main (574ad3fb9) vs feat/nnx-post-train-fixes (574ad3fb9)| Metric | main 574ad3fb9 | feat/nnx-post-train-fixes 574ad3fb9 | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 7.1810 | 7.1810 | 0 |
| TFLOP/s | 92.144 | 92.332 | +0.188 |
| Tok/s | 13889.0 | 13917.3 | +28.315 |
| Avg s/step | 4.659 | 4.847 | +0.188 |
| 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: Mon Apr 20 07:09:18 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-20 07:09:43.053847: 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) I0420 07:09:43.234245 137142412810048 max_utils.py:273] Attempting to initialize the jax distributed system... I0420 07:09:52.275719 137142412810048 distributed.py:149] Starting JAX distributed service on [::]:8482 I0420 07:09:52.278049 137142412810048 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-4yx59-slice-job-0-0.mt-15-ocdbt-false-sv-4yx59:8482 I0420 07:09:53.541022 137142412810048 max_utils.py:284] Jax distributed system initialized! I0420 07:09:58.818147 137142412810048 max_utils.py:800] System Information: Jax Version: 0.9.2 I0420 07:09:58.818249 137142412810048 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0420 07:09:58.818287 137142412810048 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 I0420 07:09:58.818320 137142412810048 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0420 07:09:59.509570 137142412810048 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 07:09:59.509858 137142412810048 checkpointing.py:677] Setting up checkpoint logger... I0420 07:09:59.509911 137142412810048 checkpointing.py:233] Creating checkpoint manager with ocdbt=False and zarr3=True I0420 07:09:59.509957 137142412810048 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0420 07:09:59.510287 137142412810048 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7cba5b71c9e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0420 07:10:02.442781 137142412810048 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0420 07:10:02.443008 137142412810048 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ca5587173b0>}, handler_registry=None I0420 07:10:02.443241 137142412810048 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ca5587173b0>` 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`. I0420 07:10:02.443289 137142412810048 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ca55871c950>` 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`. I0420 07:10:02.443324 137142412810048 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 0x7ca5587173b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ca5587173b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ca55871c950>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ca55871c950>}). I0420 07:10:02.443666 137142412810048 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0420 07:10:02.443736 137142412810048 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7ca558697d80> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0420 07:10:03.665135 137142412810048 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints I0420 07:10:03.667277 137142412810048 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=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_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ca558716780> I0420 07:10:03.667400 137142412810048 checkpointing.py:301] Checkpoint manager created! I0420 07:10:04.593490 137142412810048 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0420 07:10:04.593604 137142412810048 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 07:10:04.970936 137142412810048 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 07:10:04.971028 137142412810048 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 07:10:04.987378 137142412810048 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 07:10:04.987434 137142412810048 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 07:10:05.010936 137142412810048 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 07:10:05.010998 137142412810048 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 07:10:05.027322 137142412810048 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 07:10:05.027393 137142412810048 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 07:10:05.043657 137142412810048 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 07:10:05.043720 137142412810048 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 07:10:05.068223 137142412810048 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0420 07:10:05.068293 137142412810048 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 07:10:05.090169 137142412810048 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0420 07:10:05.090230 137142412810048 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0420 07:10:05.306899 137142412810048 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0420 07:10:05.307004 137142412810048 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0420 07:10:06.727974 137142412810048 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0420 07:10:06.728096 137142412810048 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) I0420 07:10:06.728147 137142412810048 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) I0420 07:10:06.728203 137142412810048 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') I0420 07:10:06.728255 137142412810048 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) I0420 07:10:06.728294 137142412810048 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) I0420 07:10:06.728358 137142412810048 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) I0420 07:10:06.728415 137142412810048 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') I0420 07:10:06.728454 137142412810048 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) I0420 07:10:06.728491 137142412810048 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) I0420 07:10:06.728537 137142412810048 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0420 07:10:06.728590 137142412810048 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0420 07:10:07.215329 137142412810048 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0420 07:10:07.215434 137142412810048 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0420 07:10:07.230874 137142412810048 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0420 07:10:07.230938 137142412810048 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0420 07:10:18.124564 