main (8a17c3d19) vs feat/nnx-post-train-fixes (7f06c99ac)| Metric | main 8a17c3d19 | feat/nnx-post-train-fixes 7f06c99ac | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 5.3910 | 5.3910 | 0 |
| TFLOP/s | 92.299 | 92.030 | -0.269 |
| Tok/s | 13912.3 | 13871.8 | -40.52 |
| Avg s/step | 8.483 | 8.503 | +0.02 |
| 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 08:18:41 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 08:19:06.143023: 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 08:19:06.353135 135788806346560 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 08:19:15.394039 135788806346560 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 08:19:15.398524 135788806346560 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-true-rs-a1mv7-slice-job-0-0.mt-15-ocdbt-true-rs-a1mv7:8482 I0422 08:19:16.640761 135788806346560 max_utils.py:284] Jax distributed system initialized! I0422 08:19:22.860661 135788806346560 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 08:19:22.860773 135788806346560 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 08:19:22.860815 135788806346560 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 08:19:22.860850 135788806346560 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 08:19:23.550011 135788806346560 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 08:19:23.550301 135788806346560 checkpointing.py:677] Setting up checkpoint logger... I0422 08:19:23.550357 135788806346560 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 08:19:23.550401 135788806346560 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 08:19:23.550757 135788806346560 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 0x7b7f3291ce00>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 08:19:24.876303 135788806346560 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 08:19:24.876550 135788806346560 checkpoint_manager.py:708] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b6af8751d30>}, handler_registry=None I0422 08:19:24.876790 135788806346560 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b6af8751d30>` 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 08:19:24.876839 135788806346560 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b6af8753a40>` 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 08:19:24.876876 135788806346560 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 0x7b6af8751d30>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b6af8751d30>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b6af8753a40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b6af8753a40>}). I0422 08:19:24.877219 135788806346560 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 08:19:24.877291 135788806346560 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b6a08341580> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 08:19:27.220639 135668126697216 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776845753162056387, 'commit_timestamp_nsecs': 1776845798312294499, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/5/_CHECKPOINT_METADATA I0422 08:19:27.837430 135668118304512 checkpoint_manager.py:1871] Missing metrics for step 9 I0422 08:19:27.884948 135788806346560 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints I0422 08:19:29.564528 135788806346560 checkpoint_manager.py:929] [process=3][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=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_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b6b081f83b0> I0422 08:19:29.564707 135788806346560 checkpointing.py:301] Checkpoint manager created! I0422 08:19:30.487247 135788806346560 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 08:19:30.487370 135788806346560 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 08:19:30.865985 135788806346560 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 08:19:30.866085 135788806346560 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 08:19:30.882581 135788806346560 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 08:19:30.882645 135788806346560 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 08:19:30.906646 135788806346560 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 08:19:30.906719 135788806346560 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 08:19:30.923339 135788806346560 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 08:19:30.923403 135788806346560 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 08:19:30.939968 135788806346560 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 08:19:30.940029 135788806346560 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 08:19:30.964985 135788806346560 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 08:19:30.965057 135788806346560 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 08:19:30.986623 135788806346560 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 08:19:30.986696 135788806346560 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 08:19:31.197517 135788806346560 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 08:19:31.197630 135788806346560 checkpointing.py:581] restoring from this run's directory step 9 I0422 08:19:31.463073 135788806346560 event_tracking.py:70] [process=3] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/9. I0422 08:19:31.713401 135788806346560 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/9. I0422 08:19:31.713842 135788806346560 event_tracking.py:125] [process=3] [sync] Finished blocking load in 0.25 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/9. I0422 08:19:33.263354 1443 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 08:19:39.177110 135788806346560 jax_array_handlers.py:843] [process=3] /jax/orbax/read/worker/io/requested throughput: 266.945 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 5.919855356216431 s) (per-host) I0422 08:19:39.183645 135788806346560 base_pytree_checkpoint_handler.py:130] [process=3] /jax/checkpoint/read/gbytes_per_sec: 250.081 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.328320026397705 