main (8a17c3d19) vs feat/nnx-post-train-fixes (7f06c99ac)| Metric | main 8a17c3d19 | feat/nnx-post-train-fixes 7f06c99ac | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1750 | 8.1750 | 0 |
| TFLOP/s | 59.284 | 59.227 | -0.057 |
| Tok/s | 8935.9 | 8927.4 | -8.452 |
| Avg s/step | 3.709 | 3.617 | -0.092 |
| Memory % | 1.38 | 1.38 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Wed Apr 22 07:23:17 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-22 07:23:43.514870: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0422 07:23:43.727126 135487188207424 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 07:23:52.768780 135487188207424 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 07:23:52.771123 135487188207424 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-ie7hg-slice-job-0-0.mt-04-int8-ie7hg:8482 I0422 07:23:53.880871 135487188207424 max_utils.py:284] Jax distributed system initialized! I0422 07:23:58.964579 135487188207424 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 07:23:58.964684 135487188207424 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 07:23:58.964741 135487188207424 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0422 07:23:58.964790 135487188207424 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 07:23:59.685686 135487188207424 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 07:23:59.686122 135487188207424 checkpointing.py:677] Setting up checkpoint logger... I0422 07:23:59.686185 135487188207424 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 07:23:59.686231 135487188207424 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 07:23:59.686559 135487188207424 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 0x7b38f3b18e30>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 07:24:02.950927 135487188207424 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 07:24:02.951169 135487188207424 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 0x7b2558709f70>}, handler_registry=None I0422 07:24:02.951404 135487188207424 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b2558709f70>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0422 07:24:02.951495 135487188207424 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b255870d910>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0422 07:24:02.951534 135487188207424 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 0x7b2558709f70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b2558709f70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b255870d910>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b255870d910>}). I0422 07:24:02.951875 135487188207424 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 07:24:02.951946 135487188207424 async_checkpointer.py:192] [process=2][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b25584bd760> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 07:24:03.668604 135487188207424 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints I0422 07:24:04.283757 135487188207424 checkpoint_manager.py:929] [process=2][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b255870b2f0> I0422 07:24:04.283928 135487188207424 checkpointing.py:301] Checkpoint manager created! I0422 07:24:05.423692 135487188207424 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 07:24:05.423830 135487188207424 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:24:05.811633 135487188207424 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:24:05.811736 135487188207424 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:24:05.828575 135487188207424 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:24:05.828636 135487188207424 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:24:05.894447 135487188207424 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:24:05.894536 135487188207424 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:24:05.911581 135487188207424 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:24:05.911645 135487188207424 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:24:05.934474 135487188207424 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:24:05.934579 135487188207424 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:24:05.960796 135487188207424 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 07:24:05.960866 135487188207424 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:24:06.027197 135487188207424 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 07:24:06.027284 135487188207424 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 07:24:06.536197 135487188207424 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 07:24:06.536308 135487188207424 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0422 07:24:09.007431 135487188207424 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 07:24:09.007565 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 07:24:09.007620 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 07:24:09.007680 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 07:24:09.007766 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 07:24:09.007817 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 07:24:09.007875 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 07:24:09.007932 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0422 07:24:09.007973 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0422 07:24:09.008010 135487188207424 maxtext_utils.py:1668] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 07:24:09.008059 135487188207424 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 07:24:09.008106 135487188207424 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 07:24:10.403888 135487188207424 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:24:10.403984 135487188207424 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 07:24:10.419726 135487188207424 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:24:10.419787 135487188207424 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 07:24:24.314273 135487188207424 max_utils.py:791] Total memory size: 1.8 GB, Output size: 0.4 GB, Temp size: 1.