main (59e0f1759) vs feat/nnx-trainstate-and-training-loop (77202a8e9)| Metric | main 59e0f1759 | feat/nnx-trainstate-and-training-loop 77202a8e9 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 9.4570 | 9.4570 | 0 |
| TFLOP/s | 93.994 | 94.010 | +0.016 |
| Tok/s | 14167.9 | 14170.2 | +2.328 |
| Avg s/step | 6.862 | 6.838 | -0.024 |
| 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: Sat Apr 25 07:27:00 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. 2026-04-25 07:27:24.860433: 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) I0425 07:27:25.070381 140069125056320 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 07:27:34.112831 140069125056320 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 07:27:34.115196 140069125056320 distributed.py:172] Connecting to JAX distributed service on mt-06-grad-accum-kirf6-slice-job-0-0.mt-06-grad-accum-kirf6:8482 I0425 07:27:35.452751 140069125056320 max_utils.py:284] Jax distributed system initialized! I0425 07:27:41.567178 140069125056320 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 07:27:41.567285 140069125056320 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 07:27:41.567327 140069125056320 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 I0425 07:27:41.567365 140069125056320 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 07:27:42.260718 140069125056320 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 07:27:42.261002 140069125056320 checkpointing.py:677] Setting up checkpoint logger... I0425 07:27:42.261059 140069125056320 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 07:27:42.261102 140069125056320 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 07:27:42.261450 140069125056320 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 0x7f63e33ee6c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 07:27:45.165586 140069125056320 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0425 07:27:45.165836 140069125056320 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 0x7f4f6436df40>}, handler_registry=None I0425 07:27:45.166073 140069125056320 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f4f6436df40>` 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`. I0425 07:27:45.166122 140069125056320 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f4f64371280>` 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`. I0425 07:27:45.166160 140069125056320 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 0x7f4f6436df40>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f4f6436df40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f4f64371280>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f4f64371280>}). I0425 07:27:45.166499 140069125056320 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 07:27:45.166568 140069125056320 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f4e9c491800> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 07:27:45.862612 140069125056320 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints I0425 07:27:45.881971 140069125056320 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_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f4f6436f0e0> I0425 07:27:45.882104 140069125056320 checkpointing.py:301] Checkpoint manager created! I0425 07:27:47.608313 140069125056320 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 07:27:47.608433 140069125056320 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 07:27:47.986923 140069125056320 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0425 07:27:47.987025 140069125056320 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 07:27:48.003697 140069125056320 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0425 07:27:48.003755 140069125056320 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 07:27:48.027830 140069125056320 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 07:27:48.027898 140069125056320 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 07:27:48.044473 140069125056320 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 07:27:48.044535 140069125056320 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 07:27:48.061157 140069125056320 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 07:27:48.061220 140069125056320 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 07:27:48.086284 140069125056320 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0425 07:27:48.086354 140069125056320 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 07:27:48.107567 140069125056320 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 07:27:48.107634 140069125056320 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 07:27:48.318704 140069125056320 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0425 07:27:48.318811 140069125056320 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 07:27:49.741478 140069125056320 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 07:27:49.741606 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0425 07:27:49.741671 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0425 07:27:49.741732 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0425 07:27:49.741784 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0425 07:27:49.741828 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0425 07:27:49.741881 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0425 07:27:49.741931 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0425 07:27:49.741969 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0425 07:27:49.742004 140069125056320 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0425 07:27:49.742051 140069125056320 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 07:27:49.742098 140069125056320 