main (586e69205) vs feat/nnx-post-train-fixes (7f06c99ac)| Metric | main 586e69205 | feat/nnx-post-train-fixes 7f06c99ac | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 7.1810 | 7.1810 | 0 |
| TFLOP/s | 92.172 | 92.175 | +0.003 |
| Tok/s | 13893.1 | 13893.6 | +0.471 |
| Avg s/step | 7.920 | 7.994 | +0.074 |
| 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: Thu Apr 23 08:36:49 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. 2026-04-23 08:37:14.653715: 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) I0423 08:37:14.867997 135368962934592 max_utils.py:273] Attempting to initialize the jax distributed system... I0423 08:37:23.909064 135368962934592 distributed.py:149] Starting JAX distributed service on [::]:8482 I0423 08:37:23.914573 135368962934592 distributed.py:172] Connecting to JAX distributed service on mt-16-zarr3-true-sv-pe65r-slice-job-0-0.mt-16-zarr3-true-sv-pe65r:8482 I0423 08:37:25.110413 135368962934592 max_utils.py:284] Jax distributed system initialized! I0423 08:37:30.309858 135368962934592 max_utils.py:800] System Information: Jax Version: 0.9.2 I0423 08:37:30.309963 135368962934592 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0423 08:37:30.310002 135368962934592 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 I0423 08:37:30.310035 135368962934592 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 08:37:31.005220 135368962934592 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 08:37:31.005502 135368962934592 checkpointing.py:677] Setting up checkpoint logger... I0423 08:37:31.005557 135368962934592 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0423 08:37:31.005601 135368962934592 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0423 08:37:31.005938 135368962934592 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 0x7b1d5324c860>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 08:37:33.853424 135368962934592 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0423 08:37:33.853663 135368962934592 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 0x7b09c42e4c20>}, handler_registry=None I0423 08:37:33.853906 135368962934592 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b09c42e4c20>` 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`. I0423 08:37:33.853953 135368962934592 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b09e41e3e00>` 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`. I0423 08:37:33.853989 135368962934592 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 0x7b09c42e4c20>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b09c42e4c20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b09e41e3e00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b09e41e3e00>}). I0423 08:37:33.854329 135368962934592 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0423 08:37:33.854399 135368962934592 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b09c4535800> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0423 08:37:35.494470 135368962934592 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints I0423 08:37:35.716478 135368962934592 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=5), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b09e427d2e0> I0423 08:37:35.716654 135368962934592 checkpointing.py:301] Checkpoint manager created! I0423 08:37:36.649889 135368962934592 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 08:37:36.650009 135368962934592 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:37:37.033569 135368962934592 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0423 08:37:37.033679 135368962934592 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:37:37.050327 135368962934592 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0423 08:37:37.050393 135368962934592 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 08:37:37.074406 135368962934592 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:37:37.074483 135368962934592 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:37:37.091107 135368962934592 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:37:37.091172 135368962934592 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:37:37.107553 135368962934592 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 08:37:37.107616 135368962934592 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:37:37.132596 135368962934592 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0423 08:37:37.132672 135368962934592 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 08:37:37.153937 135368962934592 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 08:37:37.154010 135368962934592 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0423 08:37:37.364036 135368962934592 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0423 08:37:37.364166 135368962934592 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0423 08:37:38.794706 135368962934592 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0423 08:37:38.794836 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 08:37:38.794895 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 08:37:38.794955 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0423 08:37:38.795007 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 08:37:38.795045 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 08:37:38.795109 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0423 08:37:38.795166 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0423 08:37:38.795206 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0423 08:37:38.795517 135368962934592 maxtext_utils.py:1707] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0423 08:37:38.795694 135368962934592 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 08:37:38.795769 135368962934592 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 08:37:39.291082 135368962934592 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 08:37:39.291186 135368962934592 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0423 08:37:39.306747 135368962934592 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 08:37:39.306805 135368962934592 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0423 08:37:50.182554 135368962934592 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. I0423 08:37:50.183390 135368962934592 metric_logger.py:301] number parameters: 1.104 billion I0423 08:38:01.543410 135368962934592 