s) (per-host) I0422 08:19:41.550178 135788806346560 event_tracking.py:138] [process=3] [sync] Finished load in 10.09 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/9 I0422 08:19:41.550339 135788806346560 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints', 'checkpointer_start_time': 1776845971.462962, 'checkpointer_duration_secs': 10.087344884872437, 'checkpoint_manager_start_time': 1776845971.1978726, 'checkpoint_manager_duration_secs': 10.352437019348145} fsdp: 32 I0422 08:19:41.971461 135788806346560 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 08:19:41.971596 135788806346560 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 08:19:41.971651 135788806346560 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 08:19:41.971710 135788806346560 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 08:19:41.971763 135788806346560 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 08:19:41.971803 135788806346560 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 08:19:41.971855 135788806346560 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 08:19:41.971906 135788806346560 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 08:19:41.971960 135788806346560 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 08:19:41.972000 135788806346560 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 08:19:41.972047 135788806346560 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 08:19:41.972094 135788806346560 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 08:19:42.460472 135788806346560 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 08:19:42.460575 135788806346560 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 08:19:42.476595 135788806346560 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 08:19:42.476655 135788806346560 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 08:19:53.349223 135788806346560 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 08:19:53.350072 135788806346560 metric_logger.py:301] number parameters: 1.104 billion I0422 08:20:04.857577 135788806346560 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0422 08:20:05.396633 135788806346560 checkpointing.py:776] Waited 0.5390346050262451 seconds for step 10 to finish before starting checkpointing. I0422 08:20:05.399139 135788806346560 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 08:20:05.401277 135788806346560 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 10 I0422 08:20:05.402732 135788806346560 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/10. I0422 08:20:06.145232 135788806346560 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 08:20:06.146152 135788806346560 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 08:20:06.146207 135788806346560 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 08:20:06.426208 135788806346560 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.281051s I0422 08:20:06.426383 135788806346560 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 5.412 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2850346565246582 s) (per-host) I0422 08:20:06.426439 135788806346560 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.285101s (batch_requests_ready=0.002266s, total_serialization_initiated=0.282760s, others=0.000075s) I0422 08:20:06.426536 135788806346560 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.289234s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.289218) I0422 08:20:06.427250 135788806346560 event_tracking.py:125] [process=3] [async] Finished blocking save in 1.03 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/10. I0422 08:20:06.427597 135667547895552 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 08:40:06.427560 I0422 08:20:06.433571 135788806346560 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0422 08:20:06.433821 135665928890112 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0422 08:20:06.433960 135788806346560 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776846005.3991222, 'wait_for_prev_duration_secs': 6.079673767089844e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776846005.4013193, 'checkpointer_blocking_duration_secs': 1.0264294147491455, 'get_old_steps_start_time': 1776846006.427773, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776846005.3971617, 'checkpoint_manager_blocking_duration_secs': 1.0367467403411865} I0422 08:20:06.434066 135788806346560 checkpointing.py:408] Started an asynchronous checkpoint save for step 10 I0422 08:20:06.434118 135788806346560 max_utils.py:750] Memstats: After params initialized: I0422 08:20:06.434166 135788806346560 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0)) I0422 08:20:06.434201 135788806346560 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0)) I0422 08:20:06.434229 135788806346560 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0)) I0422 08:20:06.434255 135788806346560 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0)) I0422 08:20:06.743119 135788806346560 metric_logger.py:196] completed step: 10, seconds: 11.507, TFLOP/s/device: 1.181, Tokens/s/device: 177.973, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869 I0422 08:20:07.018950 135788806346560 metric_logger.py:196] completed step: 11, seconds: 1.884, TFLOP/s/device: 7.212, Tokens/s/device: 1087.063, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659 I0422 08:20:07.166318 135788806346560 metric_logger.py:196] completed step: 12, seconds: 0.129, TFLOP/s/device: 105.650, Tokens/s/device: 15924.731, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068 I0422 08:20:07.313866 135788806346560 metric_logger.py:196] completed step: 13, seconds: 0.153, TFLOP/s/device: 88.618, Tokens/s/device: 13357.422, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001 I0422 08:20:07.318419 135788806346560 checkpointing.py:772] Waiting for step 15 to finish before checkpoint... I0422 08:20:07.607613 135788806346560 checkpointing.py:776] Waited 0.28915858268737793 seconds for step 15 to finish before starting