MaxView

Timing Test Reference Updates
‹ 15_ocdbt_false_saveCase: 15_ocdbt_true_resume15_ocdbt_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/s92.31592.356+0.041
Tok/s13914.613920.9+6.243
Avg s/step8.5038.346-0.157
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 08:04:21 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 08:04:45.781670: 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 08:04:45.987311 132995791423296 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 08:04:55.028265 132995791423296 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 08:04:55.030549 132995791423296 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-true-rs-dfhxx-slice-job-0-0.mt-15-ocdbt-true-rs-dfhxx:8482
I0424 08:04:56.172222 132995791423296 max_utils.py:284] Jax distributed system initialized!
I0424 08:05:02.353711 132995791423296 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 08:05:02.353821 132995791423296 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 08:05:02.353861 132995791423296 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 08:05:02.353896 132995791423296 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 08:05:03.078393 132995791423296 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 08:05:03.078673 132995791423296 checkpointing.py:677] Setting up checkpoint logger...
I0424 08:05:03.078726 132995791423296 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 08:05:03.078773 132995791423296 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 08:05:03.079129 132995791423296 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 0x78f4de116ff0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 08:05:04.329909 132995791423296 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0424 08:05:04.330159 132995791423296 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 0x78e0601a0350>}, handler_registry=None
I0424 08:05:04.330400 132995791423296 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78e0601a0350>` 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 08:05:04.330450 132995791423296 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78e0601a3fb0>` 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 08:05:04.330489 132995791423296 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 0x78e0601a0350>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78e0601a0350>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78e0601a3fb0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78e0601a3fb0>}).
I0424 08:05:04.330832 132995791423296 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 08:05:04.330901 132995791423296 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x78e048609a80> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 08:05:05.853984 132873050904320 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777017741383856738, 'commit_timestamp_nsecs': 1777017788340820620, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/9/_CHECKPOINT_METADATA
I0424 08:05:06.105169 132873050904320 checkpoint_manager.py:1871] Missing metrics for step 9
I0424 08:05:06.296608 132995791423296 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints
I0424 08:05:06.435004 132995791423296 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_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x78e0601a3ef0>
I0424 08:05:06.435212 132995791423296 checkpointing.py:301] Checkpoint manager created!
I0424 08:05:07.372801 132995791423296 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 08:05:07.372924 132995791423296 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:05:07.758501 132995791423296 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 08:05:07.758598 132995791423296 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:05:07.782028 132995791423296 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 08:05:07.782139 132995791423296 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 08:05:07.806379 132995791423296 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 08:05:07.806450 132995791423296 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:05:07.823125 132995791423296 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 08:05:07.823189 132995791423296 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:05:07.839829 132995791423296 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 08:05:07.839889 132995791423296 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:05:07.864864 132995791423296 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 08:05:07.864938 132995791423296 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 08:05:07.886324 132995791423296 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 08:05:07.886391 132995791423296 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 08:05:08.097963 132995791423296 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 08:05:08.098074 132995791423296 checkpointing.py:581] restoring from this run's directory step 9
I0424 08:05:08.775144 132995791423296 event_tracking.py:70] [process=6] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/9.
I0424 08:05:09.030171 132995791423296 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/9.
I0424 08:05:09.030622 132995791423296 event_tracking.py:125] [process=6] [sync] Finished blocking load in 0.26 seconds. Continuing load @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/9.