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0423 08:38:01.804739 135368962934592 checkpointing.py:776] Waited 0.26131272315979004 seconds for step 0 to finish before starting checkpointing. I0423 08:38:01.806924 135368962934592 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:38:01.808710 135368962934592 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0423 08:38:01.810072 135368962934592 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/0. I0423 08:38:02.581786 135368962934592 signaling_client.py:364] Using JaxDistributedSignalingClient I0423 08:38:02.582688 135368962934592 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:38:02.582743 135368962934592 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 I0423 08:38:02.863920 135368962934592 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.282207s I0423 08:38:02.864106 135368962934592 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.357 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28795337677001953 s) (per-host) I0423 08:38:02.864159 135368962934592 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.288029s (batch_requests_ready=0.002599s, total_serialization_initiated=0.285347s, others=0.000083s) I0423 08:38:02.864248 135368962934592 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.292202s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.292183) I0423 08:38:02.865057 135368962934592 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.06 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/0. I0423 08:38:02.865437 135243285657344 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:58:02.865400 I0423 08:38:02.867431 135368962934592 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0423 08:38:02.867744 135243235301120 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0423 08:38:02.867872 135368962934592 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776933481.8069065, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776933481.8087475, 'checkpointer_blocking_duration_secs': 1.0568439960479736, 'get_old_steps_start_time': 1776933482.8656182, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776933481.8052402, 'checkpoint_manager_blocking_duration_secs': 1.0625941753387451} I0423 08:38:02.868029 135368962934592 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0423 08:38:02.868078 135368962934592 max_utils.py:750] Memstats: After params initialized: I0423 08:38:02.868138 135368962934592 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0423 08:38:02.868171 135368962934592 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0423 08:38:02.868198 135368962934592 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0423 08:38:02.868224 135368962934592 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0423 08:38:03.188740 135368962934592 metric_logger.py:196] completed step: 0, seconds: 11.360, TFLOP/s/device: 1.196, Tokens/s/device: 180.284, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 08:38:03.363563 135368962934592 metric_logger.py:196] completed step: 1, seconds: 1.644, TFLOP/s/device: 8.265, Tokens/s/device: 1245.864, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 08:38:03.795979 135368962934592 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 490.989, Tokens/s/device: 74007.155, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0423 08:38:03.943283 135368962934592 metric_logger.py:196] completed step: 3, seconds: 0.433, TFLOP/s/device: 31.410, Tokens/s/device: 4734.428, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0423 08:38:03.947756 135368962934592 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0423 08:38:04.237398 135368962934592 checkpointing.py:776] Waited 0.2896089553833008 seconds for step 5 to finish before starting checkpointing. I0423 08:38:04.240462 135368962934592 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 08:38:06.292193 2764 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 08:38:08.677573 135243243693824 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/0/items/array_metadatas/process_6 I0423 08:38:39.717530 135243285657344 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 42.529 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.14133286476135 s) (per-host) I0423 08:38:39.717651 135243285657344 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.852103s. I0423 08:38:49.883186 135243285657344 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 47.017623s. I0423 08:38:49.883486 135243235301120 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 08:38:49.883630 135243235301120 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0423 08:38:49.883693 135243235301120 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:38:49.885507 135243235301120 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0423 08:38:49.885681 135368962934592 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. W0423 08:38:49.885813 135368962934592 checkpoint_manager.py:1452] Waiting for previous save to complete took 45.645351 seconds. If this number is high, consider checkpointing less frequently. I0423 08:38:49.887979 135368962934592 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5 I0423 08:38:49.890556 135368962934592 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/5. I0423 08:38:50.190853 135368962934592 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:38:50.190948 135368962934592 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 I0423 08:38:50.228307 135368962934592 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038539s I0423 08:38:50.228482 135368962934592 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 36.691 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.042041778564453125 s) (per-host) I0423 08:38:50.228533 135368962934592 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.042102s (batch_requests_ready=0.001857s, total_serialization_initiated=0.040178s, others=0.000067s) I0423 08:38:50.228623 135368962934592 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.047108s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=0.047093) I0423 08:38:50.229284 135368962934592 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/5. I0423 08:38:50.229536 135243243693824 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:58:50.229516 I0423 08:38:50.245387 135368962934592 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0423 08:38:50.245660 135243235301120 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0423 