main (8a17c3d19) vs feat/nnx-trainstate-and-training-loop (29bbe47ed)| Metric | main 8a17c3d19 | feat/nnx-trainstate-and-training-loop 29bbe47ed | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1790 | 8.1790 | 0 |
| TFLOP/s | 92.242 | 91.953 | -0.289 |
| Tok/s | 13903.8 | 13860.1 | -43.661 |
| Avg s/step | 2.863 | 2.910 | +0.047 |
| 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 07:15:01 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 07:15:26.512843: 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 07:15:26.722170 140316527699776 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 07:15:35.762890 140316527699776 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 07:15:35.765169 140316527699776 distributed.py:172] Connecting to JAX distributed service on mt-01-base-zwj6a-slice-job-0-0.mt-01-base-zwj6a:8482 I0422 07:15:37.373047 140316527699776 max_utils.py:284] Jax distributed system initialized! I0422 07:15:44.432591 140316527699776 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 07:15:44.432708 140316527699776 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 07:15:44.432755 140316527699776 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 07:15:44.432792 140316527699776 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 07:15:45.132276 140316527699776 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 07:15:45.132596 140316527699776 checkpointing.py:677] Setting up checkpoint logger... I0422 07:15:45.132658 140316527699776 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 07:15:45.132729 140316527699776 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 07:15:45.133094 140316527699776 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 0x7f9d6e1ea6c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 07:15:49.268826 140316527699776 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 07:15:49.269061 140316527699776 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f88e85c07d0>}, handler_registry=None I0422 07:15:49.269305 140316527699776 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f88e85c07d0>` 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 07:15:49.269355 140316527699776 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f88e85c4470>` 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 07:15:49.269391 140316527699776 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 0x7f88e85c07d0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f88e85c07d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f88e85c4470>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f88e85c4470>}). I0422 07:15:49.269721 140316527699776 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 07:15:49.269792 140316527699776 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f88b86a9760> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 07:15:50.833351 140316527699776 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints I0422 07:15:51.317048 140316527699776 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=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_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f88e85c5b20> I0422 07:15:51.317233 140316527699776 checkpointing.py:301] Checkpoint manager created! I0422 07:15:52.263661 140316527699776 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 07:15:52.263802 140316527699776 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:15:52.644909 140316527699776 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:15:52.645006 140316527699776 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:15:52.661463 140316527699776 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:15:52.661525 140316527699776 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:15:52.685244 140316527699776 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:15:52.685312 140316527699776 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:15:52.701889 140316527699776 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:15:52.701953 140316527699776 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:15:52.718328 140316527699776 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:15:52.718389 140316527699776 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:15:52.742961 140316527699776 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 07:15:52.743033 140316527699776 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:15:52.764258 140316527699776 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 07:15:52.764323 140316527699776 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 07:15:52.973218 140316527699776 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 07:15:52.973325 140316527699776 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0422 07:15:54.390054 140316527699776 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 07:15:54.390210 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 07:15:54.390284 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 07:15:54.390372 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 07:15:54.390471 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 07:15:54.390542 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 07:15:54.390633 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 07:15:54.390747 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0422 07:15:54.390826 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0422 07:15:54.390898 140316527699776 maxtext_utils.py:1668] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 07:15:54.390988 140316527699776 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 07:15:54.391076 140316527699776 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 07:15:54.882288 140316527699776 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:15:54.882382 140316527699776 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 07:15:54.897633 140316527699776 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:15:54.897704 140316527699776 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 07:16:05.860280 140316527699776 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 07:16:05.861088 140316527699776 metric_logger.py:301] number parameters: 1.104 billion I0422 07:16:17.478053 140316527699776 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 07:16:17.631637 140316527699776 checkpointing.py:776] Waited 0.1535651683807373 seconds for step 0 to finish before starting checkpointing. I0422 07:16:17.635429 140316527699776 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:16:17.636765 140316527699776 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0422 07:16:17.638306 140316527699776 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints/0. I0422 07:16:18.809293 140316527699776 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 07:16:18.810253 