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 07:27:49.765508 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0425 07:27:49.765574 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0425 07:27:49.780780 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0425 07:27:49.780836 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0425 07:27:49.810708 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0425 07:27:49.810774 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0425 07:27:49.825822 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0425 07:27:49.825879 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0425 07:27:49.856085 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0425 07:27:49.856148 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0425 07:27:49.871161 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0425 07:27:49.871222 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0425 07:27:49.916074 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0425 07:27:49.916146 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0425 07:27:49.931321 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0425 07:27:49.931376 140069125056320 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0425 07:27:50.626265 140069125056320 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 07:27:50.626361 140069125056320 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0425 07:27:50.642061 140069125056320 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 07:27:50.642123 140069125056320 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0425 07:28:02.097687 140069125056320 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0425 07:28:02.098491 140069125056320 metric_logger.py:301] number parameters: 1.104 billion I0425 07:28:03.976206 140069125056320 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0425 07:28:15.045052 140069125056320 checkpointing.py:776] Waited 11.06882095336914 seconds for step 0 to finish before starting checkpointing. I0425 07:28:15.047645 140069125056320 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 07:28:15.049352 140069125056320 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0425 07:28:15.050816 140069125056320 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints/0. I0425 07:28:15.369734 140069125056320 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 07:28:15.370630 140069125056320 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 07:28:15.370702 140069125056320 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 I0425 07:28:15.644348 140069125056320 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274709s I0425 07:28:15.644526 140069125056320 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.508 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28008270263671875 s) (per-host) I0425 07:28:15.644579 140069125056320 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.280146s (batch_requests_ready=0.002219s, total_serialization_initiated=0.277855s, others=0.000072s) I0425 07:28:15.644683 140069125056320 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.284317s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.284298) I0425 07:28:15.645426 140069125056320 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.60 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints/0. I0425 07:28:15.645771 139941330880256 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 07:48:15.645732 I0425 07:28:15.655091 140069125056320 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 07:28:15.655345 139939171550976 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0425 07:28:15.655467 140069125056320 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777102095.047627, 'wait_for_prev_duration_secs': 7.653236389160156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777102095.0493925, 'checkpointer_blocking_duration_secs': 0.5965237617492676, 'get_old_steps_start_time': 1777102095.6459415, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1777102095.0456178, 'checkpoint_manager_blocking_duration_secs': 0.6098098754882812} I0425 07:28:15.655572 140069125056320 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0425 07:28:15.655621 140069125056320 max_utils.py:750] Memstats: After params initialized: I0425 07:28:15.655681 140069125056320 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0)) I0425 07:28:15.655715 140069125056320 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0425 07:28:15.655743 140069125056320 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0425 07:28:15.655768 140069125056320 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0425 07:28:15.965520 140069125056320 metric_logger.py:196] completed step: 0, seconds: 1.878, TFLOP/s/device: 28.946, Tokens/s/device: 4362.991, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0425 07:28:16.573669 140069125056320 metric_logger.py:196] completed step: 1, seconds: 11.988, TFLOP/s/device: 4.534, Tokens/s/device: 683.362, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0425 07:28:17.151841 140069125056320 metric_logger.py:196] completed step: 2, seconds: 0.030, TFLOP/s/device: 1812.403, Tokens/s/device: 273185.047, total_weights: 262144, loss: 10.563, lm_loss: 10.563, perplexity: 38662.707 I0425 07:28:17.730170 140069125056320 metric_logger.py:196] completed step: 3, seconds: 0.584, TFLOP/s/device: 92.985, Tokens/s/device: 14015.733, total_weights: 262144, loss: 10.272, lm_loss: 10.272, perplexity: 28909.668 I0425 07:28:18.570502 2583 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 07:28:18.887206 140069125056320 metric_logger.py:196] completed step: 4, seconds: 0.578, TFLOP/s/device: 94.064, Tokens/s/device: 14178.358, total_weights: 262144, loss: 