main (574ad3fb9) vs test/pipeline-scan-nnx (574ad3fb9)| Metric | main 574ad3fb9 | test/pipeline-scan-nnx 574ad3fb9 | Diff (test/pipeline-scan-nnx − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 7.1810 | 7.1810 | 0 |
| TFLOP/s | 92.112 | 92.265 | +0.153 |
| Tok/s | 13884.1 | 13907.2 | +23.099 |
| Avg s/step | 14.541 | 13.528 | -1.013 |
| Memory % | 1.38 | 1.38 | 0 |
| JAX | 0.9.2 | 0.8.1 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Sat Apr 18 19:11:48 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-18 19:12:12.015509: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0418 19:12:12.548936 139016938563392 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 19:12:21.589657 139016938563392 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 19:12:21.592174 139016938563392 distributed.py:172] Connecting to JAX distributed service on mt-14-async-ckpt-false--825vo-slice-job-0-0.mt-14-async-ckpt-false--825vo:8482 I0418 19:12:22.345292 139016938563392 max_utils.py:284] Jax distributed system initialized! I0418 19:12:28.331783 139016938563392 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 19:12:28.331888 139016938563392 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 19:12:28.331928 139016938563392 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0418 19:12:28.331964 139016938563392 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 19:12:29.025496 139016938563392 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 19:12:29.025770 139016938563392 checkpointing.py:677] Setting up checkpoint logger... I0418 19:12:29.025824 139016938563392 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 19:12:29.025879 139016938563392 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 19:12:29.026230 139016938563392 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 0x7e6ec4510170>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 19:12:31.948239 139016938563392 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0418 19:12:31.948471 139016938563392 checkpoint_manager.py:708] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e5a7c112ae0>}, handler_registry=None I0418 19:12:31.948704 139016938563392 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e5a7c112ae0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0418 19:12:31.948750 139016938563392 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e5a7c113140>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0418 19:12:31.948785 139016938563392 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 0x7e5a7c112ae0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e5a7c112ae0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e5a7c113140>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e5a7c113140>}). I0418 19:12:31.949123 139016938563392 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 19:12:33.603208 139016938563392 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints I0418 19:12:34.463975 139016938563392 checkpoint_manager.py:929] [process=7][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=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_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e5a7c2fcf20> I0418 19:12:34.464167 139016938563392 checkpointing.py:301] Checkpoint manager created! I0418 19:12:35.417415 139016938563392 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 19:12:35.417529 139016938563392 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 19:12:35.796562 139016938563392 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 19:12:35.796649 139016938563392 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 19:12:35.812982 139016938563392 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 19:12:35.813041 139016938563392 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 19:12:35.837281 139016938563392 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 19:12:35.837351 139016938563392 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 19:12:35.853809 139016938563392 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 19:12:35.853874 139016938563392 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 19:12:35.870159 139016938563392 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 19:12:35.870220 139016938563392 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 19:12:35.894760 139016938563392 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 19:12:35.894828 139016938563392 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 19:12:35.915779 139016938563392 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 19:12:35.915844 139016938563392 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 19:12:36.134122 139016938563392 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0418 19:12:36.134234 139016938563392 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 19:12:37.578081 139016938563392 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 19:12:37.578206 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 19:12:37.578258 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 19:12:37.578314 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0418 19:12:37.578366 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 19:12:37.578404 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 19:12:37.578454 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0418 19:12:37.578506 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0418 19:12:37.578547 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0418 19:12:37.578583 139016938563392 maxtext_utils.py:1654] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0418 19:12:37.578634 139016938563392 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 