08:38:50.245820 135368962934592 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776933484.2404325, 'wait_for_prev_duration_secs': 45.64535140991211, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776933529.8880181, 'checkpointer_blocking_duration_secs': 0.34160709381103516, 'get_old_steps_start_time': 1776933530.2296457, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1776933484.238017, 'checkpoint_manager_blocking_duration_secs': 46.007768869400024} I0423 08:38:50.245934 135368962934592 checkpointing.py:408] Started an asynchronous checkpoint save for step 5 I0423 08:38:50.246734 135368962934592 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.878, Tokens/s/device: 13396.653, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0423 08:38:50.252558 135368962934592 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.246, Tokens/s/device: 13904.259, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0423 08:38:55.831786 135221366171392 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/5/items/array_metadatas/process_6 I0423 08:39:12.676117 135368962934592 metric_logger.py:196] completed step: 6, seconds: 46.303, TFLOP/s/device: 0.293, Tokens/s/device: 44.230, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0423 08:39:12.823347 135368962934592 metric_logger.py:196] completed step: 7, seconds: 22.277, TFLOP/s/device: 0.610, Tokens/s/device: 91.932, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0423 08:39:12.970734 135368962934592 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.309, Tokens/s/device: 13461.551, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0423 08:39:13.117765 135368962934592 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0423 08:39:13.118415 135368962934592 checkpointing.py:776] Waited 0.00067138671875 seconds for step 9 to finish before starting checkpointing. I0423 08:39:13.120330 135368962934592 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 08:39:33.423714 135243243693824 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.533 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.237234354019165 s) (per-host) I0423 08:39:33.423840 135243243693824 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.194236s. I0423 08:39:41.145125 135243243693824 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.915479s. I0423 08:39:41.145412 135243235301120 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 08:39:41.145539 135243235301120 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0423 08:39:41.145587 135243235301120 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:39:41.147131 135243235301120 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0423 08:39:41.147304 135368962934592 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5. W0423 08:39:41.147433 135368962934592 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.027107 seconds. If this number is high, consider checkpointing less frequently. I0423 08:39:41.151378 135368962934592 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0423 08:39:41.153325 135368962934592 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/9. I0423 08:39:41.454108 135368962934592 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 08:39:41.454200 135368962934592 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 I0423 08:39:41.487798 135368962934592 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034488s I0423 08:39:41.487969 135368962934592 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 40.528 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.038062334060668945 s) (per-host) I0423 08:39:41.488018 135368962934592 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038121s (batch_requests_ready=0.001739s, total_serialization_initiated=0.036318s, others=0.000064s) I0423 08:39:41.488124 135368962934592 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042347s (all_items=0.000012s, per_item={'items': '0.00001216'}, temp_paths=0.042335) I0423 08:39:41.488759 135368962934592 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/9. I0423 08:39:41.489092 135243235301120 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 08:59:41.489042 I0423 08:39:41.492862 135368962934592 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0423 08:39:41.493139 135241129780992 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0423 08:39:41.493296 135368962934592 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776933553.1202993, 'wait_for_prev_duration_secs': 28.027106523513794, 'time_between_consecutive_saves_sec': 23.23474907875061, 'checkpointer_blocking_start_time': 1776933581.1514218, 'checkpointer_blocking_duration_secs': 0.3378334045410156, 'get_old_steps_start_time': 1776933581.489271, 'get_old_steps_duration_secs': 2.47955322265625e-05, 'checkpoint_manager_blocking_start_time': 1776933553.1186552, 'checkpoint_manager_blocking_duration_secs': 28.37460994720459} I0423 08:39:41.493405 135368962934592 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0423 08:39:41.493448 135368962934592 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 08:39:46.950294 135221366171392 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_16_zarr3_true/checkpoints/9/items/array_metadatas/process_6 I0423 08:40:23.403280 135243235301120 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.651 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.95333290100098 s) (per-host) I0423 08:40:23.403383 135243235301120 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.914159s. I0423 08:40:30.514263 135243235301120 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.025023s. I0423 08:40:30.514549 135241129780992 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 08:40:30.514668 135241129780992 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0423 08:40:30.514711 135241129780992 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0423 08:40:30.516186 135241129780992 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0423 08:40:30.516354 135368962934592 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. I0423 08:40:30.516503 135368962934592 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 08:40:30.517484 135368962934592 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.172, Tokens/s/device: 13893.129, total_weights: 65536, loss: 7.181, lm_loss: 7.181, perplexity: 1314.495 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Thu Apr 23 08:40:39 UTC 2026 EXIT_CODE=0