10.022, lm_loss: 10.022, perplexity: 22524.992 I0425 07:28:18.895367 140069125056320 metric_logger.py:196] completed step: 5, seconds: 0.578, TFLOP/s/device: 93.974, Tokens/s/device: 14164.825, total_weights: 262144, loss: 9.820, lm_loss: 9.820, perplexity: 18401.865 I0425 07:28:20.674853 139940783597312 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints/0/items/array_metadatas/process_5 I0425 07:28:52.650055 139941330880256 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 42.365 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.28557515144348 s) (per-host) I0425 07:28:52.650183 139941330880256 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.004303s. I0425 07:29:01.887714 139941330880256 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 46.241812s. I0425 07:29:01.887964 139939171550976 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 07:29:01.888024 139939171550976 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0425 07:29:01.888087 139939171550976 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 07:29:01.890399 139939171550976 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 07:29:05.156389 140069125056320 metric_logger.py:196] completed step: 6, seconds: 1.159, TFLOP/s/device: 46.881, Tokens/s/device: 7066.364, total_weights: 262144, loss: 9.667, lm_loss: 9.667, perplexity: 15787.604 I0425 07:29:05.734675 140069125056320 metric_logger.py:196] completed step: 7, seconds: 45.684, TFLOP/s/device: 1.190, Tokens/s/device: 179.319, total_weights: 262144, loss: 9.561, lm_loss: 9.561, perplexity: 14203.827 I0425 07:29:06.312846 140069125056320 metric_logger.py:196] completed step: 8, seconds: 0.583, TFLOP/s/device: 93.219, Tokens/s/device: 14050.928, total_weights: 262144, loss: 9.496, lm_loss: 9.496, perplexity: 13302.920 I0425 07:29:06.890474 140069125056320 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0425 07:29:06.891149 140069125056320 checkpointing.py:776] Waited 0.0006966590881347656 seconds for step 9 to finish before starting checkpointing. I0425 07:29:06.893331 140069125056320 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 07:29:06.895239 140069125056320 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0425 07:29:06.896613 140069125056320 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints/9. I0425 07:29:07.183225 140069125056320 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 07:29:07.183323 140069125056320 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 I0425 07:29:07.218246 140069125056320 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035933s I0425 07:29:07.218429 140069125056320 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 38.521 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04004502296447754 s) (per-host) I0425 07:29:07.218484 140069125056320 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040111s (batch_requests_ready=0.001673s, total_serialization_initiated=0.038362s, others=0.000075s) I0425 07:29:07.218579 140069125056320 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044058s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.044043) I0425 07:29:07.219251 140069125056320 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints/9. I0425 07:29:07.219574 139945083639552 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 07:49:07.219536 I0425 07:29:07.221423 140069125056320 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0425 07:29:07.221682 139940826146560 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0425 07:29:07.221802 140069125056320 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777102146.8932981, 'wait_for_prev_duration_secs': 7.319450378417969e-05, 'time_between_consecutive_saves_sec': 5.002856969833374, 'checkpointer_blocking_start_time': 1777102146.8952801, 'checkpointer_blocking_duration_secs': 0.3244054317474365, 'get_old_steps_start_time': 1777102147.2197084, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1777102146.8913836, 'checkpoint_manager_blocking_duration_secs': 0.3303844928741455} I0425 07:29:07.221973 140069125056320 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0425 07:29:07.222021 140069125056320 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 07:29:13.123493 139940834539264 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_06_grad_accum/checkpoints/9/items/array_metadatas/process_5 I0425 07:29:50.615292 139945083639552 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 36.365 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.43687438964844 s) (per-host) I0425 07:29:50.615414 139945083639552 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.395215s. I0425 07:29:58.247735 139945083639552 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 51.027520s. I0425 07:29:58.248032 139940826146560 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 07:29:58.248148 139940826146560 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0425 07:29:58.248193 139940826146560 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0425 07:29:58.249879 139940826146560 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0425 07:29:58.250053 140069125056320 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. I0425 07:29:58.250195 140069125056320 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 07:29:58.251155 140069125056320 metric_logger.py:196] completed step: 9, seconds: 0.578, TFLOP/s/device: 93.994, Tokens/s/device: 14167.863, total_weights: 262144, loss: 9.457, lm_loss: 9.457, perplexity: 12802.546 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sat Apr 25 07:30:10 UTC 2026 EXIT_CODE=0
77202a8e9
· feat_nnx_trainstate_and_training_loop_20260425_092114
· full log
