MaxView

Timing Test Reference Updates
‹ 16_zarr3_false_saveCase: 16_zarr3_true_resume16_zarr3_true_save ›

Metrics: main (b117f50cf) vs feat/nnx-post-train-fixes (28b1e4a16)

Metricmain  b117f50cffeat/nnx-post-train-fixes  28b1e4a16Diff (feat/nnx-post-train-fixes − main)
Parameters1.104 billion1.104 billion—
Final loss5.39105.39100
TFLOP/s92.21192.298+0.087
Tok/s13899.013912.2+13.218
Avg s/step8.4268.390-0.036
Memory %1.381.380
JAX0.9.20.9.2—

Diff = branch value − main value. Green = branch improved. Red = branch regressed.

main  ·  b117f50cf  ·  main_20260424_070227  ·  full log 
XPK Start: Fri Apr 24 08:21:13 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-24 08:21:38.745545: 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)
I0424 08:21:38.949790 135029628073792 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 08:21:47.991477 135029628073792 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 08:21:47.993819 135029628073792 distributed.py:172] Connecting to JAX distributed service on mt-16-zarr3-true-rs-fbs7h-slice-job-0-0.mt-16-zarr3-true-rs-fbs7h:8482
I0424 08:21:49.082346 135029628073792 max_utils.py:284] Jax distributed system initialized!
I0424 08:21:55.426391 135029628073792 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 08:21:55.426501 135029628073792 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 08:21:55.426542 135029628073792 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
I0424 08:21:55.426578 135029628073792 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 08:21:56.122242 135029628073792 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 08:21:56.122531 135029628073792 checkpointing.py:677] Setting up checkpoint logger...
I0424 08:21:56.122585 135029628073792 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 08:21:56.122628 135029628073792 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 08:21:56.122963 135029628073792 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 0x7ace6a918050>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 08:21:57.374549 135029628073792 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0424 08:21:57.374792 135029628073792 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 0x7ab9583c4170>}, handler_registry=None
I0424 08:21:57.375032 135029628073792 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ab9583c4170>` 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`.
I0424 08:21:57.375082 135029628073792 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ab9583c7c80>` 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`.
I0424 08:21:57.375132 135029628073792 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 0x7ab9583c4170>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ab9583c4170>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ab9583c7c80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ab9583c7c80>}).
I0424 08:21:57.375515 135029628073792 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 08:21:57.375588 135029628073792 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7ab9387bd800> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 08:21:58.946738 134908278208256 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777018703357808009, 'commit_timestamp_nsecs': 1777018749412887090, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/5/_CHECKPOINT_METADATA
I0424 08:21:59.191918 134908269815552 checkpoint_manager.py:1871] Missing metrics for step 9
I0424 08:21:59.428234 135029628073792 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints
I0424 08:21:59.617854 135029628073792 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_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ab9583c21e0>
I0424 08:21:59.618027 135029628073792 checkpointing.py:301] Checkpoint manager created!
I0424 08:22:00.570661 135029628073792 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 08:22:00.570778 135029628073792 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:22:00.957359 135029628073792 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 08:22:00.957450 135029628073792 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:22:00.979108 135029628073792 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 08:22:00.979209 135029628073792 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:22:01.003235 135029628073792 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 08:22:01.003303 135029628073792 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:22:01.019754 135029628073792 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 08:22:01.019814 135029628073792 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:22:01.036261 135029628073792 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 08:22:01.036323 135029628073792 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:22:01.061016 135029628073792 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 08:22:01.061086 135029628073792 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:22:01.082214 135029628073792 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 08:22:01.082279 135029628073792 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 08:22:01.292886 135029628073792 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 08:22:01.292996 135029628073792 checkpointing.py:581] restoring from this run's directory step 9
I0424 08:22:01.541228 135029628073792 event_tracking.py:70] [process=6] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/9.
I0424 08:22:01.793422 135029628073792 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/9.
I0424 08:22:01.793869 135029628073792 event_tracking.py:125] [process=6] [sync] Finished blocking load in 0.25 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/9.