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 07:24:24.315144 135487188207424 metric_logger.py:301] number parameters: 1.104 billion I0422 07:24:38.662400 135487188207424 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 07:24:38.900667 135487188207424 checkpointing.py:776] Waited 0.23824763298034668 seconds for step 0 to finish before starting checkpointing. I0422 07:24:38.903410 135487188207424 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:24:38.905352 135487188207424 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 0 I0422 07:24:38.906785 135487188207424 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints/0. I0422 07:24:40.048319 135487188207424 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 07:24:40.049300 135487188207424 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:24:40.049355 135487188207424 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0422 07:24:40.332577 135487188207424 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.284330s I0422 07:24:40.332767 135487188207424 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 5.320 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28995275497436523 s) (per-host) I0422 07:24:40.332826 135487188207424 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.290025s (batch_requests_ready=0.002370s, total_serialization_initiated=0.287573s, others=0.000081s) I0422 07:24:40.332923 135487188207424 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.294079s (all_items=0.000020s, per_item={'items': '0.00001979'}, temp_paths=0.294059) I0422 07:24:40.333761 135487188207424 event_tracking.py:125] [process=2] [async] Finished blocking save in 1.43 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints/0. I0422 07:24:40.334065 135358056601344 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:44:40.334031 I0422 07:24:40.340101 135487188207424 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:24:40.340419 135357525903104 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:24:40.340584 135487188207424 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776842678.9033911, 'wait_for_prev_duration_secs': 6.556510925292969e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776842678.9053926, 'checkpointer_blocking_duration_secs': 1.428823709487915, 'get_old_steps_start_time': 1776842680.334247, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776842678.9014575, 'checkpoint_manager_blocking_duration_secs': 1.4390859603881836} I0422 07:24:40.340690 135487188207424 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 07:24:40.340752 135487188207424 max_utils.py:750] Memstats: After params initialized: I0422 07:24:40.340805 135487188207424 max_utils.py:756] Using (GB) 0.44 / 31.25 (1.408000%) on TPU_8(process=2,(0,2,0,0)) I0422 07:24:40.340838 135487188207424 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_9(process=2,(1,2,0,0)) I0422 07:24:40.340866 135487188207424 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_12(process=2,(0,3,0,0)) I0422 07:24:40.340893 135487188207424 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_13(process=2,(1,3,0,0)) I0422 07:24:40.657872 135487188207424 metric_logger.py:196] completed step: 0, seconds: 14.347, TFLOP/s/device: 0.947, Tokens/s/device: 142.746, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0422 07:24:40.914981 135487188207424 metric_logger.py:196] completed step: 1, seconds: 1.994, TFLOP/s/device: 6.815, Tokens/s/device: 1027.199, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0422 07:24:41.332946 135487188207424 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 479.484, Tokens/s/device: 72273.000, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820 I0422 07:24:41.562482 135487188207424 metric_logger.py:196] completed step: 3, seconds: 0.419, TFLOP/s/device: 32.464, Tokens/s/device: 4893.305, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199 I0422 07:24:42.021328 135487188207424 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.859, Tokens/s/device: 8721.128, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531 I0422 07:24:42.027689 135487188207424 metric_logger.py:196] completed step: 5, seconds: 0.229, TFLOP/s/device: 59.254, Tokens/s/device: 8931.414, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941 I0422 07:24:44.592485 2849 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 07:24:46.676018 135365036300032 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints/0/items/array_metadatas/process_2 I0422 07:25:11.812191 135487188207424 metric_logger.py:196] completed step: 6, seconds: 0.459, TFLOP/s/device: 29.573, Tokens/s/device: 4457.494, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468 I0422 07:25:12.041544 135487188207424 metric_logger.py:196] completed step: 7, seconds: 29.557, TFLOP/s/device: 0.460, Tokens/s/device: 69.291, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779 I0422 07:25:12.271075 135487188207424 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.071, Tokens/s/device: 8753.034, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453 I0422 07:25:12.499887 135487188207424 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 07:25:12.500603 135487188207424 checkpointing.py:776] Waited 0.0007374286651611328 seconds for step 9 to finish before starting checkpointing. I0422 07:25:12.502832 135487188207424 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:25:15.040786 135358056601344 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 45.134 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.99793553352356 s) (per-host) I0422 07:25:15.040910 135358056601344 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 34.706732s. I0422 07:25:23.092803 135358056601344 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 42.758609s. I0422 07:25:23.093081 135357525903104 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:25:23.093200 135357525903104 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:25:23.093251 135357525903104 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:25:23.094972 135357525903104 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 07:25:23.095142 135487188207424 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0422 07:25:23.095278 135487188207424 checkpoint_manager.py:1452] Waiting for previous save to complete took 10.592446 seconds. If this number is high, consider checkpointing less frequently. I0422 07:25:23.097186 135487188207424 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 