checkpointing. I0422 08:20:07.610754 135788806346560 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 08:20:12.100668 135665937282816 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/10/items/array_metadatas/process_3 I0422 08:20:48.521658 135667547895552 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.272 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.38027381896973 s) (per-host) I0422 08:20:48.521789 135667547895552 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.094083s. I0422 08:20:55.946271 135667547895552 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 49.518548s. I0422 08:20:55.946564 135665928890112 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 08:20:55.946688 135665928890112 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0422 08:20:55.946735 135665928890112 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0422 08:20:55.948509 135665928890112 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0422 08:20:55.948676 135788806346560 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0422 08:20:55.948812 135788806346560 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.338065 seconds. If this number is high, consider checkpointing less frequently. I0422 08:20:55.950635 135788806346560 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 15 I0422 08:20:55.953122 135788806346560 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/15. I0422 08:20:56.670698 135788806346560 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 08:20:56.670797 135788806346560 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 08:20:56.717080 135788806346560 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.047504s I0422 08:20:56.717243 135788806346560 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 30.221 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.051043033599853516 s) (per-host) I0422 08:20:56.717293 135788806346560 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.051101s (batch_requests_ready=0.001896s, total_serialization_initiated=0.049139s, others=0.000067s) I0422 08:20:56.717381 135788806346560 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.056150s (all_items=0.000013s, per_item={'items': '0.00001287'}, temp_paths=0.056138) I0422 08:20:56.718141 135788806346560 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/15. I0422 08:20:56.718489 135667539502848 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 08:40:56.718449 I0422 08:20:56.720465 135788806346560 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0422 08:20:56.720769 135665928890112 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0422 08:20:56.720888 135788806346560 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776846007.6107175, 'wait_for_prev_duration_secs': 48.338064670562744, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776846055.950674, 'checkpointer_blocking_duration_secs': 0.7679681777954102, 'get_old_steps_start_time': 1776846056.7186658, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776846007.6082897, 'checkpoint_manager_blocking_duration_secs': 49.11256718635559} I0422 08:20:56.721009 135788806346560 checkpointing.py:408] Started an asynchronous checkpoint save for step 15 I0422 08:20:56.721904 135788806346560 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 92.083, Tokens/s/device: 13879.758, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193 I0422 08:20:56.728123 135788806346560 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.259, Tokens/s/device: 13906.336, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661 I0422 08:21:01.726833 135665937282816 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/15/items/array_metadatas/process_3 I0422 08:21:18.025433 135788806346560 metric_logger.py:196] completed step: 16, seconds: 49.408, TFLOP/s/device: 0.275, Tokens/s/device: 41.451, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872 I0422 08:21:18.172687 135788806346560 metric_logger.py:196] completed step: 17, seconds: 21.151, TFLOP/s/device: 0.642, Tokens/s/device: 96.826, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975 I0422 08:21:18.320041 135788806346560 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.301, Tokens/s/device: 13460.489, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333 I0422 08:21:18.466668 135788806346560 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0422 08:21:18.467426 135788806346560 checkpointing.py:776] Waited 0.0007765293121337891 seconds for step 19 to finish before starting checkpointing. I0422 08:21:18.469439 135788806346560 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 08:21:38.674385 135667539502848 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.602 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.008145570755005 s) (per-host) I0422 08:21:38.674523 135667539502848 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.955910s. I0422 08:21:44.383018 135667539502848 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 47.664389s. I0422 08:21:44.383262 135665928890112 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 08:21:44.383326 135665928890112 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0422 08:21:44.383382 135665928890112 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0422 08:21:44.384845 135665928890112 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0422 08:21:44.385012 135788806346560 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15. W0422 08:21:44.385135 135788806346560 checkpoint_manager.py:1452] Waiting for previous save to complete took 25.915713 seconds. If this number is high, consider checkpointing less frequently. I0422 08:21:44.386627 135788806346560 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 19 I0422 08:21:44.388686 135788806346560 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/19. I0422 08:21:45.094586 135788806346560 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 08:21:45.094680 135788806346560 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 08:21:45.139824 135788806346560 