I0424 08:22:03.215710    1451 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 08:22:09.297115 135029628073792 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 259.603 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.08726692199707 s) (per-host)
I0424 08:22:09.302857 135029628073792 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 243.820 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.490836143493652 s) (per-host)
I0424 08:22:10.942591 135029628073792 event_tracking.py:138] [process=6] [sync] Finished load in 9.40 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/9
I0424 08:22:10.942744 135029628073792 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints', 'checkpointer_start_time': 1777018921.5411298, 'checkpointer_duration_secs': 9.401583194732666, 'checkpoint_manager_start_time': 1777018921.2932594, 'checkpoint_manager_duration_secs': 9.649455308914185}
fsdp: 32
I0424 08:22:11.361551 135029628073792 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 08:22:11.361707 135029628073792 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)
I0424 08:22:11.361785 135029628073792 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)
I0424 08:22:11.361876 135029628073792 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')
I0424 08:22:11.361962 135029628073792 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)
I0424 08:22:11.362025 135029628073792 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)
I0424 08:22:11.362117 135029628073792 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)
I0424 08:22:11.362202 135029628073792 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')
I0424 08:22:11.362267 135029628073792 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)
I0424 08:22:11.362326 135029628073792 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)
I0424 08:22:11.362398 135029628073792 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0424 08:22:11.362487 135029628073792 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0424 08:22:11.856984 135029628073792 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 08:22:11.857083 135029628073792 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 08:22:11.872624 135029628073792 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 08:22:11.872684 135029628073792 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 08:22:23.085952 135029628073792 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.
I0424 08:22:23.086787 135029628073792 metric_logger.py:301] number parameters: 1.104 billion
I0424 08:22:34.438058 135029628073792 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0424 08:22:34.595395 135029628073792 checkpointing.py:776] Waited 0.15731549263000488 seconds for step 10 to finish before starting checkpointing.
I0424 08:22:34.597619 135029628073792 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 08:22:34.599753 135029628073792 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0424 08:22:34.601738 135029628073792 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/10.
I0424 08:22:34.933153 135029628073792 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 08:22:34.934047 135029628073792 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:22:34.934118 135029628073792 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
I0424 08:22:35.211136 135029628073792 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278075s
I0424 08:22:35.211306 135029628073792 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.472 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28191161155700684 s) (per-host)
I0424 08:22:35.211357 135029628073792 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.281971s (batch_requests_ready=0.002153s, total_serialization_initiated=0.279750s, others=0.000069s)
I0424 08:22:35.211445 135029628073792 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.286368s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.286352)
I0424 08:22:35.212234 135029628073792 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/10.
I0424 08:22:35.212589 134908227852032 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:42:35.212547
I0424 08:22:35.217180 135029628073792 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:22:35.217474 134904436217600 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:22:35.217628 135029628073792 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777018954.5976, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777018954.5997941, 'checkpointer_blocking_duration_secs': 0.6129465103149414, 'get_old_steps_start_time': 1777018955.2127624, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1777018954.5958607, 'checkpoint_manager_blocking_duration_secs': 0.6217317581176758}
I0424 08:22:35.217730 135029628073792 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0424 08:22:35.217779 135029628073792 max_utils.py:750] 
Memstats: After params initialized:
I0424 08:22:35.217829 135029628073792 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0424 08:22:35.217874 135029628073792 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0424 08:22:35.217903 135029628073792 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0424 08:22:35.217927 135029628073792 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0424 08:22:35.526713 135029628073792 metric_logger.py:196] completed step: 10, seconds: 11.351, TFLOP/s/device: 1.197, Tokens/s/device: 180.422, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0424 08:22:35.723246 135029628073792 metric_logger.py:196] completed step: 11, seconds: 1.087, TFLOP/s/device: 12.499, Tokens/s/device: 1883.963, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0424 08:22:35.870828 135029628073792 metric_logger.py:196] completed step: 12, seconds: 0.049, TFLOP/s/device: 275.021, Tokens/s/device: 41454.133, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0424 08:22:36.017968 135029628073792 metric_logger.py:196] completed step: 13, seconds: 0.153, TFLOP/s/device: 88.604, Tokens/s/device: 13355.418, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0424 08:22:36.022533 135029628073792 checkpointing.py:772] Waiting for step 15 to finish before checkpoint...