137142412810048 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. I0420 07:10:18.125315 137142412810048 metric_logger.py:301] number parameters: 1.104 billion I0420 07:10:29.234916 137142412810048 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0420 07:10:29.542640 137142412810048 checkpointing.py:776] Waited 0.30770349502563477 seconds for step 0 to finish before starting checkpointing. I0420 07:10:29.545270 137142412810048 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 07:10:29.546876 137142412810048 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0420 07:10:29.548512 137142412810048 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/0. I0420 07:10:29.956531 137142412810048 signaling_client.py:364] Using JaxDistributedSignalingClient I0420 07:10:29.957489 137142412810048 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 07:10:29.957545 137142412810048 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 I0420 07:10:30.231169 137142412810048 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274711s I0420 07:10:30.231356 137142412810048 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.507 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.280123233795166 s) (per-host) I0420 07:10:30.231413 137142412810048 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.280208s (batch_requests_ready=0.002211s, total_serialization_initiated=0.277905s, others=0.000092s) I0420 07:10:30.231506 137142412810048 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.284128s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.284110) I0420 07:10:30.232362 137142412810048 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.69 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/0. I0420 07:10:30.232701 137014355687168 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 07:30:30.232662 I0420 07:10:30.239993 137142412810048 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0420 07:10:30.240251 137012745074432 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0420 07:10:30.240387 137142412810048 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776669029.5452518, 'wait_for_prev_duration_secs': 6.031990051269531e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776669029.5469139, 'checkpointer_blocking_duration_secs': 0.6858887672424316, 'get_old_steps_start_time': 1776669030.232828, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776669029.5431995, 'checkpoint_manager_blocking_duration_secs': 0.6971490383148193} I0420 07:10:30.240493 137142412810048 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0420 07:10:30.240545 137142412810048 max_utils.py:750] Memstats: After params initialized: I0420 07:10:30.240596 137142412810048 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0)) I0420 07:10:30.240629 137142412810048 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0420 07:10:30.240656 137142412810048 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0420 07:10:30.240681 137142412810048 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0420 07:10:30.551624 137142412810048 metric_logger.py:196] completed step: 0, seconds: 11.109, TFLOP/s/device: 1.223, Tokens/s/device: 184.347, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0420 07:10:30.727224 137142412810048 metric_logger.py:196] completed step: 1, seconds: 1.315, TFLOP/s/device: 10.331, Tokens/s/device: 1557.179, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0420 07:10:31.147266 137142412810048 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 479.856, Tokens/s/device: 72329.154, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0420 07:10:31.294656 137142412810048 metric_logger.py:196] completed step: 3, seconds: 0.420, TFLOP/s/device: 32.329, Tokens/s/device: 4872.953, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0420 07:10:31.299791 137142412810048 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0420 07:10:31.588995 137142412810048 checkpointing.py:776] Waited 0.2891807556152344 seconds for step 5 to finish before starting checkpointing. I0420 07:10:31.591951 137142412810048 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 07:10:33.826515 2833 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0420 07:10:36.471354 137013226391296 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_4 I0420 07:10:41.001918 137014355687168 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 142.941 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.050664901733398 s) (per-host) I0420 07:10:41.002040 137014355687168 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.768707s. I0420 07:10:47.572778 137014355687168 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 17.339429s. I0420 07:10:47.573086 137012745074432 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 07:10:47.573213 137012745074432 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0420 07:10:47.573264 137012745074432 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0420 07:10:47.574650 137012745074432 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0420 07:10:47.574860 137142412810048 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0420 07:10:47.574993 137142412810048 checkpoint_manager.py:1452] Waiting for previous save to complete took 15.983043 seconds. If this number is high, consider checkpointing less frequently. I0420 07:10:47.576889 137142412810048 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 5 I0420 07:10:47.579312 137142412810048 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/5. I0420 07:10:47.873512 137142412810048 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 07:10:47.873609 137142412810048 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 I0420 07:10:47.915518 137142412810048 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.043214s I0420 07:10:47.915690 137142412810048 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 33.123 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04657125473022461 s) (per-host) I0420 07:10:47.915743 137142412810048 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046633s (batch_requests_ready=0.001745s, total_serialization_initiated=0.044818s, others=0.000070s) I0420 07:10:47.915830 137142412810048 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.051811s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.051796) I0420 07:10:47.916493 137142412810048 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/5. I0420 