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.046195s I0422 08:21:45.140004 135788806346560 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 30.970 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04980874061584473 s) (per-host) I0422 08:21:45.140057 135788806346560 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.049872s (batch_requests_ready=0.001768s, total_serialization_initiated=0.048035s, others=0.000069s) I0422 08:21:45.140146 135788806346560 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.054147s (all_items=0.000011s, per_item={'items': '0.00001121'}, temp_paths=0.054136) I0422 08:21:45.140870 135788806346560 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/19. I0422 08:21:45.141230 135662674093824 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 08:41:45.141191 I0422 08:21:45.143111 135788806346560 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0422 08:21:45.143493 135665928890112 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0422 08:21:45.143623 135788806346560 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776846078.4693987, 'wait_for_prev_duration_secs': 25.91571283340454, 'time_between_consecutive_saves_sec': 22.520848512649536, 'checkpointer_blocking_start_time': 1776846104.3866677, 'checkpointer_blocking_duration_secs': 0.7547128200531006, 'get_old_steps_start_time': 1776846105.1413987, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1776846078.467665, 'checkpoint_manager_blocking_duration_secs': 26.675926446914673} I0422 08:21:45.143788 135788806346560 checkpointing.py:408] Started an asynchronous checkpoint save for step 19 I0422 08:21:45.143834 135788806346560 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 08:21:50.449974 135665937282816 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_15_ocdbt_true/checkpoints/19/items/array_metadatas/process_3 I0422 08:22:27.313747 135662674093824 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.410 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.2235312461853 s) (per-host) I0422 08:22:27.313854 135662674093824 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.172504s. I0422 08:22:34.056103 135662674093824 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 48.914736s. I0422 08:22:34.056395 135665928890112 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 08:22:34.056523 135665928890112 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0422 08:22:34.056565 135665928890112 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0422 08:22:34.061542 135665928890112 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0422 08:22:34.061728 135788806346560 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0422 08:22:34.061879 135788806346560 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 08:22:34.062740 135788806346560 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.299, Tokens/s/device: 13912.287, total_weights: 65536, loss: 5.391, lm_loss: 5.391, perplexity: 219.349 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 08:22:46 UTC 2026 EXIT_CODE=0
XPK Start: Wed Apr 22 13:45:30 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-22 13:45:54.270273: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0422 13:45:54.449308 135837790684992 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 13:46:03.490942 135837790684992 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 13:46:03.493331 135837790684992 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-true-rs-pb7hg-slice-job-0-0.mt-15-ocdbt-true-rs-pb7hg:8482 I0422 13:46:05.335506 135837790684992 max_utils.py:284] Jax distributed system initialized! I0422 13:46:11.621773 135837790684992 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 13:46:11.621878 135837790684992 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 13:46:11.621918 135837790684992 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0422 13:46:11.621950 135837790684992 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 13:46:12.688336 135837790684992 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 13:46:12.688920 135837790684992 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 13:46:12.689117 135837790684992 checkpointing.py:688] Setting up checkpoint logger... I0422 13:46:12.689170 135837790684992 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 13:46:12.689212 135837790684992 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 13:46:12.689557 135837790684992 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 0x7b8a99f18620>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 13:46:13.926952 135837790684992 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0422 13:46:13.927315 135837790684992 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 0x7b767c1740b0>}, handler_registry=None I0422 13:46:13.927561 135837790684992 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b767c1740b0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0422 13:46:13.927610 135837790684992 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b896fc3b470>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0422 13:46:13.927646 135837790684992 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 0x7b767c1740b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b767c1740b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b896fc3b470>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b896fc3b470>}). I0422 13:46:13.927960 135837790684992 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 13:46:13.928029 135837790684992 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b764c3539c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 13:46:15.399838 135715824314112 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776865309653716616, 'commit_timestamp_nsecs': 1776865352059399770, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0422 13:46:15.651266 135715262297856 checkpoint_manager.py:1871] Missing metrics for step 9 I0422 13:46:15.664289 135837790684992 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints I0422 