XPK Start: Sat Apr 25 09:33:13 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. 2026-04-25 09:33:38.032044: 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) I0425 09:33:38.241962 136901049722688 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 09:33:47.281324 136901049722688 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 09:33:47.283666 136901049722688 distributed.py:172] Connecting to JAX distributed service on mt-06-grad-accum-v1qwa-slice-job-0-0.mt-06-grad-accum-v1qwa:8482 I0425 09:33:48.860798 136901049722688 max_utils.py:284] Jax distributed system initialized! I0425 09:33:54.967457 136901049722688 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 09:33:54.967565 136901049722688 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 09:33:54.967605 136901049722688 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 I0425 09:33:54.967642 136901049722688 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 09:33:55.675110 136901049722688 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 09:33:55.675693 136901049722688 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 09:33:55.675874 136901049722688 checkpointing.py:688] Setting up checkpoint logger... I0425 09:33:55.675939 136901049722688 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 09:33:55.675985 136901049722688 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 09:33:55.676321 136901049722688 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 0x7c820ac4c680>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 09:33:58.525344 136901049722688 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0425 09:33:58.525541 136901049722688 checkpoint_manager.py:708] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c6dd433fbc0>}, handler_registry=None I0425 09:33:58.525774 136901049722688 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c6dd433fbc0>` 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`. I0425 09:33:58.525838 136901049722688 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c6dd4340e60>` 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`. I0425 09:33:58.525884 136901049722688 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 0x7c6dd433fbc0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c6dd433fbc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c6dd4340e60>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c6dd4340e60>}). I0425 09:33:58.526220 136901049722688 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 09:33:58.526291 136901049722688 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c6d046d9e40> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 09:34:00.077852 136901049722688 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints I0425 09:34:00.624110 136901049722688 checkpoint_manager.py:929] [process=3][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c6dd4341700> I0425 09:34:00.624282 136901049722688 checkpointing.py:302] Checkpoint manager created! I0425 09:34:02.336354 136901049722688 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 09:34:02.336460 136901049722688 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 09:34:02.715470 136901049722688 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 09:34:02.715570 136901049722688 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 09:34:02.738723 136901049722688 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 09:34:02.738851 136901049722688 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 09:34:02.763176 136901049722688 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:34:02.763249 136901049722688 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 09:34:02.780028 136901049722688 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:34:02.780089 136901049722688 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 09:34:02.796943 136901049722688 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:34:02.797002 136901049722688 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 09:34:02.821874 136901049722688 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 09:34:02.821956 136901049722688 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 09:34:02.843070 136901049722688 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 09:34:02.843136 136901049722688 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 09:34:03.060891 136901049722688 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0425 09:34:03.061013 136901049722688 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 09:34:04.490966 136901049722688 maxtext_utils.py:1874] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 09:34:04.491092 136901049722688 maxtext_utils.py:1874] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0425 09:34:04.491146 136901049722688 maxtext_utils.py:1874] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0425 09:34:04.491200 136901049722688 maxtext_utils.py:1874] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0425 09:34:04.491248 136901049722688 maxtext_utils.py:1874] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0425 09:34:04.491284 136901049722688 maxtext_utils.py:1874] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0425 09:34:04.491337 136901049722688 maxtext_utils.py:1874] 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) I0425 09:34:04.491391 136901049722688 maxtext_utils.py:1874] 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') I0425 09:34:04.491437 136901049722688 maxtext_utils.py:1874] 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) I0425 09:34:04.491476 136901049722688 maxtext_utils.py:1874] 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) I0425 09:34:04.491523 136901049722688 maxtext_utils.py:1874] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 09:34:04.491575 136901049722688 