07:10:47.916848 137012745074432 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 07:30:47.916807 I0420 07:10:47.918894 137142412810048 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0420 07:10:47.919194 137014338901760 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0420 07:10:47.919367 137142412810048 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776669031.5919218, 'wait_for_prev_duration_secs': 15.98304271697998, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776669047.57693, 'checkpointer_blocking_duration_secs': 0.34006595611572266, 'get_old_steps_start_time': 1776669047.9170167, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776669031.5896683, 'checkpoint_manager_blocking_duration_secs': 16.329653024673462} I0420 07:10:47.919480 137142412810048 checkpointing.py:408] Started an asynchronous checkpoint save for step 5 I0420 07:10:47.920261 137142412810048 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.666, Tokens/s/device: 13364.657, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0420 07:10:47.925996 137142412810048 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 91.978, Tokens/s/device: 13863.879, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0420 07:10:53.473248 137013226391296 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_4 I0420 07:10:58.357959 137012745074432 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 150.598 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.488804578781128 s) (per-host) I0420 07:10:58.358067 137012745074432 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.441104s. I0420 07:11:04.712649 137012745074432 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 16.795668s. I0420 07:11:04.712932 137014338901760 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 07:11:04.713049 137014338901760 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0420 07:11:04.713097 137014338901760 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0420 07:11:04.714656 137014338901760 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0420 07:11:11.013418 137142412810048 metric_logger.py:196] completed step: 6, seconds: 16.625, TFLOP/s/device: 0.817, Tokens/s/device: 123.189, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0420 07:11:11.160748 137142412810048 metric_logger.py:196] completed step: 7, seconds: 22.941, TFLOP/s/device: 0.592, Tokens/s/device: 89.272, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0420 07:11:11.308440 137142412810048 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.314, Tokens/s/device: 13462.347, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0420 07:11:11.455109 137142412810048 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0420 07:11:11.455721 137142412810048 checkpointing.py:776] Waited 0.00063323974609375 seconds for step 9 to finish before starting checkpointing. I0420 07:11:11.457905 137142412810048 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 07:11:11.459610 137142412810048 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0420 07:11:11.461052 137142412810048 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/9. I0420 07:11:12.016849 137142412810048 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 07:11:12.016941 137142412810048 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 I0420 07:11:12.057393 137142412810048 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041591s I0420 07:11:12.057543 137142412810048 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 34.302 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.044970035552978516 s) (per-host) I0420 07:11:12.057592 137142412810048 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045034s (batch_requests_ready=0.001620s, total_serialization_initiated=0.043346s, others=0.000068s) I0420 07:11:12.057671 137142412810048 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049020s (all_items=0.000011s, per_item={'items': '0.00001144'}, temp_paths=0.049009) I0420 07:11:12.058375 137142412810048 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.60 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/9. I0420 07:11:12.058679 136988770359040 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 07:31:12.058646 I0420 07:11:12.060577 137142412810048 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0420 07:11:12.060765 136988820715264 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0420 07:11:12.060870 137142412810048 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776669071.4578726, 'wait_for_prev_duration_secs': 7.176399230957031e-05, 'time_between_consecutive_saves_sec': 6.743176221847534, 'checkpointer_blocking_start_time': 1776669071.459647, 'checkpointer_blocking_duration_secs': 0.599170446395874, 'get_old_steps_start_time': 1776669072.0588386, 'get_old_steps_duration_secs': 2.4080276489257812e-05, 'checkpoint_manager_blocking_start_time': 1776669071.4559577, 'checkpoint_manager_blocking_duration_secs': 0.6048836708068848} I0420 07:11:12.061015 137142412810048 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0420 07:11:12.061062 137142412810048 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 07:11:17.511127 136988904642304 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_4 I0420 07:11:23.138358 136988770359040 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 141.977 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.125738620758057 s) (per-host) I0420 07:11:23.138476 136988770359040 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.079689s. I0420 07:11:29.191485 136988770359040 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 17.132682s. I0420 07:11:29.191693 136988820715264 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 07:11:29.191749 136988820715264 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0420 07:11:29.191789 136988820715264 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0420 07:11:29.193185 136988820715264 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0420 07:11:29.193352 137142412810048 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0420 07:11:29.193499 137142412810048 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 07:11:29.194430 137142412810048 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.144, Tokens/s/device: 13888.983, 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: Mon Apr 20 07:11:39 UTC 2026 EXIT_CODE=0