19:12:37.578682 139016938563392 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 19:12:38.067620 139016938563392 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 19:12:38.067723 139016938563392 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 19:12:38.083106 139016938563392 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 19:12:38.083166 139016938563392 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 19:12:48.924404 139016938563392 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. I0418 19:12:48.925253 139016938563392 metric_logger.py:301] number parameters: 1.104 billion I0418 19:13:00.018760 139016938563392 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0418 19:13:00.363813 139016938563392 checkpointing.py:776] Waited 0.3450334072113037 seconds for step 0 to finish before starting checkpointing. I0418 19:13:00.366234 139016938563392 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 19:13:00.367920 139016938563392 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0418 19:13:00.369482 139016938563392 event_tracking.py:70] [process=7] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/0. I0418 19:13:04.677109 139016938563392 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 19:13:04.678389 139016938563392 future.py:372] [process=7][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. I0418 19:13:06.473264 139016938563392 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 19:13:06.473362 139016938563392 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 19:13:06.754823 139016938563392 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.282868s I0418 19:13:06.755020 139016938563392 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 5.346 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2885310649871826 s) (per-host) I0418 19:13:06.755094 139016938563392 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.288601s (batch_requests_ready=0.002357s, total_serialization_initiated=0.286166s, others=0.000078s) I0418 19:13:06.755196 139016938563392 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.078214s (all_items=0.000033s, per_item={'items': '0.00003290'}, temp_paths=2.078181) I0418 19:13:06.755245 139016938563392 future.py:372] [process=7][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. I0418 19:13:06.755286 139016938563392 future.py:372] [process=7][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. I0418 19:13:06.768987 2778 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 19:13:10.590637 138886445901568 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/0/items/array_metadatas/process_7 I0418 19:13:38.891208 139016938563392 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 48.716 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 32.42467951774597 s) (per-host) I0418 19:13:46.057825 139016938563392 event_tracking.py:125] [process=7] [sync] Finished blocking save in 45.69 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/0. I0418 19:13:51.037813 139016938563392 event_tracking.py:138] [process=7] [sync] Finished save in 50.67 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/0 I0418 19:13:51.040398 139016938563392 checkpoint_manager.py:2137] [process=7][thread=MainThread][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 19:13:51.042356 139016938563392 checkpoint_manager.py:2146] [process=7][thread=MainThread][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 19:13:51.042408 139016938563392 checkpoint_manager.py:1581] [process=7][thread=MainThread][step=0] Finished synchronous save. I0418 19:13:51.042464 139016938563392 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776539580.3662155, 'wait_for_prev_duration_secs': 6.222724914550781e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776539580.3679574, 'checkpointer_blocking_duration_secs': 50.669997692108154, 'get_old_steps_start_time': 1776539631.0379791, 'get_old_steps_duration_secs': 2.5510787963867188e-05, 'checkpoint_manager_blocking_start_time': 1776539580.3643627, 'checkpoint_manager_blocking_duration_secs': 50.678072929382324} I0418 19:13:51.042553 139016938563392 checkpointing.py:410] Saved a checkpoint at step 0. I0418 19:13:51.042598 139016938563392 max_utils.py:750] Memstats: After params initialized: I0418 19:13:51.042644 139016938563392 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_26(process=7,(2,6,0,0)) I0418 19:13:51.042675 139016938563392 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0)) I0418 19:13:51.042703 139016938563392 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0)) I0418 19:13:51.042727 139016938563392 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0)) I0418 19:13:51.363059 139016938563392 metric_logger.py:196] completed step: 0, seconds: 11.093, TFLOP/s/device: 1.225, Tokens/s/device: 184.614, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 19:13:51.530789 139016938563392 metric_logger.py:196] completed step: 1, seconds: 51.337, TFLOP/s/device: 0.265, Tokens/s/device: 39.893, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0418 19:13:51.949755 139016938563392 metric_logger.py:196] completed step: 2, seconds: 0.026, TFLOP/s/device: 527.389, Tokens/s/device: 79493.848, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0418 19:13:52.097509 139016938563392 metric_logger.py:196] completed step: 3, seconds: 0.419, TFLOP/s/device: 32.398, Tokens/s/device: 4883.306, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0418 19:13:52.102066 139016938563392 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0418 19:13:52.391683 139016938563392 checkpointing.py:776] Waited 0.2895810604095459 seconds for step 5 to finish before starting