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 | 5.8920 | 5.8920 | 0 |
| TFLOP/s | 58.252 | 58.455 | +0.203 |
| Tok/s | 8780.4 | 8811.0 | +30.674 |
| Avg s/step | 3.338 | 3.386 | +0.048 |
| 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 07:45:51 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-22 07:46:16.326214: 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 07:46:16.540029 138718500976448 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 07:46:25.581702 138718500976448 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 07:46:25.584154 138718500976448 distributed.py:172] Connecting to JAX distributed service on mt-09-pdb-lt-1-y9mb9-slice-job-0-0.mt-09-pdb-lt-1-y9mb9:8482 I0422 07:46:26.465721 138718500976448 max_utils.py:284] Jax distributed system initialized! I0422 07:46:32.719414 138718500976448 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 07:46:32.719518 138718500976448 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 07:46:32.719559 138718500976448 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 07:46:32.719596 138718500976448 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 07:46:33.412041 138718500976448 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 8, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0422 07:46:33.412338 138718500976448 checkpointing.py:677] Setting up checkpoint logger... I0422 07:46:33.412393 138718500976448 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 07:46:33.412438 138718500976448 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 07:46:33.412769 138718500976448 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 0x7e2933251070>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 07:46:36.380231 138718500976448 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 07:46:36.380492 138718500976448 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e2826a0fe00>}, handler_registry=None I0422 07:46:36.380730 138718500976448 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e2826a0fe00>` 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 07:46:36.380778 138718500976448 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e1530179eb0>` 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 07:46:36.380813 138718500976448 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 0x7e2826a0fe00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e2826a0fe00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e1530179eb0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e1530179eb0>}). I0422 07:46:36.381130 138718500976448 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 07:46:36.381200 138718500976448 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e14184858a0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 07:46:37.845393 138718500976448 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints I0422 07:46:37.847739 138718500976448 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=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_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e1530173ec0> I0422 07:46:37.847851 138718500976448 checkpointing.py:301] Checkpoint manager created! I0422 07:46:38.779193 138718500976448 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 07:46:38.779330 138718500976448 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 07:46:39.165205 138718500976448 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:46:39.165311 138718500976448 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 07:46:39.187848 138718500976448 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:46:39.187945 138718500976448 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 07:46:39.212785 138718500976448 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:46:39.212854 138718500976448 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 07:46:39.230208 138718500976448 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:46:39.230269 138718500976448 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 07:46:39.247452 138718500976448 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:46:39.247520 138718500976448 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 07:46:39.273055 138718500976448 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 07:46:39.273128 138718500976448 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 07:46:39.294973 138718500976448 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 07:46:39.295040 138718500976448 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, 'tensor'). I0422 07:46:39.511651 138718500976448 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 07:46:39.511765 138718500976448 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 8 tensor: 4 I0422 07:46:40.994091 138718500976448 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: ('tensor',) I0422 07:46:40.994222 138718500976448 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, 'tensor') I0422 07:46:40.994277 138718500976448 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, 'tensor') I0422 07:46:40.994355 138718500976448 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0422 07:46:40.994417 138718500976448 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0422 07:46:40.994485 138718500976448 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0422 07:46:40.994566 138718500976448 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, 'tensor', None) I0422 07:46:40.994640 138718500976448 