I0424 08:05:10.848150    1472 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 08:05:16.712714 132995791423296 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 269.174 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 5.870823383331299 s) (per-host)
I0424 08:05:16.717359 132995791423296 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 252.640 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.26422381401062 s) (per-host)
I0424 08:05:16.928149 132995791423296 event_tracking.py:138] [process=6] [sync] Finished load in 8.15 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/9
I0424 08:05:16.928309 132995791423296 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints', 'checkpointer_start_time': 1777017908.7750282, 'checkpointer_duration_secs': 8.15324878692627, 'checkpoint_manager_start_time': 1777017908.0983343, 'checkpoint_manager_duration_secs': 8.829945087432861}
fsdp: 32
I0424 08:05:17.348405 132995791423296 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 08:05:17.348537 132995791423296 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 08:05:17.348590 132995791423296 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 08:05:17.348650 132995791423296 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 08:05:17.348703 132995791423296 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 08:05:17.348741 132995791423296 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 08:05:17.348793 132995791423296 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 08:05:17.348847 132995791423296 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 08:05:17.348888 132995791423296 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 08:05:17.348924 132995791423296 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 08:05:17.348971 132995791423296 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0424 08:05:17.349019 132995791423296 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0424 08:05:17.836534 132995791423296 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 08:05:17.836625 132995791423296 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 08:05:17.852375 132995791423296 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 08:05:17.852433 132995791423296 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 08:05:28.989715 132995791423296 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 08:05:28.990511 132995791423296 metric_logger.py:301] number parameters: 1.104 billion
I0424 08:05:40.769477 132995791423296 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0424 08:05:40.923074 132995791423296 checkpointing.py:776] Waited 0.15357375144958496 seconds for step 10 to finish before starting checkpointing.
I0424 08:05:40.925379 132995791423296 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 08:05:40.927207 132995791423296 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0424 08:05:40.928696 132995791423296 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/10.
I0424 08:05:41.514603 132995791423296 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 08:05:41.515604 132995791423296 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:05:41.515661 132995791423296 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 08:05:41.798963 132995791423296 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.284443s
I0424 08:05:41.799143 132995791423296 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.350 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2883458137512207 s) (per-host)
I0424 08:05:41.799197 132995791423296 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.288411s (batch_requests_ready=0.002166s, total_serialization_initiated=0.286172s, others=0.000074s)
I0424 08:05:41.799293 132995791423296 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.292597s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.292580)
I0424 08:05:41.800015 132995791423296 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.87 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/10.
I0424 08:05:41.800330 132873017333504 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:25:41.800296
I0424 08:05:41.802287 132995791423296 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:05:41.802543 132869242484480 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:05:41.802645 132995791423296 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777017940.9253564, 'wait_for_prev_duration_secs': 6.341934204101562e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777017940.9272463, 'checkpointer_blocking_duration_secs': 0.8732378482818604, 'get_old_steps_start_time': 1777017941.8005073, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1777017940.9235432, 'checkpoint_manager_blocking_duration_secs': 0.8790662288665771}
I0424 08:05:41.802793 132995791423296 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0424 08:05:41.802844 132995791423296 max_utils.py:750] 
Memstats: After params initialized:
I0424 08:05:41.802894 132995791423296 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0424 08:05:41.802925 132995791423296 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0424 08:05:41.802952 132995791423296 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0424 08:05:41.802975 132995791423296 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0424 08:05:42.121570 132995791423296 metric_logger.py:196] completed step: 10, seconds: 11.779, TFLOP/s/device: 1.154, Tokens/s/device: 173.871, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0424 08:05:42.303108 132995791423296 metric_logger.py:196] completed step: 11, seconds: 1.351, TFLOP/s/device: 10.061, Tokens/s/device: 1516.452, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0424 08:05:42.450906 132995791423296 metric_logger.py:196] completed step: 12, seconds: 0.034, TFLOP/s/device: 394.504, Tokens/s/device: 59464.011, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0424 08:05:42.598750 132995791423296 metric_logger.py:196] completed step: 13, seconds: 0.153, TFLOP/s/device: 88.610, Tokens/s/device: 13356.289, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0424 08:05:42.603438 132995791423296 checkpointing.py:772] Waiting for step 15 to finish before checkpoint...
I0424 08:05:42.893000 132995791423296 checkpointing.py:776] Waited 0.28952836990356445 seconds for step 15 to finish before starting checkpointing.