checkpointing. I0418 19:13:52.394587 139016938563392 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 19:13:52.396659 139016938563392 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 5 I0418 19:13:52.398689 139016938563392 event_tracking.py:70] [process=7] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/5. I0418 19:13:55.701721 139016938563392 future.py:372] [process=7][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. I0418 19:13:57.514626 139016938563392 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 19:13:57.514728 139016938563392 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 19:13:57.550301 139016938563392 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036952s I0418 19:13:57.550453 139016938563392 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 37.901 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.040700435638427734 s) (per-host) I0418 19:13:57.550508 139016938563392 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040763s (batch_requests_ready=0.001791s, total_serialization_initiated=0.038902s, others=0.000070s) I0418 19:13:57.550594 139016938563392 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.850245s (all_items=0.000028s, per_item={'items': '0.00002789'}, temp_paths=1.850218) I0418 19:13:57.550636 139016938563392 future.py:372] [process=7][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. I0418 19:13:57.550672 139016938563392 future.py:372] [process=7][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. I0418 19:14:00.366059 138888633988864 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/5/items/array_metadatas/process_7 I0418 19:14:34.958979 139016938563392 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 42.180 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.44917106628418 s) (per-host) I0418 19:14:42.133447 139016938563392 event_tracking.py:125] [process=7] [sync] Finished blocking save in 49.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/5. I0418 19:14:46.982785 139016938563392 event_tracking.py:138] [process=7] [sync] Finished save in 54.59 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/5 I0418 19:14:46.985069 139016938563392 checkpoint_manager.py:2137] [process=7][thread=MainThread][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0418 19:14:46.987099 139016938563392 checkpoint_manager.py:2146] [process=7][thread=MainThread][step=5] CheckpointManager Save Finalize is done on all hosts. I0418 19:14:46.987150 139016938563392 checkpoint_manager.py:1581] [process=7][thread=MainThread][step=5] Finished synchronous save. I0418 19:14:46.987205 139016938563392 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776539632.394567, 'wait_for_prev_duration_secs': 6.318092346191406e-05, 'time_between_consecutive_saves_sec': 1.3521697521209717, 'checkpointer_blocking_start_time': 1776539632.3966985, 'checkpointer_blocking_duration_secs': 54.586207151412964, 'get_old_steps_start_time': 1776539686.9829257, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1776539632.3923445, 'checkpoint_manager_blocking_duration_secs': 54.59483242034912} I0418 19:14:46.987294 139016938563392 checkpointing.py:410] Saved a checkpoint at step 5. I0418 19:14:46.988112 139016938563392 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.658, Tokens/s/device: 13363.523, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0418 19:14:47.004925 139016938563392 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.112, Tokens/s/device: 13884.087, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0418 19:15:10.739608 139016938563392 metric_logger.py:196] completed step: 6, seconds: 54.891, TFLOP/s/device: 0.248, Tokens/s/device: 37.311, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0418 19:15:10.887140 139016938563392 metric_logger.py:196] completed step: 7, seconds: 23.599, TFLOP/s/device: 0.576, Tokens/s/device: 86.782, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0418 19:15:11.034745 139016938563392 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.189, Tokens/s/device: 13443.525, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0418 19:15:11.181392 139016938563392 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0418 19:15:11.182067 139016938563392 checkpointing.py:776] Waited 0.0006973743438720703 seconds for step 9 to finish before starting checkpointing. I0418 19:15:11.184140 139016938563392 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 19:15:11.186104 139016938563392 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 9 I0418 19:15:11.187450 139016938563392 event_tracking.py:70] [process=7] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/9. I0418 19:15:14.614771 139016938563392 future.py:372] [process=7][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. I0418 19:15:16.393754 139016938563392 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0418 19:15:16.393860 139016938563392 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 19:15:16.428682 139016938563392 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037516s I0418 19:15:16.428861 139016938563392 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 37.449 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04119086265563965 s) (per-host) I0418 19:15:16.428914 139016938563392 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.041255s (batch_requests_ready=0.001850s, total_serialization_initiated=0.039335s, others=0.000069s) I0418 19:15:16.429015 139016938563392 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.816090s (all_items=0.000032s, per_item={'items': '0.00003219'}, temp_paths=1.816058) I0418 19:15:16.429056 139016938563392 future.py:372] [process=7][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. I0418 19:15:16.429112 139016938563392 future.py:372] [process=7][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. I0418 19:15:18.917462 138892372068096 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/9/items/array_metadatas/process_7 I0418 19:15:53.644312 139016938563392 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 42.398 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.25660276412964 s) (per-host) I0418 19:16:03.932551 139016938563392 event_tracking.py:125] [process=7] [sync] Finished blocking save in 52.