140316527699776 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:16:18.810310 140316527699776 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 07:16:19.084852 140316527699776 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.275632s I0422 07:16:19.085027 140316527699776 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.487 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2811152935028076 s) (per-host) I0422 07:16:19.085081 140316527699776 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.281180s (batch_requests_ready=0.002196s, total_serialization_initiated=0.278911s, others=0.000073s) I0422 07:16:19.085182 140316527699776 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.285327s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.285310) I0422 07:16:19.085998 140316527699776 event_tracking.py:125] [process=4] [async] Finished blocking save in 1.45 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints/0. I0422 07:16:19.086278 140188202301184 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:36:19.086246 I0422 07:16:19.510393 140316527699776 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:16:19.510795 140187213117184 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:16:19.510977 140316527699776 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776842177.635411, 'wait_for_prev_duration_secs': 6.079673767089844e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776842177.636802, 'checkpointer_blocking_duration_secs': 1.449615478515625, 'get_old_steps_start_time': 1776842179.0864444, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1776842177.6321578, 'checkpoint_manager_blocking_duration_secs': 1.8787660598754883} I0422 07:16:19.511101 140316527699776 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 07:16:19.511154 140316527699776 max_utils.py:750] Memstats: After params initialized: I0422 07:16:19.511220 140316527699776 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0)) I0422 07:16:19.511255 140316527699776 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0422 07:16:19.511283 140316527699776 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0422 07:16:19.511311 140316527699776 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0422 07:16:19.822357 140316527699776 metric_logger.py:196] completed step: 0, seconds: 11.617, TFLOP/s/device: 1.170, Tokens/s/device: 176.296, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0422 07:16:19.994666 140316527699776 metric_logger.py:196] completed step: 1, seconds: 2.343, TFLOP/s/device: 5.800, Tokens/s/device: 874.232, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0422 07:16:20.473641 140316527699776 metric_logger.py:196] completed step: 2, seconds: 0.025, TFLOP/s/device: 542.747, Tokens/s/device: 81808.740, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0422 07:16:20.620892 140316527699776 metric_logger.py:196] completed step: 3, seconds: 0.480, TFLOP/s/device: 28.334, Tokens/s/device: 4270.822, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0422 07:16:20.915948 140316527699776 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.791, Tokens/s/device: 13383.522, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0422 07:16:20.921850 140316527699776 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.391, Tokens/s/device: 13926.194, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0422 07:16:23.290975 2918 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 07:16:25.319835 140187221509888 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints/0/items/array_metadatas/process_4 I0422 07:16:43.092731 140316527699776 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.977, Tokens/s/device: 6930.180, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0422 07:16:43.239900 140316527699776 metric_logger.py:196] completed step: 7, seconds: 22.024, TFLOP/s/device: 0.617, Tokens/s/device: 92.989, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0422 07:16:43.387319 140316527699776 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.336, Tokens/s/device: 13465.711, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0422 07:16:43.533977 140316527699776 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 07:16:43.534661 140316527699776 checkpointing.py:776] Waited 0.0007050037384033203 seconds for step 9 to finish before starting checkpointing. I0422 07:16:43.536738 140316527699776 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:16:57.731692 140188202301184 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 40.578 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.92774844169617 s) (per-host) I0422 07:16:57.731810 140188202301184 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.645426s. I0422 07:17:06.491195 140188202301184 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 47.404796s. I0422 07:17:06.491465 140187213117184 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:17:06.491616 140187213117184 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:17:06.491680 140187213117184 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:17:06.493492 140187213117184 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 07:17:06.493671 140316527699776 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0422 07:17:06.493820 140316527699776 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.957085 seconds. If this number is high, consider checkpointing less frequently. I0422 07:17:06.495635 140316527699776 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0422 07:17:06.497742 140316527699776 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints/9. I0422 07:17:07.210859 140316527699776 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:17:07.210949 140316527699776 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 07:17:07.249529 140316527699776 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039744s I0422 07:17:07.249692 140316527699776 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 35.692 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.043219566345214844 s) (per-host) I0422 07:17:07.249756 140316527699776 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043293s (batch_requests_ready=0.001797s, total_serialization_initiated=0.041415s, others=0.000081s) I0422 07:17:07.249850 140316527699776 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.047397s (all_items=0.000014s, per_item={'items': '0.00001383'}, temp_paths=0.047383) I0422 07:17:07.250551 140316527699776 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints/9. I0422 07:17:07.250847 140187213117184 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:37:07.250815 I0422 07:17:07.252829 140316527699776 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:17:07.253130 140165628544768 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:17:07.253295 