I0424 08:05:42.895966 132995791423296 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:05:47.045474 132869730465536 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_15_ocdbt_true/checkpoints/10/items/array_metadatas/process_6
I0424 08:06:23.446265 132873017333504 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.667 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.93542456626892 s) (per-host)
I0424 08:06:23.446403 132873017333504 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.645959s.
I0424 08:06:32.263351 132873017333504 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.462875s.
I0424 08:06:32.263671 132869242484480 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:06:32.263790 132869242484480 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:06:32.263837 132869242484480 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:06:32.266029 132869242484480 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0424 08:06:32.266181 132995791423296 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 08:06:32.266267 132995791423296 checkpoint_manager.py:1452] Waiting for previous save to complete took 49.370308 seconds. If this number is high, consider checkpointing less frequently.
I0424 08:06:32.268320 132995791423296 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15
I0424 08:06:32.270826 132995791423296 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/15.
I0424 08:06:32.563499 132995791423296 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:06:32.563594 132995791423296 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 08:06:32.589970 132995791423296 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027660s
I0424 08:06:32.590155 132995791423296 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 49.524 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03114795684814453 s) (per-host)
I0424 08:06:32.590208 132995791423296 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031212s (batch_requests_ready=0.001781s, total_serialization_initiated=0.029359s, others=0.000072s)
I0424 08:06:32.590310 132995791423296 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.036241s (all_items=0.000013s, per_item={'items': '0.00001311'}, temp_paths=0.036228)
I0424 08:06:32.591012 132995791423296 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_15_ocdbt_true/checkpoints/15.
I0424 08:06:32.591351 132869242484480 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:26:32.591313
I0424 08:06:32.604626 132995791423296 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:06:32.604876 132869730465536 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:06:32.605019 132995791423296 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777017942.8959346, 'wait_for_prev_duration_secs': 49.3703076839447, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777017992.268359, 'checkpointer_blocking_duration_secs': 0.32316017150878906, 'get_old_steps_start_time': 1777017992.5915434, 'get_old_steps_duration_secs': 3.075599670410156e-05, 'checkpoint_manager_blocking_start_time': 1777017942.8936818, 'checkpoint_manager_blocking_duration_secs': 49.71130347251892}
I0424 08:06:32.605137 132995791423296 checkpointing.py:408] Started an asynchronous checkpoint save for step 15
I0424 08:06:32.605987 132995791423296 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 91.884, Tokens/s/device: 13849.722, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0424 08:06:32.614163 132995791423296 metric_logger.py:196] completed step: 15, seconds: 0.148, TFLOP/s/device: 92.004, Tokens/s/device: 13867.822, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0424 08:06:38.318435 132868664346368 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_15_ocdbt_true/checkpoints/15/items/array_metadatas/process_6
I0424 08:06:53.873650 132995791423296 metric_logger.py:196] completed step: 16, seconds: 50.009, TFLOP/s/device: 0.272, Tokens/s/device: 40.953, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0424 08:06:54.021030 132995791423296 metric_logger.py:196] completed step: 17, seconds: 21.113, TFLOP/s/device: 0.644, Tokens/s/device: 97.000, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0424 08:06:54.168282 132995791423296 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.157, Tokens/s/device: 13438.761, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0424 08:06:54.314886 132995791423296 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0424 08:06:54.315618 132995791423296 checkpointing.py:776] Waited 0.0007538795471191406 seconds for step 19 to finish before starting checkpointing.
I0424 08:06:54.317665 132995791423296 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:07:14.210482 132869242484480 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.924 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.651445627212524 s) (per-host)
I0424 08:07:14.210592 132869242484480 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.619121s.
I0424 08:07:24.172067 132869242484480 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 51.580581s.
I0424 08:07:24.172337 132869730465536 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:07:24.172394 132869730465536 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:07:24.172439 132869730465536 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:07:24.174672 132869730465536 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts.
I0424 08:07:24.174844 132995791423296 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 08:07:24.174973 132995791423296 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.857313 seconds. If this number is high, consider checkpointing less frequently.