XPK Start: Thu Apr 23 14:03:03 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. 2026-04-23 14:03:28.289385: 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) I0423 14:03:28.501512 134933083035456 max_utils.py:273] Attempting to initialize the jax distributed system... I0423 14:03:37.542666 134933083035456 distributed.py:149] Starting JAX distributed service on [::]:8482 I0423 14:03:37.544992 134933083035456 distributed.py:172] Connecting to JAX distributed service on mt-16-zarr3-true-sv-92839-slice-job-0-0.mt-16-zarr3-true-sv-92839:8482 I0423 14:03:38.569286 134933083035456 max_utils.py:284] Jax distributed system initialized! I0423 14:03:44.767813 134933083035456 max_utils.py:800] System Information: Jax Version: 0.9.2 I0423 14:03:44.767920 134933083035456 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0423 14:03:44.767961 134933083035456 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 I0423 14:03:44.767997 134933083035456 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 14:03:45.465431 134933083035456 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 14:03:45.466014 134933083035456 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 14:03:45.466208 134933083035456 checkpointing.py:688] Setting up checkpoint logger... I0423 14:03:45.466259 134933083035456 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0423 14:03:45.466303 134933083035456 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0423 14:03:45.466645 134933083035456 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 0x7ab7ed71c080>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 14:03:48.377109 134933083035456 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0423 14:03:48.377347 134933083035456 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 0x7aa3783ee9c0>}, handler_registry=None I0423 14:03:48.377591 134933083035456 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7aa3783ee9c0>` 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`. I0423 14:03:48.377641 134933083035456 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa3781f6570>` 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`. I0423 14:03:48.377677 134933083035456 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 0x7aa3783ee9c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7aa3783ee9c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa3781f6570>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7aa3781f6570>}). I0423 14:03:48.378012 134933083035456 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0423 14:03:48.378084 134933083035456 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7aa29466dd00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0423 14:03:49.128312 134933083035456 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints I0423 14:03:49.173269 134933083035456 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=5), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7aa3783ef230> I0423 14:03:49.173418 134933083035456 checkpointing.py:302] Checkpoint manager created! I0423 14:03:50.112799 134933083035456 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 14:03:50.112913 134933083035456 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 14:03:50.494811 134933083035456 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 14:03:50.494908 134933083035456 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 14:03:50.511407 134933083035456 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 14:03:50.511470 134933083035456 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 14:03:50.535467 134933083035456 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 14:03:50.535539 134933083035456 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 14:03:50.552082 134933083035456 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 14:03:50.552157 134933083035456 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 14:03:50.568582 134933083035456 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 14:03:50.568641 134933083035456 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 14:03:50.593540 134933083035456 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0423 14:03:50.593611 134933083035456 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 14:03:50.614450 134933083035456 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 14:03:50.614524 134933083035456 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0423 14:03:50.834131 134933083035456 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0423 14:03:50.834248 134933083035456 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0423 14:03:52.280550 134933083035456 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0423 14:03:52.280681 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 14:03:52.280736 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 14:03:52.280796 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0423 14:03:52.280857 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 14:03:52.280900 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 14:03:52.280954 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0423 14:03:52.281009 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0423 14:03:52.281051 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0423 14:03:52.281090 134933083035456 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0423 14:03:52.281163 134933083035456 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 14:03:52.281220 134933083035456 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 14:03:52.778636 134933083035456 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 14:03:52.778737 134933083035456 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0423 14:03:52.794391 134933083035456 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 14:03:52.794455 134933083035456 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0423 14:04:03.668822 134933083035456 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. I0423 14:04:03.669648 134933083035456 metric_logger.py:301] number parameters: 1.104 billion I0423 14:04:14.776298 134933083035456 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0423 14:04:14.960922 134933083035456 checkpointing.py:798] Waited 0.18460988998413086 seconds for step 0 to finish before starting checkpointing. I0423 14:04:14.963379 134933083035456 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 