140316527699776 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776842203.5366917, 'wait_for_prev_duration_secs': 22.957084894180298, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776842226.4956741, 'checkpointer_blocking_duration_secs': 0.7553098201751709, 'get_old_steps_start_time': 1776842227.2510092, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776842203.5349169, 'checkpoint_manager_blocking_duration_secs': 23.718341588974} I0422 07:17:07.253410 140316527699776 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 07:17:07.253456 140316527699776 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:17:13.871974 140187221509888 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_01_base/checkpoints/9/items/array_metadatas/process_4 I0422 07:17:50.599343 140187213117184 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 36.402 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.39284563064575 s) (per-host) I0422 07:17:50.599456 140187213117184 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.348497s. I0422 07:18:00.477965 140187213117184 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 53.226990s. I0422 07:18:00.478232 140165628544768 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:18:00.478347 140165628544768 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:18:00.478395 140165628544768 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:18:00.479907 140165628544768 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 07:18:00.480103 140316527699776 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 07:18:00.480257 140316527699776 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:18:00.481197 140316527699776 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.242, Tokens/s/device: 13903.787, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 07:18:12 UTC 2026 EXIT_CODE=0
29bbe47ed
· feat_nnx_trainstate_and_training_loop_20260422_093106
· full log
XPK Start: Wed Apr 22 09:32:47 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 09:33:21.652340: 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 09:33:21.863708 137638995367744 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 09:33:30.907554 137638995367744 distributed.py:172] Connecting to JAX distributed service on mt-01-base-c8i0u-slice-job-0-0.mt-01-base-c8i0u:8482 I0422 09:33:43.936339 137638995367744 max_utils.py:284] Jax distributed system initialized! I0422 09:33:49.044614 137638995367744 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 09:33:49.044728 137638995367744 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 09:33:49.044779 137638995367744 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 09:33:49.044823 137638995367744 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 09:33:49.725780 137638995367744 maxtext_utils.py:1732] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 09:33:49.726364 137638995367744 maxtext_utils.py:1732] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 09:33:49.726547 137638995367744 checkpointing.py:688] Setting up checkpoint logger... I0422 09:33:49.726611 137638995367744 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 09:33:49.726655 137638995367744 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 09:33:49.726988 137638995367744 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 0x7d2df271cfb0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 09:33:53.411594 137638995367744 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0422 09:33:53.411825 137638995367744 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d1b540d2720>}, handler_registry=None I0422 09:33:53.412059 137638995367744 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d1b540d2720>` 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 09:33:53.412110 137638995367744 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d1b540da0f0>` 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 09:33:53.412147 137638995367744 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 0x7d1b540d2720>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d1b540d2720>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d1b540da0f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d1b540da0f0>}). I0422 09:33:53.412474 137638995367744 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 09:33:53.412543 137638995367744 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d1b1c151d00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 09:33:54.407482 137638995367744 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints I0422 09:33:54.828804 137638995367744 checkpoint_manager.py:929] [process=0][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_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d1b540d38f0> I0422 09:33:54.828979 137638995367744 checkpointing.py:302] Checkpoint manager created! I0422 09:33:55.728529 137638995367744 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 09:33:55.728665 137638995367744 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 09:33:56.105807 137638995367744 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 09:33:56.105902 137638995367744 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 09:33:56.122099 137638995367744 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 09:33:56.122161 137638995367744 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 09:33:56.145697 137638995367744 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 09:33:56.145763 137638995367744 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 09:33:56.161999 137638995367744 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 09:33:56.162061 137638995367744 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 09:33:56.178415 137638995367744 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 09:33:56.178476 137638995367744 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 09:33:56.203212 137638995367744 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 09:33:56.203282 137638995367744 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 09:33:56.223835 137638995367744 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 09:33:56.223899 137638995367744 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 09:33:56.447371 137638995367744 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0422 09:33:56.447482 137638995367744 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0422 09:33:57.883190 137638995367744 maxtext_utils.py:1835] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 09:33:57.883317 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 09:33:57.883376 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 09:33:57.883436 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 09:33:57.883492 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 09:33:57.883533 