9 I0422 07:25:23.099190 135487188207424 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints/9. I0422 07:25:23.392748 135487188207424 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:25:23.392839 135487188207424 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0422 07:25:23.427655 135487188207424 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036013s I0422 07:25:23.427800 135487188207424 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 39.087 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.039464712142944336 s) (per-host) I0422 07:25:23.427846 135487188207424 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.039521s (batch_requests_ready=0.001799s, total_serialization_initiated=0.037659s, others=0.000062s) I0422 07:25:23.427926 135487188207424 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043551s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.043535) I0422 07:25:23.428573 135487188207424 event_tracking.py:125] [process=2] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints/9. I0422 07:25:23.428848 135357525903104 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:45:23.428825 I0422 07:25:23.430757 135487188207424 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:25:23.431042 135365036300032 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:25:23.431206 135487188207424 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776842712.5028007, 'wait_for_prev_duration_secs': 10.592446088790894, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776842723.0972252, 'checkpointer_blocking_duration_secs': 0.33171844482421875, 'get_old_steps_start_time': 1776842723.428967, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776842712.500846, 'checkpoint_manager_blocking_duration_secs': 10.930325746536255} I0422 07:25:23.431324 135487188207424 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 07:25:23.431370 135487188207424 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:25:28.558007 135343154247424 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_04_int8/checkpoints/9/items/array_metadatas/process_2 I0422 07:26:05.526808 135357525903104 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 37.486 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.138450622558594 s) (per-host) I0422 07:26:05.526915 135357525903104 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.097994s. I0422 07:26:14.418445 135357525903104 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 50.989507s. I0422 07:26:14.418766 135365036300032 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:26:14.418891 135365036300032 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:26:14.418946 135365036300032 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:26:14.420670 135365036300032 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 07:26:14.420861 135487188207424 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 07:26:14.421008 135487188207424 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:26:14.421907 135487188207424 metric_logger.py:196] completed step: 9, seconds: 0.229, TFLOP/s/device: 59.284, Tokens/s/device: 8935.856, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3552.333 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 07:26:22 UTC 2026 EXIT_CODE=0
XPK Start: Wed Apr 22 12:50:23 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-22 12:50:48.526211: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0422 12:50:48.732642 133194166814528 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 12:50:57.773684 133194166814528 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 12:50:57.776007 133194166814528 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-zctjs-slice-job-0-0.mt-04-int8-zctjs:8482 I0422 12:50:58.992321 133194166814528 max_utils.py:284] Jax distributed system initialized! I0422 12:51:05.347939 133194166814528 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 12:51:05.348050 133194166814528 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 12:51:05.348093 133194166814528 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0422 12:51:05.348128 133194166814528 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 12:51:06.041412 133194166814528 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 12:51:06.042002 133194166814528 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 12:51:06.042303 133194166814528 checkpointing.py:688] Setting up checkpoint logger... I0422 12:51:06.042356 133194166814528 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 12:51:06.042398 133194166814528 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 12:51:06.042740 133194166814528 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 0x792318118800>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 12:51:09.416418 133194166814528 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0422 12:51:09.416673 133194166814528 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x790ec863e360>}, handler_registry=None I0422 12:51:09.416908 133194166814528 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x790ec863e360>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0422 12:51:09.416975 133194166814528 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x790ec863e4e0>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0422 12:51:09.417012 133194166814528 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 0x790ec863e360>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x790ec863e360>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x790ec863e4e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x790ec863e4e0>}). I0422 12:51:09.417349 133194166814528 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 12:51:09.417419 133194166814528 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x790ec83d5e40> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 12:51:10.491106 133194166814528 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints I0422 12:51:10.557390 133194166814528 checkpoint_manager.py:929] [process=5][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x790ec863fb00> I0422 12:51:10.557555 133194166814528 checkpointing.py:302] Checkpoint manager created! I0422 12:51:11.701959 133194166814528 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 