13:46:16.135668 135837790684992 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_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b896fc0cb00> I0422 13:46:16.135840 135837790684992 checkpointing.py:302] Checkpoint manager created! I0422 13:46:17.077551 135837790684992 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 13:46:17.077656 135837790684992 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 13:46:17.454839 135837790684992 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 13:46:17.454931 135837790684992 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 13:46:17.471419 135837790684992 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 13:46:17.471476 135837790684992 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 13:46:17.495280 135837790684992 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 13:46:17.495348 135837790684992 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 13:46:17.511778 135837790684992 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 13:46:17.511838 135837790684992 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 13:46:17.528275 135837790684992 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 13:46:17.528332 135837790684992 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 13:46:17.552962 135837790684992 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 13:46:17.553032 135837790684992 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 13:46:17.575114 135837790684992 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 13:46:17.575179 135837790684992 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 13:46:17.789214 135837790684992 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0422 13:46:17.789323 135837790684992 checkpointing.py:582] restoring from this run's directory step 9 I0422 13:46:18.051668 135837790684992 event_tracking.py:70] [process=6] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/9. I0422 13:46:18.293586 135837790684992 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/9. I0422 13:46:18.294007 135837790684992 event_tracking.py:125] [process=6] [sync] Finished blocking load in 0.24 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/9. I0422 13:46:19.716697 1464 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 13:46:25.941365 135837790684992 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 253.635 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.230514049530029 s) (per-host) I0422 13:46:25.946257 135837790684992 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 238.028 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.648773193359375 s) (per-host) I0422 13:46:26.725732 135837790684992 event_tracking.py:138] [process=6] [sync] Finished load in 8.67 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/9 I0422 13:46:26.725883 135837790684992 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints', 'checkpointer_start_time': 1776865578.0515702, 'checkpointer_duration_secs': 8.674280881881714, 'checkpoint_manager_start_time': 1776865577.7895696, 'checkpoint_manager_duration_secs': 8.936283826828003} fsdp: 32 I0422 13:46:27.139480 135837790684992 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 13:46:27.139607 135837790684992 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 13:46:27.139659 135837790684992 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 13:46:27.139714 135837790684992 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 13:46:27.139767 135837790684992 maxtext_utils.py:1827] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 13:46:27.139805 135837790684992 maxtext_utils.py:1827] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 13:46:27.139858 135837790684992 maxtext_utils.py:1827] 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 13:46:27.139911 135837790684992 maxtext_utils.py:1827] 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 13:46:27.139950 135837790684992 maxtext_utils.py:1827] 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 13:46:27.139986 135837790684992 maxtext_utils.py:1827] 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 13:46:27.140030 135837790684992 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 13:46:27.140075 135837790684992 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 13:46:27.623553 135837790684992 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 13:46:27.623649 135837790684992 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 13:46:27.639259 135837790684992 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 13:46:27.639324 135837790684992 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 13:46:38.776639 135837790684992 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 13:46:38.777425 135837790684992 metric_logger.py:301] number parameters: 1.104 billion I0422 13:46:50.024170 135837790684992 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0422 13:46:50.179499 135837790684992 checkpointing.py:798] Waited 0.15530943870544434 seconds for step 10 to finish before starting checkpointing. I0422 13:46:50.181857 135837790684992 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 13:46:50.183562 135837790684992 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10 I0422 13:46:50.185028 135837790684992 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/10. I0422 13:46:50.512914 135837790684992 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 13:46:50.513909 135837790684992 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 13:46:50.513964 135837790684992 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0422 13:46:50.790721 135837790684992 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.277887s I0422 13:46:50.790896 135837790684992 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.476 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2816939353942871 s) (per-host) I0422 13:46:50.790949 135837790684992 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.281759s (batch_requests_ready=0.002135s, total_serialization_initiated=0.279552s, others=0.000073s) I0422 