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:13:00.164697 132090224158528 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=9] CheckpointManager Save Finalize is done on all hosts. I0423 08:13:00.164750 132090224158528 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=9] Finished synchronous save. I0423 08:13:00.164812 132090224158528 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776931929.507994, 'wait_for_prev_duration_secs': 5.745887756347656e-05, 'time_between_consecutive_saves_sec': 23.270823001861572, 'checkpointer_blocking_start_time': 1776931929.510251, 'checkpointer_blocking_duration_secs': 50.65075159072876, 'get_old_steps_start_time': 1776931980.161017, 'get_old_steps_duration_secs': 1.9550323486328125e-05, 'checkpoint_manager_blocking_start_time': 1776931929.5063179, 'checkpoint_manager_blocking_duration_secs': 50.658464431762695} I0423 08:13:00.164923 132090224158528 checkpointing.py:410] Saved a checkpoint at step 9. I0423 08:13:00.164972 132090224158528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:13:00.165018 132090224158528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:13:00.165697 132090224158528 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.223, Tokens/s/device: 13900.861, total_weights: 65536, loss: 7.181, lm_loss: 7.181, perplexity: 1314.495 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Thu Apr 23 08:13:11 UTC 2026 EXIT_CODE=0
XPK Start: Thu Apr 23 22:00:29 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. 2026-04-23 22:00:54.770399: 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) I0423 22:00:54.982676 138216919734080 max_utils.py:273] Attempting to initialize the jax distributed system... I0423 22:01:04.025260 138216919734080 distributed.py:149] Starting JAX distributed service on [::]:8482 I0423 22:01:04.027741 138216919734080 distributed.py:172] Connecting to JAX distributed service on mt-14-async-ckpt-false--316qf-slice-job-0-0.mt-14-async-ckpt-false--316qf:8482 I0423 22:01:04.941275 138216919734080 max_utils.py:284] Jax distributed system initialized! I0423 22:01:11.190452 138216919734080 max_utils.py:800] System Information: Jax Version: 0.9.2 I0423 22:01:11.190562 138216919734080 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0423 22:01:11.190602 138216919734080 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 I0423 22:01:11.190637 138216919734080 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 22:01:11.896543 138216919734080 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 22:01:11.896829 138216919734080 checkpointing.py:677] Setting up checkpoint logger... I0423 22:01:11.896887 138216919734080 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0423 22:01:11.896932 138216919734080 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0423 22:01:11.897295 138216919734080 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 0x7db46ad39d60>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 22:01:14.879194 138216919734080 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0423 22:01:14.879438 138216919734080 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7da050473710>}, handler_registry=None I0423 22:01:14.879702 138216919734080 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7da050473710>` 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`. I0423 22:01:14.879759 138216919734080 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7db35e7d1430>` 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`. I0423 22:01:14.879804 138216919734080 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 0x7da050473710>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7da050473710>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7db35e7d1430>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7db35e7d1430>}). I0423 22:01:14.880147 138216919734080 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0423 22:01:16.692003 138216919734080 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints I0423 22:01:16.694243 138216919734080 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=False, 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=5), 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_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7da050475a60> I0423 22:01:16.694358 138216919734080 checkpointing.py:301] Checkpoint manager created! I0423 22:01:17.627597 138216919734080 nnx_wrappers.py:453] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 22:01:17.627715 138216919734080 nnx_wrappers.py:453] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 22:01:18.008504 138216919734080 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 22:01:18.008598 138216919734080 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 22:01:18.025174 138216919734080 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 22:01:18.025238 138216919734080 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 22:01:18.048952 138216919734080 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 22:01:18.049020 138216919734080 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 22:01:18.065678 138216919734080 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 22:01:18.065739 138216919734080 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 22:01:18.082252 138216919734080 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 22:01:18.082318 138216919734080 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 22:01:18.107250 138216919734080 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0423 22:01:18.107325 138216919734080 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 22:01:18.128361 138216919734080 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 22:01:18.128433 138216919734080 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0423 22:01:18.341600 138216919734080 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0423 22:01:18.341712 138216919734080 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0423 22:01:19.785551 138216919734080 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0423 22:01:19.785703 138216919734080 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) I0423 22:01:19.785791 138216919734080 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) I0423 22:01:19.785880 138216919734080 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') I0423 22:01:19.785964 138216919734080 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) I0423 22:01:19.786025 138216919734080 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) I0423 22:01:19.786133 138216919734080 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) I0423 22:01:19.786219 