maxtext_utils.py:1874] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 09:34:04.520333 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0425 09:34:04.520434 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0425 09:34:04.535513 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0425 09:34:04.535573 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0425 09:34:04.565349 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0425 09:34:04.565415 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0425 09:34:04.580327 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0425 09:34:04.580386 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0425 09:34:04.610255 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0425 09:34:04.610326 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0425 09:34:04.625235 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0425 09:34:04.625295 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0425 09:34:04.669950 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0425 09:34:04.670021 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0425 09:34:04.684962 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0425 09:34:04.685020 136901049722688 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0425 09:34:05.357256 136901049722688 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 09:34:05.357351 136901049722688 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0425 09:34:05.373295 136901049722688 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 09:34:05.373356 136901049722688 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0425 09:34:16.646212 136901049722688 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0425 09:34:16.646996 136901049722688 metric_logger.py:301] number parameters: 1.104 billion I0425 09:34:18.524713 136901049722688 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0425 09:34:29.782969 136901049722688 checkpointing.py:798] Waited 11.258214473724365 seconds for step 0 to finish before starting checkpointing. I0425 09:34:29.785228 136901049722688 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 09:34:29.787235 136901049722688 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0 I0425 09:34:29.789632 136901049722688 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints/0. I0425 09:34:30.396456 136901049722688 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 09:34:30.397541 136901049722688 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 09:34:30.397602 136901049722688 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 I0425 09:34:30.671117 136901049722688 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274737s I0425 09:34:30.671286 136901049722688 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 5.502 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2803456783294678 s) (per-host) I0425 09:34:30.671339 136901049722688 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.280409s (batch_requests_ready=0.002203s, total_serialization_initiated=0.278136s, others=0.000071s) I0425 09:34:30.671430 136901049722688 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.284344s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.284328) I0425 09:34:30.673835 136901049722688 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.89 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints/0. I0425 09:34:30.674192 136772889601792 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 09:54:30.674153 I0425 09:34:30.680567 136901049722688 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 09:34:30.680842 136772347193088 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0425 09:34:30.680994 136901049722688 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777109669.7852085, 'wait_for_prev_duration_secs': 6.222724914550781e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777109669.7872741, 'checkpointer_blocking_duration_secs': 0.8870739936828613, 'get_old_steps_start_time': 1777109670.6743731, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1777109669.7834842, 'checkpoint_manager_blocking_duration_secs': 0.8974699974060059} I0425 09:34:30.681147 136901049722688 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0425 09:34:30.681200 136901049722688 max_utils.py:750] Memstats: After params initialized: I0425 09:34:30.681250 136901049722688 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0)) I0425 09:34:30.681285 136901049722688 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0)) I0425 09:34:30.681312 136901049722688 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0)) I0425 09:34:30.681337 136901049722688 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0)) I0425 09:34:30.995998 136901049722688 metric_logger.py:196] completed step: 0, seconds: 1.878, TFLOP/s/device: 28.946, Tokens/s/device: 4363.014, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0425 09:34:31.618233 136901049722688 metric_logger.py:196] completed step: 1, seconds: 12.470, TFLOP/s/device: 4.358, Tokens/s/device: 656.947, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0425 09:34:32.196435 136901049722688 metric_logger.py:196] completed step: 2, seconds: 0.044, TFLOP/s/device: 1230.579, Tokens/s/device: 185486.245, total_weights: 262144, loss: 10.563, lm_loss: 10.563, perplexity: 38662.707 I0425 09:34:32.774601 136901049722688 metric_logger.py:196] completed step: 3, seconds: 0.584, TFLOP/s/device: 93.062, Tokens/s/device: 14027.373, total_weights: 262144, loss: 10.272, lm_loss: 10.272, perplexity: 28909.668 I0425 09:34:33.931013 136901049722688 metric_logger.py:196] completed step: 4, seconds: 0.578, TFLOP/s/device: 94.001, Tokens/s/device: 14168.843, total_weights: 262144, loss: 10.022, lm_loss: 10.022, perplexity: 