13:46:50.791038 135837790684992 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.286142s (all_items=0.000038s, per_item={'items': '0.00003839'}, temp_paths=0.286103) I0422 13:46:50.791788 135837790684992 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/10. I0422 13:46:50.792151 135715228727040 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 14:06:50.792112 I0422 13:46:50.798364 135837790684992 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0422 13:46:50.798612 135711465252608 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0422 13:46:50.798761 135837790684992 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776865610.1818388, 'wait_for_prev_duration_secs': 5.9604644775390625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776865610.1835995, 'checkpointer_blocking_duration_secs': 0.608708381652832, 'get_old_steps_start_time': 1776865610.792332, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776865610.1800065, 'checkpoint_manager_blocking_duration_secs': 0.6187193393707275} I0422 13:46:50.798864 135837790684992 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0422 13:46:50.798912 135837790684992 max_utils.py:750] Memstats: After params initialized: I0422 13:46:50.798959 135837790684992 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0422 13:46:50.798992 135837790684992 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0422 13:46:50.799016 135837790684992 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0422 13:46:50.799041 135837790684992 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0422 13:46:51.109603 135837790684992 metric_logger.py:196] completed step: 10, seconds: 11.247, TFLOP/s/device: 1.208, Tokens/s/device: 182.099, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869 I0422 13:46:51.691529 135837790684992 metric_logger.py:196] completed step: 11, seconds: 1.084, TFLOP/s/device: 12.535, Tokens/s/device: 1889.405, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659 I0422 13:46:51.838999 135837790684992 metric_logger.py:196] completed step: 12, seconds: 0.436, TFLOP/s/device: 31.151, Tokens/s/device: 4695.385, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068 I0422 13:46:51.986448 135837790684992 metric_logger.py:196] completed step: 13, seconds: 0.152, TFLOP/s/device: 89.576, Tokens/s/device: 13501.932, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001 I0422 13:46:51.991117 135837790684992 checkpointing.py:794] Waiting for step 15 to finish before checkpoint... I0422 13:46:52.280645 135837790684992 checkpointing.py:798] Waited 0.28951215744018555 seconds for step 15 to finish before starting checkpointing. I0422 13:46:52.284320 135837790684992 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 13:46:56.058692 135713626507008 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_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/10/items/array_metadatas/process_6 I0422 13:47:32.433703 135715228727040 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.677 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.92447853088379 s) (per-host) I0422 13:47:32.433808 135715228727040 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.641543s. I0422 13:47:42.429001 135715228727040 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 51.636718s. I0422 13:47:42.429352 135711465252608 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 13:47:42.429474 135711465252608 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0422 13:47:42.429522 135711465252608 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0422 13:47:42.430805 135711465252608 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0422 13:47:42.430994 135837790684992 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0422 13:47:42.431149 135837790684992 checkpoint_manager.py:1452] Waiting for previous save to complete took 50.146826 seconds. If this number is high, consider checkpointing less frequently. I0422 13:47:42.432982 135837790684992 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15 I0422 13:47:42.435512 135837790684992 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/15. I0422 13:47:42.736530 135837790684992 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 13:47:42.736627 135837790684992 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0422 13:47:42.777676 135837790684992 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042202s I0422 13:47:42.777843 135837790684992 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 33.784 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.045659780502319336 s) (per-host) I0422 13:47:42.777892 135837790684992 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045717s (batch_requests_ready=0.001773s, total_serialization_initiated=0.043879s, others=0.000066s) I0422 13:47:42.777981 135837790684992 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050709s (all_items=0.000012s, per_item={'items': '0.00001240'}, temp_paths=0.050697) I0422 13:47:42.778713 135837790684992 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/15. I0422 13:47:42.779061 135711465252608 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 14:07:42.779024 I0422 13:47:42.781082 135837790684992 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize I0422 13:47:42.781390 135713626507008 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0422 13:47:42.781552 135837790684992 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776865612.2842913, 'wait_for_prev_duration_secs': 50.14682626724243, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776865662.433023, 'checkpointer_blocking_duration_secs': 0.34615612030029297, 'get_old_steps_start_time': 1776865662.779204, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776865612.2812965, 'checkpoint_manager_blocking_duration_secs': 50.50022029876709} I0422 13:47:42.781666 135837790684992 checkpointing.py:409] Started an asynchronous checkpoint save for step 15 I0422 13:47:42.782487 135837790684992 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 92.062, Tokens/s/device: 