I0424 08:22:36.312064 135029628073792 checkpointing.py:776] Waited 0.28949427604675293 seconds for step 15 to finish before starting checkpointing.
I0424 08:22:36.315264 135029628073792 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:22:40.182287 134904444610304 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/10/items/array_metadatas/process_6
I0424 08:23:16.245570 134908227852032 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.232 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.31613850593567 s) (per-host)
I0424 08:23:16.245696 134908227852032 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.032994s.
I0424 08:23:25.870934 134908227852032 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.658216s.
I0424 08:23:25.871245 134904436217600 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:23:25.871415 134904436217600 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:23:25.871654 134904436217600 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:23:25.872983 134904436217600 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0424 08:23:25.873187 135029628073792 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.
W0424 08:23:25.873325 135029628073792 checkpoint_manager.py:1452] Waiting for previous save to complete took 49.558070 seconds. If this number is high, consider checkpointing less frequently.
I0424 08:23:25.875414 135029628073792 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15
I0424 08:23:25.877899 135029628073792 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/15.
I0424 08:23:26.172949 135029628073792 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:23:26.173040 135029628073792 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
I0424 08:23:26.199207 135029628073792 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027434s
I0424 08:23:26.199384 135029628073792 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 50.017 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03084111213684082 s) (per-host)
I0424 08:23:26.199438 135029628073792 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.030905s (batch_requests_ready=0.001764s, total_serialization_initiated=0.029069s, others=0.000072s)
I0424 08:23:26.199533 135029628073792 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035907s (all_items=0.000012s, per_item={'items': '0.00001192'}, temp_paths=0.035895)
I0424 08:23:26.200249 135029628073792 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/15.
I0424 08:23:26.200580 134901223384832 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:43:26.200543
I0424 08:23:26.212237 135029628073792 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:23:26.212508 134904436217600 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:23:26.212661 135029628073792 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777018956.3152258, 'wait_for_prev_duration_secs': 49.558069944381714, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777019005.8754537, 'checkpointer_blocking_duration_secs': 0.3252739906311035, 'get_old_steps_start_time': 1777019006.200752, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1777018956.3127203, 'checkpoint_manager_blocking_duration_secs': 49.899906158447266}
I0424 08:23:26.212769 135029628073792 checkpointing.py:408] Started an asynchronous checkpoint save for step 15
I0424 08:23:26.213581 135029628073792 metric_logger.py:196] completed step: 14, seconds: 0.147, TFLOP/s/device: 92.187, Tokens/s/device: 13895.391, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0424 08:23:26.219878 135029628073792 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.325, Tokens/s/device: 13916.163, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0424 08:23:31.310150 134904444610304 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/15/items/array_metadatas/process_6
I0424 08:23:47.195981 135029628073792 metric_logger.py:196] completed step: 16, seconds: 50.196, TFLOP/s/device: 0.271, Tokens/s/device: 40.800, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0424 08:23:47.343359 135029628073792 metric_logger.py:196] completed step: 17, seconds: 20.830, TFLOP/s/device: 0.652, Tokens/s/device: 98.319, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0424 08:23:47.490513 135029628073792 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.264, Tokens/s/device: 13454.830, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0424 08:23:47.637207 135029628073792 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0424 08:23:47.637917 135029628073792 checkpointing.py:776] Waited 0.000728607177734375 seconds for step 19 to finish before starting checkpointing.
I0424 08:23:47.640114 135029628073792 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:24:08.436939 134901223384832 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.371 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.26836085319519 s) (per-host)
I0424 08:24:08.437054 134901223384832 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.236358s.
I0424 08:24:16.412476 134901223384832 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.211763s.
I0424 08:24:16.412763 134904436217600 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:24:16.412885 134904436217600 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:24:16.412934 134904436217600 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:24:16.414429 134904436217600 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts.
I0424 08:24:16.414600 135029628073792 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.
W0424 08:24:16.414723 135029628073792 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.774626 seconds. If this number is high, consider checkpointing less frequently.
I0424 08:24:16.416390 135029628073792 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0424 08:24:16.418527 135029628073792 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/19.