14:04:14.965208 134933083035456 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0423 14:04:14.966844 134933083035456 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/0. I0423 14:04:15.684252 134933083035456 signaling_client.py:364] Using JaxDistributedSignalingClient I0423 14:04:15.685316 134933083035456 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 14:04:15.685376 134933083035456 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 I0423 14:04:15.968423 134933083035456 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.284246s I0423 14:04:15.968606 134933083035456 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.319 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2899892330169678 s) (per-host) I0423 14:04:15.968661 134933083035456 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.290053s (batch_requests_ready=0.002491s, total_serialization_initiated=0.287490s, others=0.000072s) I0423 14:04:15.968751 134933083035456 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.294150s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.294132) I0423 14:04:15.969582 134933083035456 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.00 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/0. I0423 14:04:15.969928 134803882694400 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:24:15.969891 I0423 14:04:16.391288 134933083035456 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0423 14:04:16.391687 134802841790208 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0423 14:04:16.391859 134933083035456 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776953054.9633603, 'wait_for_prev_duration_secs': 6.914138793945312e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776953054.9652534, 'checkpointer_blocking_duration_secs': 1.004835605621338, 'get_old_steps_start_time': 1776953055.9701269, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776953054.9614353, 'checkpoint_manager_blocking_duration_secs': 1.430379867553711} I0423 14:04:16.391964 134933083035456 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0423 14:04:16.392015 134933083035456 max_utils.py:750] Memstats: After params initialized: I0423 14:04:16.392072 134933083035456 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0423 14:04:16.392120 134933083035456 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0423 14:04:16.392151 134933083035456 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0423 14:04:16.392177 134933083035456 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0423 14:04:16.707343 134933083035456 metric_logger.py:196] completed step: 0, seconds: 11.107, TFLOP/s/device: 1.223, Tokens/s/device: 184.396, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 14:04:16.889810 134933083035456 metric_logger.py:196] completed step: 1, seconds: 1.930, TFLOP/s/device: 7.042, Tokens/s/device: 1061.377, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0423 14:04:17.309514 134933083035456 metric_logger.py:196] completed step: 2, seconds: 0.035, TFLOP/s/device: 384.719, Tokens/s/device: 57989.070, total_weights: 65536, loss: 10.560, lm_loss: 10.560, perplexity: 38547.719 I0423 14:04:17.456891 134933083035456 metric_logger.py:196] completed step: 3, seconds: 0.420, TFLOP/s/device: 32.335, Tokens/s/device: 4873.881, total_weights: 65536, loss: 9.980, lm_loss: 9.980, perplexity: 21590.506 I0423 14:04:17.461414 134933083035456 checkpointing.py:794] Waiting for step 5 to finish before checkpoint... I0423 14:04:17.750926 134933083035456 checkpointing.py:798] Waited 0.28948068618774414 seconds for step 5 to finish before starting checkpointing. I0423 14:04:17.753955 134933083035456 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 14:04:19.067247 2811 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 14:04:21.389004 134802850182912 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/0/items/array_metadatas/process_6 I0423 14:04:58.440325 134803882694400 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.940 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.761675119400024 s) (per-host) I0423 14:04:58.440431 134803882694400 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.470387s. I0423 14:05:03.566476 134803882694400 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 47.596416s. I0423 14:05:03.566807 134802841790208 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 14:05:03.566930 134802841790208 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0423 14:05:03.566983 134802841790208 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0423 14:05:03.568750 134802841790208 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0423 14:05:03.568942 134933083035456 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. W0423 14:05:03.569109 134933083035456 checkpoint_manager.py:1452] Waiting for previous save to complete took 45.815136 seconds. If this number is high, consider checkpointing less frequently. I0423 14:05:03.571027 134933083035456 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 5 I0423 14:05:03.573648 134933083035456 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/5. I0423 14:05:03.874556 134933083035456 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 14:05:03.874647 134933083035456 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 I0423 14:05:03.910978 134933083035456 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037511s I0423 14:05:03.911157 134933083035456 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 37.697 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04092001914978027 s) (per-host) I0423 14:05:03.911209 134933083035456 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040982s (batch_requests_ready=0.001720s, total_serialization_initiated=0.039193s, others=0.000069s) I0423 14:05:03.911295 134933083035456 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.045944s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.045927) I0423 14:05:03.912024 134933083035456 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/5. I0423 14:05:03.912373 134802841790208 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:25:03.912337 I0423 14:05:03.919947 134933083035456 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0423 14:05:03.920222 134797967988480 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0423 14:05:03.920385 134933083035456 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776953057.7539153, 