XPK Start: Mon Apr 20 17:07:29 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-20 17:07:53.293152: 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) I0420 17:07:53.472885 134928181425984 max_utils.py:273] Attempting to initialize the jax distributed system... I0420 17:08:02.513166 134928181425984 distributed.py:149] Starting JAX distributed service on [::]:8482 I0420 17:08:02.515488 134928181425984 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-false-sv-2d10z-slice-job-0-0.mt-15-ocdbt-false-sv-2d10z:8482 I0420 17:08:03.684596 134928181425984 max_utils.py:284] Jax distributed system initialized! I0420 17:08:09.729644 134928181425984 max_utils.py:800] System Information: Jax Version: 0.9.2 I0420 17:08:09.729750 134928181425984 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0420 17:08:09.729791 134928181425984 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 I0420 17:08:09.729824 134928181425984 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0420 17:08:10.425420 134928181425984 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 17:08:10.426037 134928181425984 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0420 17:08:10.426218 134928181425984 checkpointing.py:688] Setting up checkpoint logger... I0420 17:08:10.426267 134928181425984 checkpointing.py:234] Creating checkpoint manager with ocdbt=False and zarr3=True I0420 17:08:10.426309 134928181425984 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0420 17:08:10.426655 134928181425984 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=False, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7ab6b2d383e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0420 17:08:13.495186 134928181425984 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0420 17:08:13.495415 134928181425984 checkpoint_manager.py:708] [process=2][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7aa2801a70b0>}, handler_registry=None I0420 17:08:13.495665 134928181425984 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7aa2801a70b0>` 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`. I0420 17:08:13.495721 134928181425984 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa2985193d0>` 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`. I0420 17:08:13.495763 134928181425984 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 0x7aa2801a70b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7aa2801a70b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa2985193d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa2985193d0>}). I0420 17:08:13.496086 134928181425984 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0420 17:08:13.496154 134928181425984 async_checkpointer.py:192] [process=2][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7aa298143740> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0420 17:08:14.576610 134928181425984 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints I0420 17:08:14.578818 134928181425984 checkpoint_manager.py:929] [process=2][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=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_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7aa298517b00> I0420 17:08:14.578927 134928181425984 checkpointing.py:302] Checkpoint manager created! I0420 17:08:15.532077 134928181425984 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0420 17:08:15.532183 134928181425984 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 17:08:15.921722 134928181425984 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 17:08:15.921809 134928181425984 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 17:08:15.941102 134928181425984 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0420 17:08:15.941162 134928181425984 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0420 17:08:15.967881 134928181425984 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 17:08:15.967950 134928181425984 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 17:08:15.987587 134928181425984 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 17:08:15.987650 134928181425984 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 17:08:16.007069 134928181425984 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0420 17:08:16.007130 134928181425984 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 17:08:16.034806 134928181425984 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0420 17:08:16.034879 134928181425984 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0420 17:08:16.056306 134928181425984 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0420 17:08:16.056369 134928181425984 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0420 17:08:16.306822 134928181425984 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0420 17:08:16.306930 134928181425984 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0420 17:08:17.756627 134928181425984 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0420 17:08:17.756754 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0420 17:08:17.756807 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0420 17:08:17.756865 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0420 17:08:17.756920 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0420 17:08:17.756959 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0420 17:08:17.757017 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0420 17:08:17.757071 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0420 17:08:17.757113 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0420 17:08:17.757150 134928181425984 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0420 17:08:17.757199 134928181425984 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0420 17:08:17.757247 134928181425984 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0420 17:08:18.249417 134928181425984 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0420 17:08:18.249513 134928181425984 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0420 17:08:18.264801 134928181425984 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0420 17:08:18.264863 134928181425984 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0420 17:08:29.128841 134928181425984 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. I0420 17:08:29.129615 134928181425984 metric_logger.py:301] number parameters: 1.104 billion I0420 17:08:40.515228 134928181425984 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0420 