I0424 08:24:16.703323 135029628073792 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:24:16.703412 135029628073792 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
I0424 08:24:16.730722 135029628073792 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.028275s
I0424 08:24:16.730848 135029628073792 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 48.710 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03166842460632324 s) (per-host)
I0424 08:24:16.730893 135029628073792 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031723s (batch_requests_ready=0.001626s, total_serialization_initiated=0.030039s, others=0.000059s)
I0424 08:24:16.730966 135029628073792 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035964s (all_items=0.000027s, per_item={'items': '0.00002694'}, temp_paths=0.035937)
I0424 08:24:16.731640 135029628073792 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/19.
I0424 08:24:16.731980 134904436217600 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:44:16.731942
I0424 08:24:16.733899 135029628073792 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:24:16.734201 134901198206720 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:24:16.734333 135029628073792 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777019027.640072, 'wait_for_prev_duration_secs': 28.7746262550354, 'time_between_consecutive_saves_sec': 21.76705026626587, 'checkpointer_blocking_start_time': 1777019056.4164298, 'checkpointer_blocking_duration_secs': 0.3157031536102295, 'get_old_steps_start_time': 1777019056.7321503, 'get_old_steps_duration_secs': 2.6464462280273438e-05, 'checkpoint_manager_blocking_start_time': 1777019027.6381714, 'checkpoint_manager_blocking_duration_secs': 29.096129894256592}
I0424 08:24:16.734501 135029628073792 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0424 08:24:16.734549 135029628073792 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:24:21.871226 134904444610304 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_16_zarr3_true/checkpoints/19/items/array_metadatas/process_6
I0424 08:24:59.078039 134904436217600 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.273 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.378817319869995 s) (per-host)
I0424 08:24:59.078179 134904436217600 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.346087s.
I0424 08:25:05.424827 134904436217600 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.692719s.
I0424 08:25:05.425044 134901198206720 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:25:05.425116 134901198206720 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:25:05.425159 134901198206720 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:25:05.426923 134901198206720 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0424 08:25:05.427118 135029628073792 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.
I0424 08:25:05.427271 135029628073792 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 08:25:05.428309 135029628073792 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.211, Tokens/s/device: 13898.975, 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: Fri Apr 24 08:25:18 UTC 2026
EXIT_CODE=0
feat/nnx-post-train-fixes  ·  28b1e4a16  ·  feat_nnx_post_train_fixes_20260424_120657  ·  full log 
XPK Start: Fri Apr 24 13:18:39 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. 
2026-04-24 13:19:04.722529: 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)
I0424 13:19:04.940335 133400064923456 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 13:19:13.981472 133400064923456 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 13:19:13.983761 133400064923456 distributed.py:172] Connecting to JAX distributed service on mt-16-zarr3-true-rs-gd1ip-slice-job-0-0.mt-16-zarr3-true-rs-gd1ip:8482
I0424 13:19:14.951608 133400064923456 max_utils.py:284] Jax distributed system initialized!
I0424 13:19:20.058963 133400064923456 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 13:19:20.059065 133400064923456 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 13:19:20.059123 133400064923456 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
I0424 13:19:20.059164 133400064923456 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 13:19:20.757930 133400064923456 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 13:19:20.758553 133400064923456 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 13:19:20.758781 133400064923456 checkpointing.py:688] Setting up checkpoint logger...
I0424 13:19:20.758834 133400064923456 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 13:19:20.758877 133400064923456 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 13:19:20.759239 133400064923456 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 0x7952e7c4c050>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 13:19:22.059557 133400064923456 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 13:19:22.059794 133400064923456 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 0x7951d81e2000>}, handler_registry=None
I0424 13:19:22.060037 133400064923456 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7951d81e2000>` 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`.
I0424 13:19:22.060091 133400064923456 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x793f40775dc0>` 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`.
I0424 13:19:22.060137 133400064923456 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 0x7951d81e2000>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7951d81e2000>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x793f40775dc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x793f40775dc0>}).
I0424 13:19:22.060524 133400064923456 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 13:19:22.060595 133400064923456 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x793e582fdd00> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 13:19:23.841171 133280879863552 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777036597503488017, 'commit_timestamp_nsecs': 1777036643843243990, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/9/_CHECKPOINT_METADATA
I0424 13:19:24.078951 133280879863552 checkpoint_manager.py:1871] Missing metrics for step 9
I0424 13:19:24.451346 133400064923456 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints
I0424 13:19:24.453967 133400064923456 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x793f8018acc0>
I0424 13:19:24.454084 133400064923456 checkpointing.py:302] Checkpoint manager created!