22524.992 I0425 09:34:33.936966 136901049722688 metric_logger.py:196] completed step: 5, seconds: 0.578, TFLOP/s/device: 93.999, Tokens/s/device: 14168.623, total_weights: 262144, loss: 9.820, lm_loss: 9.820, perplexity: 18401.865 I0425 09:34:33.971740 2574 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 09:34:35.941731 136772355585792 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints/0/items/array_metadatas/process_3 I0425 09:35:08.345474 136772889601792 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 41.618 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.9544997215271 s) (per-host) I0425 09:35:08.345583 136772889601792 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.671280s. I0425 09:35:14.473419 136772889601792 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 43.799100s. I0425 09:35:14.473706 136772347193088 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 09:35:14.473983 136772347193088 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0425 09:35:14.474038 136772347193088 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 09:35:14.476469 136772347193088 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 09:35:19.483664 136901049722688 metric_logger.py:196] completed step: 6, seconds: 1.157, TFLOP/s/device: 46.978, Tokens/s/device: 7081.017, total_weights: 262144, loss: 9.667, lm_loss: 9.667, perplexity: 15787.604 I0425 09:35:20.061715 136901049722688 metric_logger.py:196] completed step: 7, seconds: 44.969, TFLOP/s/device: 1.209, Tokens/s/device: 182.169, total_weights: 262144, loss: 9.561, lm_loss: 9.561, perplexity: 14203.827 I0425 09:35:20.640033 136901049722688 metric_logger.py:196] completed step: 8, seconds: 0.583, TFLOP/s/device: 93.230, Tokens/s/device: 14052.687, total_weights: 262144, loss: 9.496, lm_loss: 9.496, perplexity: 13302.920 I0425 09:35:21.217478 136901049722688 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0425 09:35:21.218164 136901049722688 checkpointing.py:798] Waited 0.0007066726684570312 seconds for step 9 to finish before starting checkpointing. I0425 09:35:21.220285 136901049722688 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 09:35:21.222122 136901049722688 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 9 I0425 09:35:21.223551 136901049722688 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints/9. I0425 09:35:21.640675 136901049722688 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0425 09:35:21.640773 136901049722688 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 I0425 09:35:21.678354 136901049722688 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038517s I0425 09:35:21.678520 136901049722688 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 36.773 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04194903373718262 s) (per-host) I0425 09:35:21.678573 136901049722688 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.042011s (batch_requests_ready=0.001661s, total_serialization_initiated=0.040280s, others=0.000070s) I0425 09:35:21.678663 136901049722688 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.046233s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=0.046218) I0425 09:35:21.679409 136901049722688 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.46 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints/9. I0425 09:35:21.679751 136766447150848 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 09:55:21.679715 I0425 09:35:21.683714 136901049722688 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0425 09:35:21.683992 136772868867840 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0425 09:35:21.684153 136901049722688 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777109721.2202525, 'wait_for_prev_duration_secs': 7.390975952148438e-05, 'time_between_consecutive_saves_sec': 6.743741750717163, 'checkpointer_blocking_start_time': 1777109721.2221608, 'checkpointer_blocking_duration_secs': 0.4577360153198242, 'get_old_steps_start_time': 1777109721.6799204, 'get_old_steps_duration_secs': 4.291534423828125e-05, 'checkpoint_manager_blocking_start_time': 1777109721.218386, 'checkpoint_manager_blocking_duration_secs': 0.465731143951416} I0425 09:35:21.684264 136901049722688 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0425 09:35:21.684310 136901049722688 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 09:35:26.865104 136772877260544 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_06_grad_accum/checkpoints/9/items/array_metadatas/process_3 I0425 09:36:04.222417 136766447150848 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.092 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.58578562736511 s) (per-host) I0425 09:36:04.222559 136766447150848 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.542694s. I0425 09:36:10.567231 136766447150848 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 48.887350s. I0425 09:36:10.567506 136772868867840 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 09:36:10.567625 136772868867840 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0425 09:36:10.567673 136772868867840 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0425 09:36:10.570183 136772868867840 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0425 09:36:10.570373 136901049722688 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0425 09:36:10.570526 136901049722688 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 09:36:10.571483 136901049722688 metric_logger.py:196] completed step: 9, seconds: 0.578, TFLOP/s/device: 94.010, Tokens/s/device: 14170.191, total_weights: 262144, loss: 9.457, lm_loss: 9.457, perplexity: 12802.546 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sat Apr 25 09:36:23 UTC 2026 EXIT_CODE=0