12:51:11.702066 133194166814528 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 12:51:12.096251 133194166814528 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 12:51:12.096358 133194166814528 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 12:51:12.112887 133194166814528 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 12:51:12.112944 133194166814528 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 12:51:12.178255 133194166814528 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 12:51:12.178347 133194166814528 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 12:51:12.194733 133194166814528 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 12:51:12.194792 133194166814528 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 12:51:12.216723 133194166814528 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 12:51:12.216827 133194166814528 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 12:51:12.242121 133194166814528 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 12:51:12.242195 133194166814528 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 12:51:12.307555 133194166814528 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 12:51:12.307638 133194166814528 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 12:51:12.823006 133194166814528 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0422 12:51:12.823115 133194166814528 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0422 12:51:15.272471 133194166814528 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 12:51:15.272597 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 12:51:15.272669 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 12:51:15.272757 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 12:51:15.272827 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 12:51:15.272889 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 12:51:15.272960 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 12:51:15.273033 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0422 12:51:15.273085 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0422 12:51:15.273133 133194166814528 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 12:51:15.273194 133194166814528 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 12:51:15.273247 133194166814528 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 12:51:16.674005 133194166814528 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 12:51:16.674105 133194166814528 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 12:51:16.689506 133194166814528 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 12:51:16.689569 133194166814528 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 12:51:30.685475 133194166814528 max_utils.py:791] Total memory size: 1.8 GB, Output size: 0.4 GB, Temp size: 1.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 12:51:30.686287 133194166814528 metric_logger.py:301] number parameters: 1.104 billion I0422 12:51:45.113431 133194166814528 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0422 12:51:45.356731 133194166814528 checkpointing.py:798] Waited 0.24328064918518066 seconds for step 0 to finish before starting checkpointing. I0422 12:51:45.359234 133194166814528 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 12:51:45.361197 133194166814528 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0422 12:51:45.362567 133194166814528 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints/0. I0422 12:51:45.772588 133194166814528 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 12:51:45.773500 133194166814528 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 12:51:45.773555 133194166814528 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0422 12:51:46.049521 133194166814528 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.277007s I0422 12:51:46.049696 133194166814528 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.460 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2825305461883545 s) (per-host) I0422 12:51:46.049752 133194166814528 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.282597s (batch_requests_ready=0.002321s, total_serialization_initiated=0.280202s, others=0.000074s) I0422 12:51:46.049845 133194166814528 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287220s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.287204) I0422 12:51:46.050698 133194166814528 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.69 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints/0. I0422 12:51:46.051055 133061993928448 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 13:11:46.051016 I0422 12:51:46.061206 133194166814528 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 12:51:46.061478 133061463230208 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0422 12:51:46.061630 133194166814528 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776862305.359215, 'wait_for_prev_duration_secs': 6.175041198730469e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776862305.3612413, 'checkpointer_blocking_duration_secs': 0.6899769306182861, 'get_old_steps_start_time': 1776862306.0512419, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776862305.3572862, 'checkpoint_manager_blocking_duration_secs': 0.7043066024780273} I0422 12:51:46.061756 133194166814528 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0422 12:51:46.061805 133194166814528 max_utils.py:750] Memstats: After params initialized: I0422 12:51:46.061853 133194166814528 max_utils.py:756] Using (GB) 0.44 / 31.25 (1.408000%) on TPU_18(process=5,(2,4,0,0)) I0422 12:51:46.061885 133194166814528 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0422 12:51:46.061913 133194166814528 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0422 12:51:46.061939 133194166814528 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0422 12:51:46.383316 133194166814528 metric_logger.py:196] completed step: 0, seconds: 14.427, TFLOP/s/device: 0.942, Tokens/s/device: 141.956, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0422 12:51:46.643260 133194166814528 metric_logger.py:196] completed step: 1, seconds: 1.268, TFLOP/s/device: 10.713, Tokens/s/device: 1614.771, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277 I0422 12:51:47.066816 133194166814528 