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: ('tensor', None, None, 'fsdp') I0422 07:46:40.994687 138718500976448 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, 'tensor', None) I0422 07:46:40.994725 138718500976448 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, 'tensor', None) I0422 07:46:40.994775 138718500976448 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0422 07:46:40.994824 138718500976448 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0422 07:46:41.511775 138718500976448 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:46:41.511870 138718500976448 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0422 07:46:41.527955 138718500976448 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:46:41.528014 138718500976448 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0422 07:46:56.585261 138718500976448 max_utils.py:791] Total memory size: 0.9 GB, Output size: 0.4 GB, Temp size: 0.5 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 07:46:56.586024 138718500976448 metric_logger.py:301] number parameters: 1.104 billion I0422 07:46:57.775649 138718500976448 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 07:47:12.348656 138718500976448 checkpointing.py:776] Waited 14.572985649108887 seconds for step 0 to finish before starting checkpointing. I0422 07:47:12.351089 138718500976448 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:47:12.352882 138718500976448 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0422 07:47:12.354495 138718500976448 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints/0. I0422 07:47:12.675990 138718500976448 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 07:47:12.676940 138718500976448 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:47:12.676998 138718500976448 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 07:47:12.945339 138718500976448 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.269421s I0422 07:47:12.945517 138718500976448 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.602 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27536487579345703 s) (per-host) I0422 07:47:12.945566 138718500976448 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.275424s (batch_requests_ready=0.002432s, total_serialization_initiated=0.272925s, others=0.000067s) I0422 07:47:12.945653 138718500976448 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.280032s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.280015) I0422 07:47:12.946445 138718500976448 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.59 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints/0. I0422 07:47:12.946800 138590574339840 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 08:07:12.946762 I0422 07:47:12.959494 138718500976448 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:47:12.959813 138590071813888 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:47:12.959969 138718500976448 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776844032.3510706, 'wait_for_prev_duration_secs': 6.008148193359375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776844032.352921, 'checkpointer_blocking_duration_secs': 0.5940506458282471, 'get_old_steps_start_time': 1776844032.9469981, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1776844032.349094, 'checkpoint_manager_blocking_duration_secs': 0.6108372211456299} I0422 07:47:12.960077 138718500976448 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 07:47:12.960133 138718500976448 max_utils.py:750] Memstats: After params initialized: I0422 07:47:12.960183 138718500976448 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0422 07:47:12.960215 138718500976448 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0422 07:47:12.960243 138718500976448 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0422 07:47:12.960270 138718500976448 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0422 07:47:13.270932 138718500976448 metric_logger.py:196] completed step: 0, seconds: 1.190, TFLOP/s/device: 2.856, Tokens/s/device: 430.429, total_weights: 16384, loss: 10.862, lm_loss: 10.862, perplexity: 52153.723 I0422 07:47:13.363400 138718500976448 metric_logger.py:196] completed step: 1, seconds: 15.494, TFLOP/s/device: 0.219, Tokens/s/device: 33.046, total_weights: 16384, loss: 10.862, lm_loss: 10.862, perplexity: 52153.723 I0422 07:47:13.776438 138718500976448 metric_logger.py:196] completed step: 2, seconds: 0.034, TFLOP/s/device: 99.333, Tokens/s/device: 14972.511, total_weights: 16384, loss: 9.763, lm_loss: 9.763, perplexity: 17373.129 I0422 07:47:13.834601 138718500976448 metric_logger.py:196] completed step: 3, seconds: 0.413, TFLOP/s/device: 8.221, Tokens/s/device: 1239.118, total_weights: 16384, loss: 8.823, lm_loss: 8.823, perplexity: 6791.817 I0422 07:47:13.951608 138718500976448 metric_logger.py:196] completed step: 4, seconds: 0.064, TFLOP/s/device: 52.950, Tokens/s/device: 7981.169, total_weights: 16384, loss: 7.947, lm_loss: 7.947, perplexity: 2828.417 I0422 07:47:13.957203 138718500976448 metric_logger.py:196] completed step: 5, seconds: 0.058, TFLOP/s/device: 58.748, Tokens/s/device: 8855.067, total_weights: 16384, loss: 7.212, lm_loss: 7.212, perplexity: 1355.339 I0422 07:47:15.901681 2544 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 07:47:17.938209 138590550959872 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_6 I0422 07:47:27.754382 138718500976448 metric_logger.py:196] completed step: 6, seconds: 0.117, TFLOP/s/device: 