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/9. I0418 19:16:09.187234 139016938563392 event_tracking.py:138] [process=7] [sync] Finished save in 58.00 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints/9 I0418 19:16:09.189231 139016938563392 checkpoint_manager.py:2137] [process=7][thread=MainThread][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0418 19:16:09.191120 139016938563392 checkpoint_manager.py:2146] [process=7][thread=MainThread][step=9] CheckpointManager Save Finalize is done on all hosts. I0418 19:16:09.191170 139016938563392 checkpoint_manager.py:1581] [process=7][thread=MainThread][step=9] Finished synchronous save. I0418 19:16:09.191217 139016938563392 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776539711.1841223, 'wait_for_prev_duration_secs': 5.698204040527344e-05, 'time_between_consecutive_saves_sec': 24.19698429107666, 'checkpointer_blocking_start_time': 1776539711.1861417, 'checkpointer_blocking_duration_secs': 58.00120496749878, 'get_old_steps_start_time': 1776539769.1873596, 'get_old_steps_duration_secs': 2.1219253540039062e-05, 'checkpoint_manager_blocking_start_time': 1776539711.1823123, 'checkpoint_manager_blocking_duration_secs': 58.00888156890869} I0418 19:16:09.191302 139016938563392 checkpointing.py:410] Saved a checkpoint at step 9. I0418 19:16:09.191334 139016938563392 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 19:16:09.191364 139016938563392 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 19:16:09.192053 139016938563392 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.112, Tokens/s/device: 13884.087, 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: Sat Apr 18 19:16:18 UTC 2026 EXIT_CODE=0
XPK Start: Sun Apr 19 09:45:56 UTC 2026 2026-04-19 09:46:00.603576: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered WARNING: All log messages before absl::InitializeLog() is called are written to STDERR E0000 00:00:1776591960.616018 10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776591960.619868 10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776591960.631325 10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776591960.631346 10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776591960.631349 10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776591960.631351 10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. 2026-04-19 09:46:19.917324: 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) I0419 09:46:20.432171 132159858333504 max_utils.py:273] Attempting to initialize the jax distributed system... INFO:2026-04-19 09:46:29,470:jax._src.distributed:140: Starting JAX distributed service on [::]:8482 I0419 09:46:29.470629 132159858333504 distributed.py:140] Starting JAX distributed service on [::]:8482 INFO:2026-04-19 09:46:29,473:jax._src.distributed:157: Connecting to JAX distributed service on mt-14-async-ckpt-false--atkh9-slice-job-0-0.mt-14-async-ckpt-false--atkh9:8482 I0419 09:46:29.473099 132159858333504 distributed.py:157] Connecting to JAX distributed service on mt-14-async-ckpt-false--atkh9-slice-job-0-0.mt-14-async-ckpt-false--atkh9:8482 I0419 09:46:30.608531 132159858333504 max_utils.py:284] Jax distributed system initialized! I0419 09:46:37.570633 132159858333504 max_utils.py:800] System Information: Jax Version: 0.8.1 I0419 09:46:37.570742 132159858333504 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0419 09:46:37.570784 132159858333504 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709 I0419 09:46:37.570818 132159858333504 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0419 09:46:38.191235 132159858333504 maxtext_utils.py:1398] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 09:46:38.191525 132159858333504 checkpointing.py:677] Setting up checkpoint logger... I0419 09:46:38.191581 132159858333504 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0419 09:46:38.191628 132159858333504 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None I0419 09:46:38.191992 132159858333504 base_pytree_checkpoint_handler.py:415] 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 0x783285d5f500>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0419 09:46:41.494119 132159858333504 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0419 09:46:41.494360 132159858333504 checkpoint_manager.py:709] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x783119c5f260>}, handler_registry=None I0419 09:46:41.494591 132159858333504 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x783119c5f260>` 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`. I0419 09:46:41.494642 132159858333504 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x781df00d0b00>` 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`. I0419 09:46:41.494679 132159858333504 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 0x783119c5f260>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x783119c5f260>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x781df00d0b00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x781df00d0b00>}). I0419 09:46:41.497316 132159858333504 