17:08:40.671123 134928181425984 checkpointing.py:798] Waited 0.15587949752807617 seconds for step 0 to finish before starting checkpointing. I0420 17:08:40.673421 134928181425984 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 17:08:40.675177 134928181425984 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 0 I0420 17:08:40.676551 134928181425984 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/0. I0420 17:08:41.395118 134928181425984 signaling_client.py:364] Using JaxDistributedSignalingClient I0420 17:08:41.396019 134928181425984 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 17:08:41.396081 134928181425984 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 I0420 17:08:41.670476 134928181425984 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.275435s I0420 17:08:41.670657 134928181425984 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 5.488 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2810659408569336 s) (per-host) I0420 17:08:41.670720 134928181425984 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.281135s (batch_requests_ready=0.002443s, total_serialization_initiated=0.278615s, others=0.000076s) I0420 17:08:41.670833 134928181425984 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.285543s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.285526) I0420 17:08:41.671696 134928181425984 event_tracking.py:125] [process=2] [async] Finished blocking save in 1.00 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/0. I0420 17:08:41.672005 134800388470528 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:28:41.671971 I0420 17:08:41.687320 134928181425984 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0420 17:08:41.687602 134798777857792 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0420 17:08:41.687734 134928181425984 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776704920.673404, 'wait_for_prev_duration_secs': 6.079673767089844e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776704920.6752152, 'checkpointer_blocking_duration_secs': 0.9969379901885986, 'get_old_steps_start_time': 1776704921.6721795, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776704920.6715674, 'checkpoint_manager_blocking_duration_secs': 1.0161197185516357} I0420 17:08:41.687840 134928181425984 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0420 17:08:41.687888 134928181425984 max_utils.py:750] Memstats: After params initialized: I0420 17:08:41.687939 134928181425984 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_8(process=2,(0,2,0,0)) I0420 17:08:41.687972 134928181425984 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_9(process=2,(1,2,0,0)) I0420 17:08:41.687999 134928181425984 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_12(process=2,(0,3,0,0)) I0420 17:08:41.688023 134928181425984 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_13(process=2,(1,3,0,0)) I0420 17:08:41.998719 134928181425984 metric_logger.py:196] completed step: 0, seconds: 11.386, TFLOP/s/device: 1.193, Tokens/s/device: 179.878, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0420 17:08:42.186368 134928181425984 metric_logger.py:196] completed step: 1, seconds: 1.482, TFLOP/s/device: 9.168, Tokens/s/device: 1381.845, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0420 17:08:42.611034 134928181425984 metric_logger.py:196] completed step: 2, seconds: 0.040, TFLOP/s/device: 336.199, Tokens/s/device: 50675.508, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0420 17:08:42.758308 134928181425984 metric_logger.py:196] completed step: 3, seconds: 0.425, TFLOP/s/device: 31.994, Tokens/s/device: 4822.557, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0420 17:08:42.762967 134928181425984 checkpointing.py:794] Waiting for step 5 to finish before checkpoint... I0420 17:08:43.052031 134928181425984 checkpointing.py:798] Waited 0.2890164852142334 seconds for step 5 to finish before starting checkpointing. I0420 17:08:43.055069 134928181425984 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 17:08:45.538852 2817 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0420 17:08:48.231731 134798795716352 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/0/items/array_metadatas/process_2 I0420 17:08:53.067497 134800388470528 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 135.264 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.677883863449097 s) (per-host) I0420 17:08:53.067623 134800388470528 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.395509s. I0420 17:09:00.632581 134800388470528 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 18.960447s. I0420 17:09:00.632805 134798777857792 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 17:09:00.632866 134798777857792 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0420 17:09:00.632916 134798777857792 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0420 17:09:00.634656 134798777857792 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0420 17:09:00.634830 134928181425984 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0420 17:09:00.634965 134928181425984 checkpoint_manager.py:1452] Waiting for previous save to complete took 17.579892 seconds. If this number is high, consider checkpointing less frequently. I0420 17:09:00.636946 134928181425984 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 5 I0420 17:09:00.639440 134928181425984 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/5. I0420 17:09:01.360713 134928181425984 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 17:09:01.360807 134928181425984 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 I0420 17:09:01.402556 134928181425984 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.043118s I0420 17:09:01.402715 134928181425984 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 33.183 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.046486854553222656 s) (per-host) I0420 17:09:01.402773 134928181425984 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046553s (batch_requests_ready=0.001734s, total_serialization_initiated=0.044745s, others=0.000074s) I0420 17:09:01.402861 134928181425984 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.051485s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.051470) I0420 17:09:01.403534 134928181425984 event_tracking.py:125] [process=2] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/5. I0420 