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 09:33:57.883599 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 09:33:57.883657 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0422 09:33:57.883698 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0422 09:33:57.883735 137638995367744 maxtext_utils.py:1835] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 09:33:57.883783 137638995367744 maxtext_utils.py:1835] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 09:33:57.883832 137638995367744 maxtext_utils.py:1835] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 09:33:58.369305 137638995367744 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 09:33:58.369400 137638995367744 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 09:33:58.384628 137638995367744 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 09:33:58.384685 137638995367744 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 09:34:08.957525 137638995367744 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 09:34:08.964995 137638995367744 metric_logger.py:301] number parameters: 1.104 billion I0422 09:34:20.208614 137638995367744 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0422 09:34:20.594354 137638995367744 checkpointing.py:798] Waited 0.3857440948486328 seconds for step 0 to finish before starting checkpointing. I0422 09:34:20.598325 137638995367744 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 09:34:20.600074 137638995367744 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0 I0422 09:34:20.601544 137638995367744 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0. I0422 09:34:21.321990 137638995367744 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 09:34:21.751209 137499709814528 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0 I0422 09:34:21.763301 137638995367744 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 09:34:21.763395 137638995367744 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 09:34:22.033226 137638995367744 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.270819s I0422 09:34:22.034260 137638995367744 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.560 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27744460105895996 s) (per-host) I0422 09:34:22.034351 137638995367744 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.277555s (batch_requests_ready=0.002169s, total_serialization_initiated=0.274410s, others=0.000976s) I0422 09:34:22.034440 137638995367744 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.282357s (all_items=0.000020s, per_item={'items': '0.00001979'}, temp_paths=0.282337) I0422 09:34:22.035175 137638995367744 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.44 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0. I0422 09:34:22.035515 137499718207232 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 09:54:22.035476 I0422 09:34:22.475130 137638995367744 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 09:34:22.475530 137494519609088 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 09:34:22.475724 137638995367744 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776850460.5983074, 'wait_for_prev_duration_secs': 5.984306335449219e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776850460.6001112, 'checkpointer_blocking_duration_secs': 1.4355194568634033, 'get_old_steps_start_time': 1776850462.0356576, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776850460.5948932, 'checkpoint_manager_blocking_duration_secs': 1.8807909488677979} I0422 09:34:22.475831 137638995367744 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0422 09:34:22.475884 137638995367744 max_utils.py:750] Memstats: After params initialized: I0422 09:34:22.475943 137638995367744 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0)) I0422 09:34:22.475979 137638995367744 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0)) I0422 09:34:22.476008 137638995367744 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0)) I0422 09:34:22.476034 137638995367744 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0)) I0422 09:34:22.822280 137638995367744 metric_logger.py:196] completed step: 0, seconds: 10.986, TFLOP/s/device: 1.237, Tokens/s/device: 186.416, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0422 09:34:22.825934 137638995367744 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/tensorboard/' I0422 09:34:23.230335 137638995367744 metric_logger.py:196] completed step: 1, seconds: 2.612, TFLOP/s/device: 5.202, Tokens/s/device: 784.120, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0422 09:34:23.376532 137638995367744 metric_logger.py:196] completed step: 2, seconds: 0.409, TFLOP/s/device: 33.244, Tokens/s/device: 5010.937, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0422 09:34:23.523711 137638995367744 metric_logger.py:196] completed step: 3, seconds: 0.010, TFLOP/s/device: 1425.273, Tokens/s/device: 214832.686, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0422 09:34:28.711164 137492455995136 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776850468291994848, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776850468291994848, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0/_CHECKPOINT_METADATA I0422 09:34:28.937565 137498277705472 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0/items I0422 09:34:30.492635 2612 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 09:34:32.689517 137496263055104 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0/items/array_metadatas/process_0 I0422 09:34:46.086607 137638995367744 metric_logger.py:196] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.374, Tokens/s/device: 13923.543, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0422 09:34:46.095474 137638995367744 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.239, Tokens/s/device: 13903.315, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0422 09:34:46.241691 137638995367744 metric_logger.py:196] completed step: 6, seconds: 22.561, TFLOP/s/device: 0.602, Tokens/s/device: 90.775, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0422 09:34:46.389205 137638995367744 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 2050.268, Tokens/s/device: 309038.781, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0422 09:34:46.536490 137638995367744 metric_logger.py:196] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.361, Tokens/s/device: 13921.650, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0422 09:34:46.683135 137638995367744 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0422 09:34:46.683862 137638995367744 checkpointing.py:798] Waited 0.0007493495941162109 seconds for step 9 to finish before starting checkpointing. I0422 09:34:46.685856 137638995367744 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 