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33 I0419 09:46:42.207648 132159858333504 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints I0419 09:46:42.966595 132159858333504 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=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_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x781df02c4d70> I0419 09:46:42.966776 132159858333504 checkpointing.py:301] Checkpoint manager created! I0419 09:46:43.855324 132159858333504 nnx_wrappers.py:453] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0419 09:46:43.855430 132159858333504 nnx_wrappers.py:453] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 09:46:44.225820 132159858333504 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 09:46:44.225915 132159858333504 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 09:46:44.242053 132159858333504 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 09:46:44.242118 132159858333504 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 09:46:44.265916 132159858333504 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 09:46:44.265998 132159858333504 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 09:46:44.282085 132159858333504 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 09:46:44.282146 132159858333504 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 09:46:44.298152 132159858333504 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 09:46:44.298212 132159858333504 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 09:46:44.322290 132159858333504 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0419 09:46:44.322367 132159858333504 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 09:46:44.342056 132159858333504 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0419 09:46:44.342125 132159858333504 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0419 09:46:44.532029 132159858333504 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0419 09:46:44.532139 132159858333504 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0419 09:46:45.961709 132159858333504 maxtext_utils.py:1501] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0419 09:46:45.961852 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 09:46:45.961908 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 09:46:45.961993 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0419 09:46:45.962054 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0419 09:46:45.962096 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0419 09:46:45.962150 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0419 09:46:45.962204 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: PartitionSpec('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0419 09:46:45.962244 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0419 09:46:45.962282 132159858333504 maxtext_utils.py:1501] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0419 09:46:45.962330 132159858333504 maxtext_utils.py:1501] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed', 'vocab') Physical: ('fsdp', None) I0419 09:46:45.962375 132159858333504 maxtext_utils.py:1501] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed') Physical: (None, 'fsdp') I0419 09:46:46.465037 132159858333504 train.py:156] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0419 09:46:46.465133 132159858333504 train.py:156] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0419 09:46:46.480257 132159858333504 train.py:163] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0419 09:46:46.480316 132159858333504 train.py:163] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0419 09:46:57.145363 132159858333504 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. I0419 09:46:57.146276 132159858333504 metric_logger.py:289] number parameters: 1.104 billion I0419 09:47:08.122581 132159858333504 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0419 09:47:08.278632 132159858333504 checkpointing.py:776] Waited 0.1560370922088623 seconds for step 0 to finish before starting checkpointing. I0419 09:47:08.281146 132159858333504 checkpoint_manager.py:2013] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 09:47:08.283061 132159858333504 checkpoint_manager.py:1518] [process=4] Saving checkpoint at step 0 I0419 09:47:08.284501 132159858333504 checkpointer.py:241] [process=4] Started saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/0. I0419 09:47:11.243968 132159858333504 signaling_client.py:364] Using JaxDistributedSignalingClient I0419 09:47:11.245347 132159858333504 future.py:363] [process=4][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. I0419 09:47:12.881540 132159858333504 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array. I0419 09:47:12.881637 132159858333504 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 I0419 09:47:13.133493 132159858333504 base_pytree_checkpoint_handler.py:153] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.253298s I0419 09:47:13.133672 132159858333504 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/blocking_gbytes_per_sec: 3.011 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5123474597930908 s) (per-host) I0419 09:47:13.133726 132159858333504 base_pytree_checkpoint_handler.py:737] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.512411s (batch_requests_ready=0.255501s, total_serialization_initiated=0.256839s, others=0.000071s) I0419 09:47:13.133822 132159858333504 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.889964s (all_items=0.000036s, per_item={'items': '0.00003600'}, temp_paths=1.889928) I0419 09:47:13.133867 132159858333504 