28.925, Tokens/s/device: 4359.933, total_weights: 16384, loss: 6.652, lm_loss: 6.652, perplexity: 774.659 I0422 07:47:27.812719 138718500976448 metric_logger.py:196] completed step: 7, seconds: 13.739, TFLOP/s/device: 0.247, Tokens/s/device: 37.265, total_weights: 16384, loss: 6.269, lm_loss: 6.269, perplexity: 527.696 I0422 07:47:27.871511 138718500976448 metric_logger.py:196] completed step: 8, seconds: 0.063, TFLOP/s/device: 53.719, Tokens/s/device: 8097.166, total_weights: 16384, loss: 6.031, lm_loss: 6.031, perplexity: 416.165 I0422 07:47:27.929187 138718500976448 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 07:47:27.929825 138718500976448 checkpointing.py:776] Waited 0.0006561279296875 seconds for step 9 to finish before starting checkpointing. I0422 07:47:27.933099 138718500976448 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:47:49.029658 138590574339840 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 43.444 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.359461545944214 s) (per-host) I0422 07:47:49.029765 138590574339840 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.082841s. I0422 07:47:57.385383 138590574339840 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 44.438443s. I0422 07:47:57.385678 138590071813888 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:47:57.385795 138590071813888 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:47:57.385845 138590071813888 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:47:57.388420 138590071813888 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 07:47:57.388604 138718500976448 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0422 07:47:57.388745 138718500976448 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.455643 seconds. If this number is high, consider checkpointing less frequently. I0422 07:47:57.390564 138718500976448 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0422 07:47:57.392548 138718500976448 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints/9. I0422 07:47:58.096515 138718500976448 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:47:58.096612 138718500976448 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 07:47:58.121503 138718500976448 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026121s I0422 07:47:58.121668 138718500976448 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 51.765 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.029799699783325195 s) (per-host) I0422 07:47:58.121727 138718500976448 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.029867s (batch_requests_ready=0.001976s, total_serialization_initiated=0.027814s, others=0.000077s) I0422 07:47:58.121835 138718500976448 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.034032s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=0.034018) I0422 07:47:58.122742 138718500976448 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.73 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints/9. I0422 07:47:58.123042 138590071813888 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 08:07:58.123009 I0422 07:47:58.126780 138718500976448 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:47:58.127037 138594861688576 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:47:58.127145 138718500976448 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776844047.9330704, 'wait_for_prev_duration_secs': 29.45564341545105, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776844077.3906043, 'checkpointer_blocking_duration_secs': 0.7325859069824219, 'get_old_steps_start_time': 1776844078.1232142, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776844047.9300482, 'checkpoint_manager_blocking_duration_secs': 30.19706082344055} I0422 07:47:58.127319 138718500976448 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 07:47:58.127367 138718500976448 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:48:03.272415 138590055028480 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_6 I0422 07:48:39.063540 138590071813888 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.553 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.97163152694702 s) (per-host) I0422 07:48:39.063654 138590071813888 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.940492s. I0422 07:48:50.139764 138590071813888 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 52.016585s. I0422 07:48:50.140059 138594861688576 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:48:50.140186 138594861688576 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:48:50.140236 138594861688576 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:48:50.141618 138594861688576 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 07:48:50.141774 138718500976448 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 07:48:50.141919 138718500976448 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:48:50.142935 138718500976448 metric_logger.py:196] completed step: 9, seconds: 0.058, TFLOP/s/device: 58.252, Tokens/s/device: 8780.354, total_weights: 16384, loss: 5.892, lm_loss: 5.892, perplexity: 361.955 Per train step: Total TFLOPs: 3.40 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 07:49:01 UTC 2026 EXIT_CODE=0