I0424 08:07:24.176469 132995791423296 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0424 08:07:24.178480 132995791423296 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/19.
I0424 08:07:24.681343 132995791423296 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 08:07:24.681436 132995791423296 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 08:07:24.707644 132995791423296 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027232s
I0424 08:07:24.707823 132995791423296 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 50.141 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03076481819152832 s) (per-host)
I0424 08:07:24.707876 132995791423296 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.030828s (batch_requests_ready=0.001670s, total_serialization_initiated=0.029089s, others=0.000069s)
I0424 08:07:24.707975 132995791423296 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035048s (all_items=0.000012s, per_item={'items': '0.00001216'}, temp_paths=0.035036)
I0424 08:07:24.708579 132995791423296 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.53 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints/19.
I0424 08:07:24.708842 132871431898880 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 08:27:24.708808
I0424 08:07:24.711776 132995791423296 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 08:07:24.712046 132869730465536 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 08:07:24.712214 132995791423296 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777018014.317635, 'wait_for_prev_duration_secs': 29.85731291770935, 'time_between_consecutive_saves_sec': 22.05156660079956, 'checkpointer_blocking_start_time': 1777018044.1765099, 'checkpointer_blocking_duration_secs': 0.5324673652648926, 'get_old_steps_start_time': 1777018044.7089915, 'get_old_steps_duration_secs': 2.3603439331054688e-05, 'checkpoint_manager_blocking_start_time': 1777018014.315853, 'checkpoint_manager_blocking_duration_secs': 30.39633059501648}
I0424 08:07:24.712327 132995791423296 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0424 08:07:24.712370 132995791423296 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 08:07:30.251598 132868664346368 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_15_ocdbt_true/checkpoints/19/items/array_metadatas/process_6
I0424 08:08:06.860844 132871431898880 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.446 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.183748722076416 s) (per-host)
I0424 08:08:06.860963 132871431898880 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.152015s.
I0424 08:08:14.451831 132871431898880 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.742868s.
I0424 08:08:14.452170 132869730465536 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 08:08:14.452298 132869730465536 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 08:08:14.452339 132869730465536 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0424 08:08:14.454108 132869730465536 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0424 08:08:14.454308 132995791423296 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 08:08:14.454459 132995791423296 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 08:08:14.455467 132995791423296 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.315, Tokens/s/device: 13914.650, 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 08:08:26 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 13:01:45 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 13:02:10.229855: 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 13:02:10.441095 133269405091648 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 13:02:19.482037 133269405091648 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 13:02:19.484422 133269405091648 distributed.py:172] Connecting to JAX distributed service on mt-15-ocdbt-true-rs-6gnb3-slice-job-0-0.mt-15-ocdbt-true-rs-6gnb3:8482
I0424 13:02:20.720692 133269405091648 max_utils.py:284] Jax distributed system initialized!
I0424 13:02:26.787085 133269405091648 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 13:02:26.787192 133269405091648 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 13:02:26.787233 133269405091648 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 13:02:26.787265 133269405091648 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 13:02:27.479345 133269405091648 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 13:02:27.479948 133269405091648 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 13:02:27.480128 133269405091648 checkpointing.py:688] Setting up checkpoint logger...
I0424 13:02:27.480178 133269405091648 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 13:02:27.480222 133269405091648 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 13:02:27.480559 133269405091648 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 0x79347c1340b0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 13:02:28.768563 133269405091648 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 13:02:28.768797 133269405091648 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 0x79213c0eb1a0>}, handler_registry=None
I0424 13:02:28.769051 133269405091648 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79213c0eb1a0>` 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 13:02:28.769103 133269405091648 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7920e431efc0>` 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 13:02:28.769139 133269405091648 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 0x79213c0eb1a0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79213c0eb1a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7920e431efc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7920e431efc0>}).