future.py:363] [process=4][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. I0419 09:47:13.133907 132159858333504 future.py:363] [process=4][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. I0419 09:47:13.144408 2597 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0419 09:47:15.460841 132036782171904 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/0/items/array_metadatas/process_4 I0419 09:47:47.458016 132159858333504 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/gbytes_per_sec: 45.343 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.836652755737305 s) (per-host) I0419 09:47:56.366444 132159858333504 checkpointer.py:290] Finished synchronous save in 48.08 seconds to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/0 I0419 09:47:56.368457 132159858333504 checkpoint_manager.py:2133] [process=4][thread=MainThread][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0419 09:47:56.370114 132159858333504 checkpoint_manager.py:2142] [process=4][thread=MainThread][step=0] CheckpointManager Save Finalize is done on all hosts. I0419 09:47:56.370162 132159858333504 checkpoint_manager.py:1587] [process=4][thread=MainThread][step=0] Finished synchronous save. I0419 09:47:56.370216 132159858333504 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776592028.2811284, 'wait_for_prev_duration_secs': 6.771087646484375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776592028.2830987, 'checkpointer_blocking_duration_secs': 48.08346247673035, 'get_old_steps_start_time': 1776592076.3665826, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1776592028.2791874, 'checkpoint_manager_blocking_duration_secs': 48.09099984169006} I0419 09:47:56.370301 132159858333504 checkpointing.py:410] Saved a checkpoint at step 0. I0419 09:47:56.370345 132159858333504 max_utils.py:750] Memstats: After params initialized: I0419 09:47:56.370391 132159858333504 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0)) I0419 09:47:56.370422 132159858333504 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0419 09:47:56.370447 132159858333504 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0419 09:47:56.370470 132159858333504 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0419 09:47:56.652896 132159858333504 metric_logger.py:185] completed step: 0, seconds: 10.976, TFLOP/s/device: 1.238, Tokens/s/device: 186.586, total_weights: 65536, loss: 10.874 I0419 09:47:56.804482 132159858333504 metric_logger.py:185] completed step: 1, seconds: 48.512, TFLOP/s/device: 0.280, Tokens/s/device: 42.216, total_weights: 65536, loss: 10.874 I0419 09:47:57.248888 132159858333504 metric_logger.py:185] completed step: 2, seconds: 0.022, TFLOP/s/device: 612.336, Tokens/s/device: 92297.985, total_weights: 65536, loss: 10.560 I0419 09:47:57.396368 132159858333504 metric_logger.py:185] completed step: 3, seconds: 0.444, TFLOP/s/device: 30.589, Tokens/s/device: 4610.702, total_weights: 65536, loss: 9.980 I0419 09:47:57.400760 132159858333504 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0419 09:47:57.691092 132159858333504 checkpointing.py:776] Waited 0.2902953624725342 seconds for step 5 to finish before starting checkpointing. I0419 09:47:57.693869 132159858333504 checkpoint_manager.py:2013] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 09:47:57.695505 132159858333504 checkpoint_manager.py:1518] [process=4] Saving checkpoint at step 5 I0419 09:47:57.697088 132159858333504 checkpointer.py:241] [process=4] Started saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/5. I0419 09:48:00.650184 132159858333504 future.py:363] [process=4][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. I0419 09:48:02.709975 132159858333504 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array. I0419 09:48:02.710079 132159858333504 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 I0419 09:48:02.751216 132159858333504 base_pytree_checkpoint_handler.py:153] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042437s I0419 09:48:02.751396 132159858333504 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.126 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.7257428169250488 s) (per-host) I0419 09:48:02.751451 132159858333504 base_pytree_checkpoint_handler.py:737] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.725807s (batch_requests_ready=0.681401s, total_serialization_initiated=0.044335s, others=0.000071s) I0419 09:48:02.751553 132159858333504 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 2.102688s (all_items=0.000027s, per_item={'items': '0.00002742'}, temp_paths=2.102660) I0419 09:48:02.751592 132159858333504 future.py:363] [process=4][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. I0419 09:48:02.751629 132159858333504 future.py:363] [process=4][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. I0419 09:48:04.164495 132036782171904 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/5/items/array_metadatas/process_4 I0419 09:48:41.700052 132159858333504 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/gbytes_per_sec: 39.814 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 39.67436647415161 s) (per-host) I0419 09:48:49.713621 132159858333504 checkpointer.py:290] Finished synchronous save in 52.02 seconds to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/5 I0419 09:48:49.715652 132159858333504 checkpoint_manager.py:2133] [process=4][thread=MainThread][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0419 09:48:49.717473 132159858333504 checkpoint_manager.py:2142] [process=4][thread=MainThread][step=5] CheckpointManager Save Finalize is done on all hosts. I0419 09:48:49.717523 132159858333504 checkpoint_manager.py:1587] [process=4][thread=MainThread][step=5] Finished