XPK Start: Wed Apr 22 13:12:57 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 13:13:22.803836: 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 13:13:23.015605 140367654737728 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 13:13:32.055490 140367654737728 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 13:13:32.057838 140367654737728 distributed.py:172] Connecting to JAX distributed service on mt-09-pdb-lt-1-hz6q3-slice-job-0-0.mt-09-pdb-lt-1-hz6q3:8482 I0422 13:13:32.868310 140367654737728 max_utils.py:284] Jax distributed system initialized! I0422 13:13:38.959683 140367654737728 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 13:13:38.959786 140367654737728 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 13:13:38.959828 140367654737728 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 13:13:38.959867 140367654737728 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 13:13:39.652363 140367654737728 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 8, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0422 13:13:39.652995 140367654737728 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 8, 1, 1, 1, 1, 4, 1, 1, 1, 1) I0422 13:13:39.653173 140367654737728 checkpointing.py:688] Setting up checkpoint logger... I0422 13:13:39.653223 140367654737728 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 13:13:39.653266 140367654737728 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 13:13:39.653600 140367654737728 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 0x7fa92c850290>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 13:13:42.582793 140367654737728 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0422 13:13:42.583030 140367654737728 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 0x7f950876a7b0>}, handler_registry=None I0422 13:13:42.583287 140367654737728 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f950876a7b0>` 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 13:13:42.583338 140367654737728 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f95085e32f0>` 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 13:13:42.583375 140367654737728 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 0x7f950876a7b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f950876a7b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f95085e32f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f95085e32f0>}). I0422 13:13:42.583717 140367654737728 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 13:13:42.583787 140367654737728 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f94f0381d00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 13:13:44.712403 140367654737728 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_09_pdb_lt_1/checkpoints I0422 13:13:44.714723 140367654737728 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_09_pdb_lt_1/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f950876acf0> I0422 13:13:44.714833 140367654737728 checkpointing.py:302] Checkpoint manager created! I0422 13:13:45.638939 140367654737728 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 13:13:45.639042 140367654737728 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 13:13:46.023099 140367654737728 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 13:13:46.023196 140367654737728 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 13:13:46.039812 140367654737728 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 13:13:46.039872 140367654737728 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, 'tensor'). I0422 13:13:46.063781 140367654737728 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 13:13:46.063850 140367654737728 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 13:13:46.080466 140367654737728 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 13:13:46.080531 140367654737728 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 13:13:46.097175 140367654737728 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 13:13:46.097241 140367654737728 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 13:13:46.122298 140367654737728 attentions.py:1198] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 13:13:46.122363 140367654737728 attentions.py:1198] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, 'tensor', None). I0422 13:13:46.143236 140367654737728 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 13:13:46.143298 140367654737728 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, 'tensor'). I0422 13:13:46.366343 140367654737728 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0422 13:13:46.366452 140367654737728 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 8 tensor: 4 I0422 13:13:47.830213 140367654737728 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: ('tensor',) I0422 13:13:47.830343 140367654737728 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, 'tensor') I0422 13:13:47.830397 140367654737728 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, 'tensor') I0422 13:13:47.830455 140367654737728 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: ('tensor', None, 'fsdp') I0422 13:13:47.830508 140367654737728 maxtext_utils.py:1880] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0422 13:13:47.830548 140367654737728 maxtext_utils.py:1880] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: ('tensor', None) I0422 13:13:47.830601 140367654737728 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, 'tensor', None) I0422 13:13:47.830666 140367654737728 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: ('tensor', None, None, 'fsdp') I0422 13:13:47.830712 140367654737728 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, 'tensor', None) I0422 13:13:47.830751 