138216919734080 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') I0423 22:01:19.786283 138216919734080 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) I0423 22:01:19.786345 138216919734080 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) I0423 22:01:19.786418 138216919734080 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 22:01:19.786490 138216919734080 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 22:01:20.281167 138216919734080 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 22:01:20.281270 138216919734080 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0423 22:01:20.296842 138216919734080 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 22:01:20.296902 138216919734080 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0423 22:01:31.154487 138216919734080 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. I0423 22:01:31.155337 138216919734080 metric_logger.py:301] number parameters: 1.104 billion I0423 22:01:42.462032 138216919734080 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0423 22:01:43.144933 138216919734080 checkpointing.py:776] Waited 0.6828780174255371 seconds for step 0 to finish before starting checkpointing. I0423 22:01:43.147494 138216919734080 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 22:01:43.149266 138216919734080 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0423 22:01:43.150861 138216919734080 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/0. I0423 22:01:45.633736 138216919734080 signaling_client.py:364] Using JaxDistributedSignalingClient I0423 22:01:45.635185 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:01:47.008816 138216919734080 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 22:01:47.008914 138216919734080 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 I0423 22:01:47.288275 138216919734080 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.280406s I0423 22:01:47.288462 138216919734080 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.381 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28666067123413086 s) (per-host) I0423 22:01:47.288520 138216919734080 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.286727s (batch_requests_ready=0.002861s, total_serialization_initiated=0.283791s, others=0.000075s) I0423 22:01:47.288618 138216919734080 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.654967s (all_items=0.000038s, per_item={'items': '0.00003791'}, temp_paths=1.654929) I0423 22:01:47.294651 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:01:47.294725 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:01:47.301393 2794 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 22:01:49.332606 138088609937152 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/0/items/array_metadatas/process_6 I0423 22:02:21.540559 138216919734080 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 45.734 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.538721323013306 s) (per-host) I0423 22:02:25.495923 138216919734080 event_tracking.py:125] [process=6] [sync] Finished blocking save in 42.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/0. I0423 22:02:29.761346 138216919734080 event_tracking.py:138] [process=6] [sync] Finished save in 46.61 seconds @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/0 I0423 22:02:29.763726 138216919734080 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0423 22:02:29.765730 138216919734080 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=0] CheckpointManager Save Finalize is done on all hosts. I0423 22:02:29.765783 138216919734080 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=0] Finished synchronous save. I0423 22:02:29.765835 138216919734080 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776981703.1474757, 'wait_for_prev_duration_secs': 6.103515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776981703.1493053, 'checkpointer_blocking_duration_secs': 46.612160444259644, 'get_old_steps_start_time': 1776981749.761488, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1776981703.1455178, 'checkpoint_manager_blocking_duration_secs': 46.62028980255127} I0423 22:02:29.765925 138216919734080 checkpointing.py:410] Saved a checkpoint at step 0. I0423 22:02:29.765968 138216919734080 max_utils.py:750] Memstats: After params initialized: I0423 22:02:29.766016 138216919734080 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0423 22:02:29.766047 138216919734080 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0423 22:02:29.766073 138216919734080 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0423 22:02:29.766110 138216919734080 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0423 22:02:30.081758 138216919734080 metric_logger.py:196] completed step: 0, seconds: 11.307, TFLOP/s/device: 1.202, Tokens/s/device: 181.133, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 22:02:30.255671 138216919734080 metric_logger.py:196] completed step: 1, seconds: 47.612, TFLOP/s/device: 0.285, Tokens/s/device: 43.014, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 22:02:30.665023 138216919734080 metric_logger.py:196] completed step: 2, seconds: 0.032, TFLOP/s/device: 420.810, Tokens/s/device: 63429.138, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0423 22:02:30.812410 138216919734080 metric_logger.py:196] completed step: 3, seconds: 0.410, TFLOP/s/device: 33.136, Tokens/s/device: 4994.549, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0423 22:02:30.816993 138216919734080 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0423 22:02:31.106813 138216919734080 checkpointing.py:776] Waited 0.28978466987609863 seconds for step 5 to finish before starting checkpointing. I0423 22:02:31.109765 138216919734080 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 22:02:31.111564 138216919734080 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5 I0423 22:02:31.113218 138216919734080 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/5. I0423 22:02:33.801351 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:02:35.249427 138216919734080 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 22:02:35.249524 138216919734080 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 I0423 22:02:35.284353 138216919734080 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036087s I0423 22:02:35.284518 138216919734080 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 38.578 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03998613357543945 s) (per-host) I0423 22:02:35.284571 138216919734080 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040050s (batch_requests_ready=0.001898s, total_serialization_initiated=0.038078s, others=0.000074s) I0423 22:02:35.284662 138216919734080 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.484590s (all_items=0.000028s, per_item={'items': '0.00002766'}, temp_paths=1.484562) I0423 22:02:35.284703 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:02:35.284738 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:02:37.765735 138088609937152 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/5/items/array_metadatas/process_6 I0423 22:03:14.153922 138216919734080 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 40.597 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.90935182571411 s) (per-host) I0423 22:03:19.514219 138216919734080 event_tracking.py:125] [process=6] [sync] Finished blocking save in 48.40 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/5. I0423 22:03:23.459844 138216919734080 event_tracking.py:138] [process=6] [sync] Finished save in 52.35 seconds @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/5 I0423 22:03:23.462459 138216919734080 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0423 22:03:23.464223 138216919734080 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=5] CheckpointManager Save Finalize is done on all hosts. I0423 22:03:23.464276 138216919734080 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=5] Finished synchronous save. I0423 22:03:23.464329 138216919734080 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776981751.1097457, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': 1.3439738750457764, 'checkpointer_blocking_start_time': 1776981751.1116016, 'checkpointer_blocking_duration_secs': 52.34835934638977, 'get_old_steps_start_time': 1776981803.4599802, 'get_old_steps_duration_secs': 2.384185791015625e-05, 'checkpoint_manager_blocking_start_time': 1776981751.1074834, 'checkpoint_manager_blocking_duration_secs': 52.356820821762085} I0423 22:03:23.464420 138216919734080 checkpointing.py:410] Saved a checkpoint at step 5. I0423 22:03:23.465278 138216919734080 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.958, Tokens/s/device: 13408.670, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0423 22:03:23.482402 138216919734080 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.249, Tokens/s/device: 13904.825, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0423 22:03:46.175282 138216919734080 metric_logger.py:196] completed step: 6, seconds: 52.653, TFLOP/s/device: 0.258, Tokens/s/device: 38.896, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0423 22:03:46.322652 138216919734080 metric_logger.py:196] completed step: 7, seconds: 22.558, TFLOP/s/device: 0.602, Tokens/s/device: 90.790, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0423 22:03:46.470226 138216919734080 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.177, Tokens/s/device: 13441.672, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0423 22:03:46.616972 138216919734080 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0423 22:03:46.617574 138216919734080 checkpointing.py:776] Waited 0.0006241798400878906 seconds for step 9 to finish before starting checkpointing. I0423 22:03:46.619630 138216919734080 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 22:03:46.621637 138216919734080 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0423 22:03:46.622944 138216919734080 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/9. I0423 22:03:49.181782 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:03:50.651309 138216919734080 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 22:03:50.651407 138216919734080 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 I0423 22:03:50.690575 138216919734080 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.040293s I0423 22:03:50.690742 138216919734080 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 34.765 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.044371604919433594 s) (per-host) I0423 22:03:50.690793 138216919734080 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044432s (batch_requests_ready=0.001872s, total_serialization_initiated=0.042495s, others=0.000065s) I0423 22:03:50.690887 138216919734080 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.510335s (all_items=0.000029s, per_item={'items': '0.00002933'}, temp_paths=1.510306) I0423 22:03:50.690928 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:03:50.690964 138216919734080 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture. I0423 22:03:52.736449 138092878993152 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/9/items/array_metadatas/process_6 I0423 22:04:29.054951 138216919734080 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 41.126 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.408541679382324 s) (per-host) I0423 22:04:33.608333 138216919734080 event_tracking.py:125] [process=6] [sync] Finished blocking save in 46.99 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/9. I0423 22:04:38.125260 138216919734080 event_tracking.py:138] [process=6] [sync] Finished save in 51.50 seconds @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints/9 I0423 22:04:38.127761 138216919734080 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0423 22:04:38.129899 138216919734080 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=9] CheckpointManager Save Finalize is done on all hosts. I0423 22:04:38.129950 138216919734080 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=9] Finished synchronous save. I0423 22:04:38.129999 138216919734080 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776981826.6196125, 'wait_for_prev_duration_secs': 5.626678466796875e-05, 'time_between_consecutive_saves_sec': 23.15534782409668, 'checkpointer_blocking_start_time': 1776981826.6216774, 'checkpointer_blocking_duration_secs': 51.50370168685913, 'get_old_steps_start_time': 1776981878.1253936, 'get_old_steps_duration_secs': 1.9311904907226562e-05, 'checkpoint_manager_blocking_start_time': 1776981826.6177917, 'checkpoint_manager_blocking_duration_secs': 51.512184143066406} I0423 22:04:38.130084 138216919734080 checkpointing.py:410] Saved a checkpoint at step 9. I0423 22:04:38.130131 138216919734080 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 22:04:38.130162 138216919734080 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 22:04:38.130899 138216919734080 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.201, Tokens/s/device: 13897.560, total_weights: 65536, loss: 7.181, lm_loss: 7.181, perplexity: 1314.495 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Thu Apr 23 22:04:50 UTC 2026 EXIT_CODE=0