'wait_for_prev_duration_secs': 45.815136432647705, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776953103.5710664, 'checkpointer_blocking_duration_secs': 0.3414475917816162, 'get_old_steps_start_time': 1776953103.9125392, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776953057.7517695, 'checkpoint_manager_blocking_duration_secs': 46.168580293655396} I0423 14:05:03.920502 134933083035456 checkpointing.py:409] Started an asynchronous checkpoint save for step 5 I0423 14:05:03.921258 134933083035456 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 89.061, Tokens/s/device: 13424.227, total_weights: 65536, loss: 9.460, lm_loss: 9.460, perplexity: 12840.355 I0423 14:05:03.927287 134933083035456 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.204, Tokens/s/device: 13897.937, total_weights: 65536, loss: 8.959, lm_loss: 8.959, perplexity: 7776.467 I0423 14:05:09.975959 134802850182912 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/5/items/array_metadatas/process_6 I0423 14:05:26.568609 134933083035456 metric_logger.py:196] completed step: 6, seconds: 46.464, TFLOP/s/device: 0.292, Tokens/s/device: 44.077, total_weights: 65536, loss: 8.469, lm_loss: 8.469, perplexity: 4762.740 I0423 14:05:26.716029 134933083035456 metric_logger.py:196] completed step: 7, seconds: 22.495, TFLOP/s/device: 0.604, Tokens/s/device: 91.041, total_weights: 65536, loss: 8.003, lm_loss: 8.003, perplexity: 2989.151 I0423 14:05:26.863371 134933083035456 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.384, Tokens/s/device: 13472.886, total_weights: 65536, loss: 7.572, lm_loss: 7.572, perplexity: 1943.098 I0423 14:05:27.010009 134933083035456 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0423 14:05:27.010725 134933083035456 checkpointing.py:798] Waited 0.0007357597351074219 seconds for step 9 to finish before starting checkpointing. I0423 14:05:27.012741 134933083035456 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 14:05:45.796218 134802841790208 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.676 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.925939321517944 s) (per-host) I0423 14:05:45.796335 134802841790208 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.883849s. I0423 14:05:54.592667 134802841790208 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.680166s. I0423 14:05:54.592957 134797967988480 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 14:05:54.593082 134797967988480 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0423 14:05:54.593148 134797967988480 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0423 14:05:54.594655 134797967988480 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0423 14:05:54.594840 134933083035456 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5. W0423 14:05:54.594970 134933083035456 checkpoint_manager.py:1452] Waiting for previous save to complete took 27.582234 seconds. If this number is high, consider checkpointing less frequently. I0423 14:05:54.596437 134933083035456 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0423 14:05:54.598587 134933083035456 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/9. I0423 14:05:55.312863 134933083035456 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 14:05:55.312955 134933083035456 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 I0423 14:05:55.349196 134933083035456 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037219s I0423 14:05:55.349364 134933083035456 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 38.000 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.040593862533569336 s) (per-host) I0423 14:05:55.349414 134933083035456 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040653s (batch_requests_ready=0.001631s, total_serialization_initiated=0.038958s, others=0.000064s) I0423 14:05:55.349498 134933083035456 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044837s (all_items=0.000012s, per_item={'items': '0.00001168'}, temp_paths=0.044825) I0423 14:05:55.350169 134933083035456 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/9. I0423 14:05:55.350505 134797967988480 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 14:25:55.350468 I0423 14:05:55.354761 134933083035456 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0423 14:05:55.355046 134803370219264 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0423 14:05:55.355215 134933083035456 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776953127.0127125, 'wait_for_prev_duration_secs': 27.582234144210815, 'time_between_consecutive_saves_sec': 23.443918704986572, 'checkpointer_blocking_start_time': 1776953154.5964806, 'checkpointer_blocking_duration_secs': 0.7541713714599609, 'get_old_steps_start_time': 1776953155.35067, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1776953127.0109508, 'checkpoint_manager_blocking_duration_secs': 28.34423303604126} I0423 14:05:55.355324 134933083035456 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0423 14:05:55.355368 134933083035456 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 14:06:00.477240 134800123864832 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_16_zarr3_true/checkpoints/9/items/array_metadatas/process_6 I0423 14:06:36.309747 134797967988480 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.526 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.00093746185303 s) (per-host) I0423 14:06:36.309869 134797967988480 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.959252s. I0423 14:06:44.785131 134797967988480 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.434481s. I0423 14:06:44.785425 134803370219264 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 14:06:44.785547 134803370219264 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0423 14:06:44.785590 134803370219264 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0423 14:06:44.787333 134803370219264 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0423 14:06:44.787511 134933083035456 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. I0423 14:06:44.787665 134933083035456 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 14:06:44.788605 134933083035456 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.175, Tokens/s/device: 13893.600, total_weights: 65536, loss: 7.181, lm_loss: 7.181, perplexity: 1314.495 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Thu Apr 23 14:06:53 UTC 2026 EXIT_CODE=0