I0424 13:19:25.402298 133400064923456 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 13:19:25.402407 133400064923456 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:19:25.785815 133400064923456 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 13:19:25.785912 133400064923456 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:19:25.803249 133400064923456 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 13:19:25.803316 133400064923456 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:19:25.828059 133400064923456 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:19:25.828139 133400064923456 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:19:25.845447 133400064923456 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:19:25.845509 133400064923456 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:19:25.862713 133400064923456 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:19:25.862774 133400064923456 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:19:25.888413 133400064923456 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 13:19:25.888490 133400064923456 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:19:25.910171 133400064923456 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 13:19:25.910234 133400064923456 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 13:19:26.129197 133400064923456 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 13:19:26.129312 133400064923456 checkpointing.py:582] restoring from this run's directory step 9
I0424 13:19:26.390811 133400064923456 event_tracking.py:70] [process=6] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/9.
I0424 13:19:26.635648 133400064923456 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/9.
I0424 13:19:26.636079 133400064923456 event_tracking.py:125] [process=6] [sync] Finished blocking load in 0.25 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/9.
I0424 13:19:28.070301    1414 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 13:19:33.708088 133400064923456 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 280.012 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 5.643588542938232 s) (per-host)
I0424 13:19:33.722762 133400064923456 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 261.057 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.062260866165161 s) (per-host)
I0424 13:19:35.923743 133400064923456 event_tracking.py:138] [process=6] [sync] Finished load in 9.53 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/9
I0424 13:19:35.923901 133400064923456 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints', 'checkpointer_start_time': 1777036766.3907125, 'checkpointer_duration_secs': 9.533156394958496, 'checkpoint_manager_start_time': 1777036766.1295626, 'checkpoint_manager_duration_secs': 9.794307947158813}
fsdp: 32
I0424 13:19:36.346368 133400064923456 maxtext_utils.py:1880]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 13:19:36.346500 133400064923456 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)
I0424 13:19:36.346560 133400064923456 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)
I0424 13:19:36.346617 133400064923456 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')
I0424 13:19:36.346666 133400064923456 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)
I0424 13:19:36.346704 133400064923456 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)
I0424 13:19:36.346756 133400064923456 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)
I0424 13:19:36.346807 133400064923456 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')
I0424 13:19:36.346847 133400064923456 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)
I0424 13:19:36.346883 133400064923456 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)
I0424 13:19:36.346930 133400064923456 maxtext_utils.py:1880]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0424 13:19:36.346975 133400064923456 maxtext_utils.py:1880]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0424 13:19:36.843608 133400064923456 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 13:19:36.843703 133400064923456 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 13:19:36.859228 133400064923456 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 13:19:36.859292 133400064923456 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 13:19:47.861575 133400064923456 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.
I0424 13:19:47.862386 133400064923456 metric_logger.py:301] number parameters: 1.104 billion
I0424 13:19:59.461582 133400064923456 checkpointing.py:794] Waiting for step 10 to finish before checkpoint...
I0424 13:19:59.618110 133400064923456 checkpointing.py:798] Waited 0.1564950942993164 seconds for step 10 to finish before starting checkpointing.
I0424 13:19:59.620444 133400064923456 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 13:19:59.622465 133400064923456 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0424 13:19:59.624126 133400064923456 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/10.
I0424 13:19:59.960871 133400064923456 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 13:19:59.961798 133400064923456 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:19:59.961864 133400064923456 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
I0424 13:20:00.239437 133400064923456 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278640s
I0424 13:20:00.239612 133400064923456 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.460 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28253698348999023 s) (per-host)
I0424 13:20:00.239666 133400064923456 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.282602s (batch_requests_ready=0.002201s, total_serialization_initiated=0.280327s, others=0.000074s)
I0424 13:20:00.239760 133400064923456 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287003s (all_items=0.000021s, per_item={'items': '0.00002074'}, temp_paths=0.286982)
I0424 13:20:00.240471 133400064923456 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.62 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/10.