synchronous save. I0419 09:48:49.717574 132159858333504 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776592077.6938505, 'wait_for_prev_duration_secs': 6.413459777832031e-05, 'time_between_consecutive_saves_sec': 1.3236985206604004, 'checkpointer_blocking_start_time': 1776592077.695544, 'checkpointer_blocking_duration_secs': 52.018197536468506, 'get_old_steps_start_time': 1776592129.713772, 'get_old_steps_duration_secs': 2.5033950805664062e-05, 'checkpoint_manager_blocking_start_time': 1776592077.691798, 'checkpoint_manager_blocking_duration_secs': 52.02575135231018} I0419 09:48:49.717661 132159858333504 checkpointing.py:410] Saved a checkpoint at step 5. I0419 09:48:49.718162 132159858333504 metric_logger.py:185] completed step: 4, seconds: 0.152, TFLOP/s/device: 89.190, Tokens/s/device: 13443.701, total_weights: 65536, loss: 9.460 I0419 09:48:49.740966 132159858333504 metric_logger.py:185] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.174, Tokens/s/device: 13893.506, total_weights: 65536, loss: 8.959 I0419 09:49:09.724740 132159858333504 metric_logger.py:185] completed step: 6, seconds: 52.322, TFLOP/s/device: 0.260, Tokens/s/device: 39.142, total_weights: 65536, loss: 8.469 I0419 09:49:09.872199 132159858333504 metric_logger.py:185] completed step: 7, seconds: 19.855, TFLOP/s/device: 0.684, Tokens/s/device: 103.150, total_weights: 65536, loss: 8.003 I0419 09:49:10.022078 132159858333504 metric_logger.py:185] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.155, Tokens/s/device: 13438.408, total_weights: 65536, loss: 7.572 I0419 09:49:10.169489 132159858333504 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0419 09:49:10.170187 132159858333504 checkpointing.py:776] Waited 0.0007205009460449219 seconds for step 9 to finish before starting checkpointing. I0419 09:49:10.172188 132159858333504 checkpoint_manager.py:2013] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 09:49:10.173827 132159858333504 checkpoint_manager.py:1518] [process=4] Saving checkpoint at step 9 I0419 09:49:10.175147 132159858333504 checkpointer.py:241] [process=4] Started saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/9. I0419 09:49:13.009303 132159858333504 future.py:363] [process=4][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. I0419 09:49:14.651391 132159858333504 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array. I0419 09:49:14.651496 132159858333504 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 I0419 09:49:14.695574 132159858333504 base_pytree_checkpoint_handler.py:153] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.045164s I0419 09:49:14.695751 132159858333504 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/blocking_gbytes_per_sec: 5.183 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2976415157318115 s) (per-host) I0419 09:49:14.695805 132159858333504 base_pytree_checkpoint_handler.py:737] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.297705s (batch_requests_ready=0.250442s, total_serialization_initiated=0.047194s, others=0.000070s) I0419 09:49:14.695904 132159858333504 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.687907s (all_items=0.000020s, per_item={'items': '0.00002027'}, temp_paths=1.687887) I0419 09:49:14.695959 132159858333504 future.py:363] [process=4][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. I0419 09:49:14.696004 132159858333504 future.py:363] [process=4][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. I0419 09:49:16.495283 132033066039040 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/9/items/array_metadatas/process_4 I0419 09:49:52.584205 132159858333504 base_pytree_checkpoint_handler.py:129] [process=4] /jax/checkpoint/write/gbytes_per_sec: 41.366 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.18605613708496 s) (per-host) I0419 09:50:02.800450 132159858333504 checkpointer.py:290] Finished synchronous save in 52.63 seconds to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints/9 I0419 09:50:02.803107 132159858333504 checkpoint_manager.py:2133] [process=4][thread=MainThread][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0419 09:50:02.805001 132159858333504 checkpoint_manager.py:2142] [process=4][thread=MainThread][step=9] CheckpointManager Save Finalize is done on all hosts. I0419 09:50:02.805056 132159858333504 checkpoint_manager.py:1587] [process=4][thread=MainThread][step=9] Finished synchronous save. I0419 09:50:02.805107 132159858333504 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260419_085454/linen_xpk_test_pipeline_scan_nnx_20260419_085454_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1776592150.172172, 'wait_for_prev_duration_secs': 6.270408630371094e-05, 'time_between_consecutive_saves_sec': 20.454659938812256, 'checkpointer_blocking_start_time': 1776592150.173866, 'checkpointer_blocking_duration_secs': 52.62670588493347, 'get_old_steps_start_time': 1776592202.8005848, 'get_old_steps_duration_secs': 1.9550323486328125e-05, 'checkpoint_manager_blocking_start_time': 1776592150.1704118, 'checkpoint_manager_blocking_duration_secs': 52.634671211242676} I0419 09:50:02.805190 132159858333504 checkpointing.py:410] Saved a checkpoint at step 9. I0419 09:50:02.805222 132159858333504 checkpoint_manager.py:2013] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 09:50:02.805252 132159858333504 checkpoint_manager.py:2013] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 09:50:02.805639 132159858333504 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.265, Tokens/s/device: 13907.186, total_weights: 65536, loss: 7.181 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sun Apr 19 09:50:16 UTC 2026 EXIT_CODE=0