17:09:01.403897 134798777857792 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:29:01.403859 I0420 17:09:01.405854 134928181425984 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0420 17:09:01.406106 134799897392896 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0420 17:09:01.406228 134928181425984 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776704923.0550425, 'wait_for_prev_duration_secs': 17.57989239692688, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776704940.6369915, 'checkpointer_blocking_duration_secs': 0.7670536041259766, 'get_old_steps_start_time': 1776704941.404067, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776704923.0526683, 'checkpoint_manager_blocking_duration_secs': 18.35352396965027} I0420 17:09:01.406346 134928181425984 checkpointing.py:409] Started an asynchronous checkpoint save for step 5 I0420 17:09:01.407208 134928181425984 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.741, Tokens/s/device: 13376.004, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0420 17:09:01.413256 134928181425984 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.207, Tokens/s/device: 13898.409, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0420 17:09:07.620503 134798795716352 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/5/items/array_metadatas/process_2 I0420 17:09:12.488266 134798777857792 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 141.897 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 11.131978034973145 s) (per-host) I0420 17:09:12.488393 134798777857792 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 11.084383s. I0420 17:09:18.537051 134798777857792 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 17.133026s. I0420 17:09:18.537304 134799897392896 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 17:09:18.537361 134799897392896 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0420 17:09:18.537409 134799897392896 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0420 17:09:18.538986 134799897392896 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0420 17:09:23.988495 134928181425984 metric_logger.py:196] completed step: 6, seconds: 18.649, TFLOP/s/device: 0.729, Tokens/s/device: 109.819, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0420 17:09:24.135881 134928181425984 metric_logger.py:196] completed step: 7, seconds: 22.429, TFLOP/s/device: 0.606, Tokens/s/device: 91.310, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0420 17:09:24.283399 134928181425984 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.222, Tokens/s/device: 13448.557, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0420 17:09:24.430077 134928181425984 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0420 17:09:24.430756 134928181425984 checkpointing.py:798] Waited 0.0007042884826660156 seconds for step 9 to finish before starting checkpointing. I0420 17:09:24.432863 134928181425984 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 17:09:24.434440 134928181425984 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 9 I0420 17:09:24.435858 134928181425984 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/9. I0420 17:09:25.146455 134928181425984 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0420 17:09:25.146564 134928181425984 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 I0420 17:09:25.187221 134928181425984 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041641s I0420 17:09:25.187370 134928181425984 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 34.302 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.044969797134399414 s) (per-host) I0420 17:09:25.187418 134928181425984 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045028s (batch_requests_ready=0.001610s, total_serialization_initiated=0.043356s, others=0.000062s) I0420 17:09:25.187500 134928181425984 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049169s (all_items=0.000010s, per_item={'items': '0.00001025'}, temp_paths=0.049159) I0420 17:09:25.188274 134928181425984 event_tracking.py:125] [process=2] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/9. I0420 17:09:25.188587 134799897392896 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:29:25.188555 I0420 17:09:25.190543 134928181425984 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0420 17:09:25.190774 134799872214784 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0420 17:09:25.190877 134928181425984 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776704964.4328318, 'wait_for_prev_duration_secs': 7.224082946777344e-05, 'time_between_consecutive_saves_sec': 5.893805265426636, 'checkpointer_blocking_start_time': 1776704964.4344804, 'checkpointer_blocking_duration_secs': 0.7542543411254883, 'get_old_steps_start_time': 1776704965.1887524, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1776704964.430985, 'checkpoint_manager_blocking_duration_secs': 0.7598593235015869} I0420 17:09:25.191033 134928181425984 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0420 17:09:25.191083 134928181425984 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0420 17:09:30.964656 134799360521984 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_15_ocdbt_false/checkpoints/9/items/array_metadatas/process_2 I0420 17:09:36.000244 134799897392896 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 145.480 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 10.857802867889404 s) (per-host) I0420 17:09:36.000362 134799897392896 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 10.811659s. I0420 17:09:42.233566 134799897392896 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 17.044811s. I0420 17:09:42.233808 134799872214784 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0420 17:09:42.233872 134799872214784 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0420 17:09:42.233911 134799872214784 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0420 17:09:42.235421 134799872214784 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0420 17:09:42.235586 134928181425984 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0420 17:09:42.235741 134928181425984 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0420 17:09:42.236582 134928181425984 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.332, Tokens/s/device: 13917.298, 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: Mon Apr 20 17:09:54 UTC 2026 EXIT_CODE=0