I0424 13:20:00.240777 133280846292736 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:40:00.240745
I0424 13:20:00.248743 133400064923456 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:20:00.248978 133277633459968 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:20:00.249120 133400064923456 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777036799.6204257, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777036799.6225026, 'checkpointer_blocking_duration_secs': 0.6184256076812744, 'get_old_steps_start_time': 1777036800.240951, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1777036799.618597, 'checkpoint_manager_blocking_duration_secs': 0.6304640769958496}
I0424 13:20:00.249223 133400064923456 checkpointing.py:409] Started an asynchronous checkpoint save for step 10
I0424 13:20:00.249272 133400064923456 max_utils.py:750] 
Memstats: After params initialized:
I0424 13:20:00.249320 133400064923456 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0424 13:20:00.249352 133400064923456 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0424 13:20:00.249378 133400064923456 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0424 13:20:00.249403 133400064923456 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0424 13:20:00.565477 133400064923456 metric_logger.py:196] completed step: 10, seconds: 11.599, TFLOP/s/device: 1.171, Tokens/s/device: 176.566, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0424 13:20:01.148823 133400064923456 metric_logger.py:196] completed step: 11, seconds: 1.102, TFLOP/s/device: 12.325, Tokens/s/device: 1857.814, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0424 13:20:01.296490 133400064923456 metric_logger.py:196] completed step: 12, seconds: 0.437, TFLOP/s/device: 31.075, Tokens/s/device: 4683.959, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0424 13:20:01.443985 133400064923456 metric_logger.py:196] completed step: 13, seconds: 0.152, TFLOP/s/device: 89.184, Tokens/s/device: 13442.819, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0424 13:20:01.448719 133400064923456 checkpointing.py:794] Waiting for step 15 to finish before checkpoint...
I0424 13:20:01.738292 133400064923456 checkpointing.py:798] Waited 0.2895395755767822 seconds for step 15 to finish before starting checkpointing.
I0424 13:20:01.741844 133400064923456 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:20:05.270915 133280837900032 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/10/items/array_metadatas/process_6
I0424 13:20:41.508108 133280846292736 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.016 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.55098104476929 s) (per-host)
I0424 13:20:41.508234 133280846292736 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.267346s.
I0424 13:20:50.121966 133280846292736 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.881055s.
I0424 13:20:50.122205 133277633459968 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:20:50.122262 133277633459968 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:20:50.122315 133277633459968 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:20:50.125007 133277633459968 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0424 13:20:50.125195 133400064923456 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.
W0424 13:20:50.125330 133400064923456 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.383484 seconds. If this number is high, consider checkpointing less frequently.
I0424 13:20:50.127442 133400064923456 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15
I0424 13:20:50.129923 133400064923456 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/15.
I0424 13:20:50.443579 133400064923456 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:20:50.443677 133400064923456 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
I0424 13:20:50.468479 133400064923456 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.025995s
I0424 13:20:50.468612 133400064923456 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 52.569 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.029343843460083008 s) (per-host)
I0424 13:20:50.468660 133400064923456 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.029400s (batch_requests_ready=0.001735s, total_serialization_initiated=0.027603s, others=0.000062s)
I0424 13:20:50.468740 133400064923456 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.034505s (all_items=0.000014s, per_item={'items': '0.00001383'}, temp_paths=0.034492)
I0424 13:20:50.469460 133400064923456 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/15.