140367654737728 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, 'tensor', None) I0422 13:13:47.830801 140367654737728 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', 'tensor') I0422 13:13:47.830855 140367654737728 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: ('tensor', 'fsdp') I0422 13:13:48.331147 140367654737728 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 13:13:48.331240 140367654737728 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0422 13:13:48.346840 140367654737728 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0422 13:13:48.346902 140367654737728 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). I0422 13:14:03.345853 140367654737728 max_utils.py:791] Total memory size: 0.9 GB, Output size: 0.4 GB, Temp size: 0.5 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 13:14:03.346643 140367654737728 metric_logger.py:301] number parameters: 1.104 billion I0422 13:14:04.532193 140367654737728 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0422 13:14:19.017856 140367654737728 checkpointing.py:798] Waited 14.485643148422241 seconds for step 0 to finish before starting checkpointing. I0422 13:14:19.020319 140367654737728 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 13:14:19.025296 140367654737728 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0422 13:14:19.026980 140367654737728 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_09_pdb_lt_1/checkpoints/0. I0422 13:14:19.322294 140367654737728 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 13:14:19.323218 140367654737728 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 13:14:19.323280 140367654737728 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 13:14:19.588534 140367654737728 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.266319s I0422 13:14:19.588704 140367654737728 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.667 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.272202730178833 s) (per-host) I0422 13:14:19.588754 140367654737728 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.272263s (batch_requests_ready=0.002393s, total_serialization_initiated=0.269803s, others=0.000067s) I0422 13:14:19.588846 140367654737728 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.276453s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.276435) I0422 13:14:19.589640 140367654737728 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.56 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_09_pdb_lt_1/checkpoints/0. I0422 13:14:19.589938 140240090621696 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 13:34:19.589906 I0422 13:14:19.608962 140367654737728 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 13:14:19.609271 140239590905600 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0422 13:14:19.609428 140367654737728 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_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776863659.0203009, 'wait_for_prev_duration_secs': 6.127357482910156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776863659.0253382, 'checkpointer_blocking_duration_secs': 0.5647475719451904, 'get_old_steps_start_time': 1776863659.590111, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776863659.0183918, 'checkpoint_manager_blocking_duration_secs': 0.5909972190856934} I0422 13:14:19.609534 140367654737728 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0422 13:14:19.609584 140367654737728 max_utils.py:750] Memstats: After params initialized: I0422 13:14:19.609635 140367654737728 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0)) I0422 13:14:19.609681 140367654737728 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0422 13:14:19.609711 140367654737728 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0422 13:14:19.609737 140367654737728 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0422 13:14:19.923069 140367654737728 metric_logger.py:196] completed step: 0, seconds: 1.185, TFLOP/s/device: 2.865, Tokens/s/device: 431.912, total_weights: 16384, loss: 10.862, lm_loss: 10.862, perplexity: 52153.723 I0422 13:14:20.007662 140367654737728 metric_logger.py:196] completed step: 1, seconds: 15.389, TFLOP/s/device: 0.221, Tokens/s/device: 33.270, total_weights: 16384, loss: 10.862, lm_loss: 10.862, perplexity: 52153.723 I0422 13:14:20.437523 140367654737728 metric_logger.py:196] completed step: 2, seconds: 0.026, TFLOP/s/device: 129.008, Tokens/s/device: 19445.499, total_weights: 16384, loss: 9.763, lm_loss: 9.763, perplexity: 17373.129 I0422 13:14:20.495886 140367654737728 metric_logger.py:196] completed step: 3, seconds: 0.430, TFLOP/s/device: 7.894, Tokens/s/device: 1189.868, total_weights: 16384, loss: 8.823, lm_loss: 8.823, perplexity: 6791.817 I0422 13:14:20.613210 140367654737728 metric_logger.py:196] completed step: 4, seconds: 0.064, TFLOP/s/device: 53.348, Tokens/s/device: 8041.211, total_weights: 16384, loss: 7.947, lm_loss: 7.947, perplexity: 2828.417 I0422 13:14:20.619009 140367654737728 metric_logger.py:196] completed step: 5, seconds: 0.058, TFLOP/s/device: 58.239, Tokens/s/device: 8778.397, total_weights: 16384, loss: 7.212, lm_loss: 7.212, perplexity: 1355.339 I0422 13:14:23.332872 2521 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 13:14:26.153626 140239599298304 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_09_pdb_lt_1/checkpoints/0/items/array_metadatas/process_5 I0422 13:14:34.944579 140367654737728 metric_logger.py:196] completed step: 