09:35:07.166052 137496209897216 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 36.673680s (commit=35.436636s, array_metadata_write=1.237043s) I0422 09:35:07.167160 137499718207232 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.785 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 45.41032433509827 s) (per-host) I0422 09:35:07.167270 137499718207232 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 45.131127s. I0422 09:35:08.325622 137499718207232 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776850468291994848, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0/_CHECKPOINT_METADATA I0422 09:35:09.512180 137492455995136 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776850468291994848, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0/_CHECKPOINT_METADATA I0422 09:35:09.909022 137499718207232 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000256s. I0422 09:35:10.290929 137499718207232 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0422 09:35:11.256692 137499718207232 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.377232s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0/items I0422 09:35:11.258374 137499718207232 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0/items I0422 09:35:11.792753 137499718207232 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0 I0422 09:35:13.219194 137499718207232 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0`. I0422 09:35:13.220015 137499718207232 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 52.62 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/0 I0422 09:35:13.221805 137499718207232 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 51.185659s. I0422 09:35:13.222032 137494519609088 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 09:35:13.222167 137494519609088 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 09:35:13.222244 137494519609088 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 09:35:13.224347 137494519609088 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 09:35:13.224536 137638995367744 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0422 09:35:13.224684 137638995367744 checkpoint_manager.py:1452] Waiting for previous save to complete took 26.538829 seconds. If this number is high, consider checkpointing less frequently. I0422 09:35:13.226616 137638995367744 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9 I0422 09:35:13.228604 137638995367744 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9. I0422 09:35:13.533279 137638995367744 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 09:35:13.533373 137638995367744 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 09:35:13.562167 137638995367744 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.029687s I0422 09:35:13.563983 137638995367744 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 43.718 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03528475761413574 s) (per-host) I0422 09:35:13.564059 137638995367744 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.035378s (batch_requests_ready=0.001874s, total_serialization_initiated=0.031773s, others=0.001731s) I0422 09:35:13.564140 137638995367744 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.039484s (all_items=0.000032s, per_item={'items': '0.00003195'}, temp_paths=0.039452) I0422 09:35:13.564838 137638995367744 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9. I0422 09:35:13.565155 137494519609088 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 09:55:13.565118 I0422 09:35:13.570440 137638995367744 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 09:35:13.570708 137496769246976 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0422 09:35:13.570842 137638995367744 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776850486.6858234, 'wait_for_prev_duration_secs': 26.53882908821106, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776850513.226655, 'checkpointer_blocking_duration_secs': 0.3386349678039551, 'get_old_steps_start_time': 1776850513.5653117, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776850486.68409, 'checkpoint_manager_blocking_duration_secs': 26.886717319488525} I0422 09:35:13.570960 137638995367744 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0422 09:35:13.571005 137638995367744 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 09:35:13.768951 137499718207232 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9 I0422 09:35:15.950234 137496263055104 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9/items I0422 09:35:18.848227 137498245748480 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9/items/array_metadatas/process_0 I0422 09:35:58.729095 137496786032384 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.644406s (commit=40.480054s, array_metadata_write=1.164351s) I0422 09:35:58.730156 137494519609088 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.946 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 45.201435565948486 s) (per-host) I0422 09:35:58.730265 137494519609088 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 45.165000s. I0422 09:36:00.458969 137494519609088 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0422 09:36:00.624284 137494519609088 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000231s. I0422 09:36:01.548747 137494519609088 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.927866s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9/items I0422 09:36:01.550265 137494519609088 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9/items I0422 09:36:02.094807 137494519609088 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9 I0422 09:36:03.511608 137494519609088 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9`. I0422 09:36:03.512424 137494519609088 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 50.29 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106/linen_xpk_feat_nnx_trainstate_and_training_loop_20260422_093106_01_base/checkpoints/9 I0422 09:36:03.514312 137494519609088 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 49.949045s. I0422 09:36:03.514518 137496769246976 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 09:36:03.514655 137496769246976 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0422 09:36:03.514718 137496769246976 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 09:36:03.517589 137496769246976 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 09:36:03.517731 137638995367744 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 09:36:03.517830 137638995367744 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 09:36:03.518823 137638995367744 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.953, Tokens/s/device: 13860.126, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 09:36:12 UTC 2026 EXIT_CODE=0