13876.561, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193 I0422 13:47:42.788858 135837790684992 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.148, Tokens/s/device: 13889.548, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661 I0422 13:47:47.932001 135714134013696 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_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/15/items/array_metadatas/process_6 I0422 13:48:03.658204 135837790684992 metric_logger.py:196] completed step: 16, seconds: 50.796, TFLOP/s/device: 0.267, Tokens/s/device: 40.318, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872 I0422 13:48:03.805549 135837790684992 metric_logger.py:196] completed step: 17, seconds: 20.723, TFLOP/s/device: 0.656, Tokens/s/device: 98.826, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975 I0422 13:48:03.952849 135837790684992 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.296, Tokens/s/device: 13459.693, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333 I0422 13:48:04.099486 135837790684992 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0422 13:48:04.100273 135837790684992 checkpointing.py:798] Waited 0.000804901123046875 seconds for step 19 to finish before starting checkpointing. I0422 13:48:04.102422 135837790684992 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 13:48:25.211611 135711465252608 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.185 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.479371786117554 s) (per-host) I0422 13:48:25.211737 135711465252608 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.432024s. I0422 13:48:33.438197 135711465252608 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.658470s. I0422 13:48:33.438528 135713626507008 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 13:48:33.438663 135713626507008 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0422 13:48:33.438710 135713626507008 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts... I0422 13:48:33.439980 135713626507008 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts. I0422 13:48:33.440173 135837790684992 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=15][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=15. W0422 13:48:33.440297 135837790684992 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.337880 seconds. If this number is high, consider checkpointing less frequently. I0422 13:48:33.441711 135837790684992 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19 I0422 13:48:33.444130 135837790684992 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/19. I0422 13:48:33.732746 135837790684992 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 13:48:33.732835 135837790684992 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0422 13:48:33.776740 135837790684992 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.044935s I0422 13:48:33.776898 135837790684992 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 31.904 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04835104942321777 s) (per-host) I0422 13:48:33.776949 135837790684992 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.048413s (batch_requests_ready=0.001635s, total_serialization_initiated=0.046712s, others=0.000066s) I0422 13:48:33.777038 135837790684992 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.052369s (all_items=0.000010s, per_item={'items': '0.00000978'}, temp_paths=0.052359) I0422 13:48:33.777763 135837790684992 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_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/19. I0422 13:48:33.778048 135713626507008 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 14:08:33.778015 I0422 13:48:33.780011 135837790684992 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0422 13:48:33.780269 135714134013696 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0422 13:48:33.780395 135837790684992 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776865684.1023943, 'wait_for_prev_duration_secs': 29.33787989616394, 'time_between_consecutive_saves_sec': 21.671539068222046, 'checkpointer_blocking_start_time': 1776865713.4417481, 'checkpointer_blocking_duration_secs': 0.3364078998565674, 'get_old_steps_start_time': 1776865713.778174, 'get_old_steps_duration_secs': 2.7179718017578125e-05, 'checkpoint_manager_blocking_start_time': 1776865684.1005332, 'checkpoint_manager_blocking_duration_secs': 29.679828643798828} I0422 13:48:33.780507 135837790684992 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0422 13:48:33.780551 135837790684992 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 13:48:38.961971 135708770223872 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_20260422_123906/linen_xpk_feat_nnx_post_train_fixes_20260422_123906_15_ocdbt_true/checkpoints/19/items/array_metadatas/process_6 I0422 13:49:16.114492 135713626507008 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.267 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.38589668273926 s) (per-host) I0422 13:49:16.114611 135713626507008 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.335960s. I0422 13:49:22.754777 135713626507008 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.976110s. I0422 13:49:22.755086 135714134013696 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 13:49:22.755160 135714134013696 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0422 13:49:22.755200 135714134013696 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0422 13:49:22.756551 135714134013696 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0422 13:49:22.756696 135837790684992 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0422 13:49:22.756852 135837790684992 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 13:49:22.757863 135837790684992 metric_logger.py:196] completed step: 19, seconds: 0.148, TFLOP/s/device: 92.030, Tokens/s/device: 13871.767, total_weights: 65536, loss: 5.391, lm_loss: 5.391, perplexity: 219.349 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 13:49:35 UTC 2026 EXIT_CODE=0