metric_logger.py:196] completed step: 2, seconds: 0.031, TFLOP/s/device: 437.617, Tokens/s/device: 65962.381, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820 I0422 12:51:47.296531 133194166814528 metric_logger.py:196] completed step: 3, seconds: 0.424, TFLOP/s/device: 32.040, Tokens/s/device: 4829.482, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199 I0422 12:51:47.755840 133194166814528 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.855, Tokens/s/device: 8720.534, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531 I0422 12:51:47.761998 133194166814528 metric_logger.py:196] completed step: 5, seconds: 0.230, TFLOP/s/device: 59.185, Tokens/s/device: 8921.026, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941 I0422 12:51:49.391613 2923 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 12:51:51.348740 133061471622912 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints/0/items/array_metadatas/process_5 I0422 12:52:17.430011 133194166814528 metric_logger.py:196] completed step: 6, seconds: 0.460, TFLOP/s/device: 29.561, Tokens/s/device: 4455.826, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468 I0422 12:52:17.659606 133194166814528 metric_logger.py:196] completed step: 7, seconds: 29.440, TFLOP/s/device: 0.462, Tokens/s/device: 69.565, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779 I0422 12:52:17.889093 133194166814528 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.151, Tokens/s/device: 8765.210, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453 I0422 12:52:18.119114 133194166814528 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0422 12:52:18.119753 133194166814528 checkpointing.py:798] Waited 0.0006611347198486328 seconds for step 9 to finish before starting checkpointing. I0422 12:52:18.121873 133194166814528 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 12:52:20.023419 133061993928448 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 46.111 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.25619101524353 s) (per-host) I0422 12:52:20.023543 133061993928448 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.972372s. I0422 12:52:25.429252 133061993928448 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 39.378063s. I0422 12:52:25.429547 133061463230208 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 12:52:25.429676 133061463230208 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0422 12:52:25.429727 133061463230208 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 12:52:25.431255 133061463230208 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 12:52:25.431441 133194166814528 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0422 12:52:25.431578 133194166814528 checkpoint_manager.py:1452] Waiting for previous save to complete took 7.309704 seconds. If this number is high, consider checkpointing less frequently. I0422 12:52:25.433495 133194166814528 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0422 12:52:25.435606 133194166814528 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints/9. I0422 12:52:25.744859 133194166814528 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 12:52:25.744953 133194166814528 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0422 12:52:25.776350 133194166814528 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.033437s I0422 12:52:25.776519 133194166814528 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 41.587 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03709244728088379 s) (per-host) I0422 12:52:25.776571 133194166814528 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.037155s (batch_requests_ready=0.001938s, total_serialization_initiated=0.035146s, others=0.000070s) I0422 12:52:25.776685 133194166814528 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.041142s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=0.041127) I0422 12:52:25.777405 133194166814528 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints/9. I0422 12:52:25.777783 133061463230208 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 13:12:25.777745 I0422 12:52:25.781672 133194166814528 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 12:52:25.781954 133061471622912 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0422 12:52:25.782132 133194166814528 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776862338.1218424, 'wait_for_prev_duration_secs': 7.309704065322876, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776862345.4335337, 'checkpointer_blocking_duration_secs': 0.3443934917449951, 'get_old_steps_start_time': 1776862345.7779486, 'get_old_steps_duration_secs': 3.075599670410156e-05, 'checkpoint_manager_blocking_start_time': 1776862338.1199849, 'checkpoint_manager_blocking_duration_secs': 7.662106990814209} I0422 12:52:25.782266 133194166814528 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0422 12:52:25.782315 133194166814528 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 12:52:31.293172 133038495196928 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_04_int8/checkpoints/9/items/array_metadatas/process_5 I0422 12:53:07.593601 133061463230208 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.741 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.854135513305664 s) (per-host) I0422 12:53:07.593747 133061463230208 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.815847s. I0422 12:53:14.280261 133061463230208 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 48.502347s. I0422 12:53:14.280553 133061471622912 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 12:53:14.280691 133061471622912 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0422 12:53:14.280739 133061471622912 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 12:53:14.282923 133061471622912 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 12:53:14.283095 133194166814528 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 12:53:14.283241 133194166814528 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 12:53:14.284205 133194166814528 metric_logger.py:196] completed step: 9, seconds: 0.229, TFLOP/s/device: 59.227, Tokens/s/device: 8927.404, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3552.333 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 12:53:23 UTC 2026 EXIT_CODE=0