6, seconds: 0.118, TFLOP/s/device: 28.831, Tokens/s/device: 4345.686, total_weights: 16384, loss: 6.652, lm_loss: 6.652, perplexity: 774.659 I0422 13:14:35.002847 140367654737728 metric_logger.py:196] completed step: 7, seconds: 14.268, TFLOP/s/device: 0.238, Tokens/s/device: 35.885, total_weights: 16384, loss: 6.269, lm_loss: 6.269, perplexity: 527.696 I0422 13:14:35.061342 140367654737728 metric_logger.py:196] completed step: 8, seconds: 0.063, TFLOP/s/device: 53.905, Tokens/s/device: 8125.179, total_weights: 16384, loss: 6.031, lm_loss: 6.031, perplexity: 416.165 I0422 13:14:35.119088 140367654737728 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0422 13:14:35.119606 140367654737728 checkpointing.py:798] Waited 0.0005335807800292969 seconds for step 9 to finish before starting checkpointing. I0422 13:14:35.122885 140367654737728 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 13:14:56.470098 140240090621696 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 42.515 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.15356159210205 s) (per-host) I0422 13:14:56.470205 140240090621696 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.880158s. I0422 13:15:06.085543 140240090621696 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 46.495480s. I0422 13:15:06.085861 140239590905600 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 13:15:06.086015 140239590905600 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0422 13:15:06.086070 140239590905600 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 13:15:06.087640 140239590905600 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 13:15:06.087854 140367654737728 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 13:15:06.087985 140367654737728 checkpoint_manager.py:1452] Waiting for previous save to complete took 30.965100 seconds. If this number is high, consider checkpointing less frequently. I0422 13:15:06.089817 140367654737728 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0422 13:15:06.091876 140367654737728 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_09_pdb_lt_1/checkpoints/9. I0422 13:15:06.377706 140367654737728 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 13:15:06.377799 140367654737728 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 13:15:06.402517 140367654737728 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.025892s I0422 13:15:06.402643 140367654737728 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 52.194 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.02955460548400879 s) (per-host) I0422 13:15:06.402705 140367654737728 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.029625s (batch_requests_ready=0.001992s, total_serialization_initiated=0.027558s, others=0.000075s) I0422 13:15:06.402783 140367654737728 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.033670s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.033653) I0422 13:15:06.403487 140367654737728 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.31 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_09_pdb_lt_1/checkpoints/9. I0422 13:15:06.403799 140233123276544 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 13:35:06.403777 I0422 13:15:06.408542 140367654737728 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 13:15:06.408814 140239590905600 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0422 13:15:06.408973 140367654737728 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_09_pdb_lt_1/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776863675.1228557, 'wait_for_prev_duration_secs': 30.965100288391113, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776863706.0898557, 'checkpointer_blocking_duration_secs': 0.3140420913696289, 'get_old_steps_start_time': 1776863706.4039202, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776863675.1198459, 'checkpoint_manager_blocking_duration_secs': 31.289092540740967} I0422 13:15:06.409083 140367654737728 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0422 13:15:06.409129 140367654737728 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 13:15:11.854687 140233114883840 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_09_pdb_lt_1/checkpoints/9/items/array_metadatas/process_5 I0422 13:15:48.088850 140233123276544 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.866 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.71572303771973 s) (per-host) I0422 13:15:48.088956 140233123276544 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.685081s. I0422 13:15:55.895950 140233123276544 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 49.492057s. I0422 13:15:55.896236 140239590905600 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 13:15:55.896363 140239590905600 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0422 13:15:55.896412 140239590905600 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 13:15:55.897805 140239590905600 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 13:15:55.897984 140367654737728 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 13:15:55.898125 140367654737728 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 13:15:55.899178 140367654737728 metric_logger.py:196] completed step: 9, seconds: 0.058, TFLOP/s/device: 58.455, Tokens/s/device: 8811.028, total_weights: 16384, loss: 5.892, lm_loss: 5.892, perplexity: 361.955 Per train step: Total TFLOPs: 3.40 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 13:16:07 UTC 2026 EXIT_CODE=0