I0424 13:20:50.469802 133280837900032 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:40:50.469765
I0424 13:20:50.483211 133400064923456 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:20:50.483533 133277633459968 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:20:50.483715 133400064923456 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777036801.7418144, 'wait_for_prev_duration_secs': 48.38348412513733, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777036850.1274843, 'checkpointer_blocking_duration_secs': 0.3424553871154785, 'get_old_steps_start_time': 1777036850.469964, 'get_old_steps_duration_secs': 3.361701965332031e-05, 'checkpoint_manager_blocking_start_time': 1777036801.7389328, 'checkpoint_manager_blocking_duration_secs': 48.74474000930786}
I0424 13:20:50.483835 133400064923456 checkpointing.py:409] Started an asynchronous checkpoint save for step 15
I0424 13:20:50.484755 133400064923456 metric_logger.py:196] completed step: 14, seconds: 0.147, TFLOP/s/device: 92.123, Tokens/s/device: 13885.781, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0424 13:20:50.491174 133400064923456 metric_logger.py:196] completed step: 15, seconds: 0.148, TFLOP/s/device: 92.064, Tokens/s/device: 13876.937, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0424 13:20:56.080777 133275998361344 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/15/items/array_metadatas/process_6
I0424 13:21:11.609204 133400064923456 metric_logger.py:196] completed step: 16, seconds: 49.041, TFLOP/s/device: 0.277, Tokens/s/device: 41.761, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0424 13:21:11.756568 133400064923456 metric_logger.py:196] completed step: 17, seconds: 20.972, TFLOP/s/device: 0.648, Tokens/s/device: 97.654, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0424 13:21:11.903893 133400064923456 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.195, Tokens/s/device: 13444.496, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0424 13:21:12.050660 133400064923456 checkpointing.py:794] Waiting for step 19 to finish before checkpoint...
I0424 13:21:12.051385 133400064923456 checkpointing.py:798] Waited 0.0007479190826416016 seconds for step 19 to finish before starting checkpointing.
I0424 13:21:12.053416 133400064923456 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:21:31.779181 133280837900032 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 38.210 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.339876890182495 s) (per-host)
I0424 13:21:31.779297 133280837900032 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.309385s.
I0424 13:21:40.848733 133280837900032 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.378805s.
I0424 13:21:40.849021 133277633459968 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:21:40.849179 133277633459968 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:21:40.849228 133277633459968 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:21:40.851245 133277633459968 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts.
I0424 13:21:40.851426 133400064923456 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.
W0424 13:21:40.851558 133400064923456 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.798140 seconds. If this number is high, consider checkpointing less frequently.
I0424 13:21:40.853152 133400064923456 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0424 13:21:40.855326 133400064923456 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/19.
I0424 13:21:41.148309 133400064923456 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:21:41.148400 133400064923456 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
I0424 13:21:41.174792 133400064923456 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027398s
I0424 13:21:41.174919 133400064923456 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 50.134 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03076910972595215 s) (per-host)
I0424 13:21:41.174963 133400064923456 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.030822s (batch_requests_ready=0.001626s, total_serialization_initiated=0.029139s, others=0.000058s)
I0424 13:21:41.175031 133400064923456 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035049s (all_items=0.000012s, per_item={'items': '0.00001216'}, temp_paths=0.035037)
I0424 13:21:41.175622 133400064923456 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/19.
I0424 13:21:41.175840 133275998361344 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:41:41.175799
I0424 13:21:41.179885 133400064923456 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:21:41.180129 133277633459968 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:21:41.180282 133400064923456 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777036872.0533864, 'wait_for_prev_duration_secs': 28.798140287399292, 'time_between_consecutive_saves_sec': 21.928340673446655, 'checkpointer_blocking_start_time': 1777036900.853194, 'checkpointer_blocking_duration_secs': 0.32279109954833984, 'get_old_steps_start_time': 1777036901.1759996, 'get_old_steps_duration_secs': 2.2172927856445312e-05, 'checkpoint_manager_blocking_start_time': 1777036872.051636, 'checkpoint_manager_blocking_duration_secs': 29.12861728668213}
I0424 13:21:41.180373 133400064923456 checkpointing.py:409] Started an asynchronous checkpoint save for step 19
I0424 13:21:41.180413 133400064923456 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:21:46.293219 133274422134528 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_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_16_zarr3_true/checkpoints/19/items/array_metadatas/process_6
I0424 13:22:22.748226 133275998361344 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.967 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.60403847694397 s) (per-host)
I0424 13:22:22.748365 133275998361344 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.572414s.
I0424 13:22:30.201618 133275998361344 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.025651s.
I0424 13:22:30.201911 133277633459968 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:22:30.202034 133277633459968 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:22:30.202076 133277633459968 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:22:30.203937 133277633459968 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0424 13:22:30.204162 133400064923456 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.
I0424 13:22:30.204307 133400064923456 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 13:22:30.205287 133400064923456 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.298, Tokens/s/device: 13912.193, 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: Fri Apr 24 13:22:39 UTC 2026
EXIT_CODE=0