I0424 13:02:28.769458 133269405091648 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 13:02:28.769530 133269405091648 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x79201c675e40> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 13:02:30.598481 133151024211712 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777035585392099934, 'commit_timestamp_nsecs': 1777035630254270950, '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_15_ocdbt_true/checkpoints/9/_CHECKPOINT_METADATA
I0424 13:02:30.859919 133151024211712 checkpoint_manager.py:1871] Missing metrics for step 9
I0424 13:02:31.006788 133269405091648 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_15_ocdbt_true/checkpoints
I0424 13:02:31.286297 133269405091648 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_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7920e431f1d0>
I0424 13:02:31.286483 133269405091648 checkpointing.py:302] Checkpoint manager created!
I0424 13:02:32.227172 133269405091648 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 13:02:32.227279 133269405091648 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:02:32.609154 133269405091648 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 13:02:32.609245 133269405091648 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:02:32.631062 133269405091648 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 13:02:32.631160 133269405091648 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:02:32.655146 133269405091648 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 13:02:32.655218 133269405091648 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:02:32.671759 133269405091648 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 13:02:32.671819 133269405091648 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:02:32.688445 133269405091648 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 13:02:32.688505 133269405091648 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:02:32.713272 133269405091648 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 13:02:32.713349 133269405091648 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:02:32.735068 133269405091648 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 13:02:32.735133 133269405091648 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 13:02:32.953541 133269405091648 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 13:02:32.953649 133269405091648 checkpointing.py:582] restoring from this run's directory step 9
I0424 13:02:33.204645 133269405091648 event_tracking.py:70] [process=3] [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_15_ocdbt_true/checkpoints/9.
I0424 13:02:33.473529 133269405091648 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_15_ocdbt_true/checkpoints/9.
I0424 13:02:33.473980 133269405091648 event_tracking.py:125] [process=3] [sync] Finished blocking load in 0.27 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_15_ocdbt_true/checkpoints/9.
I0424 13:02:34.878644    1420 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 13:02:40.628184 133269405091648 jax_array_handlers.py:843] [process=3] /jax/orbax/read/worker/io/requested throughput: 274.587 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 5.755095481872559 s) (per-host)
I0424 13:02:40.632703 133269405091648 base_pytree_checkpoint_handler.py:130] [process=3] /jax/checkpoint/read/gbytes_per_sec: 257.066 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 6.156361103057861 s) (per-host)
I0424 13:02:41.935333 133269405091648 event_tracking.py:138] [process=3] [sync] Finished load in 8.73 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_15_ocdbt_true/checkpoints/9
I0424 13:02:41.935503 133269405091648 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_15_ocdbt_true/checkpoints', 'checkpointer_start_time': 1777035753.2045472, 'checkpointer_duration_secs': 8.730924606323242, 'checkpoint_manager_start_time': 1777035752.9539073, 'checkpoint_manager_duration_secs': 8.981566429138184}
fsdp: 32
I0424 13:02:42.351093 133269405091648 maxtext_utils.py:1880]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 13:02:42.351226 133269405091648 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 13:02:42.351281 133269405091648 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 13:02:42.351339 133269405091648 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 13:02:42.351396 133269405091648 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 13:02:42.351436 133269405091648 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 13:02:42.351489 133269405091648 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 13:02:42.351542 133269405091648 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 13:02:42.351581 133269405091648 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 13:02:42.351617 133269405091648 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 13:02:42.351669 133269405091648 maxtext_utils.py:1880]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0424 13:02:42.351717 133269405091648 maxtext_utils.py:1880]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0424 13:02:42.842214 133269405091648 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 13:02:42.842314 133269405091648 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 13:02:42.858128 133269405091648 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 13:02:42.858194 133269405091648 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 13:02:53.844728 133269405091648 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 13:02:53.845518 133269405091648 metric_logger.py:301] number parameters: 1.104 billion
I0424 13:03:05.251195 133269405091648 checkpointing.py:794] Waiting for step 10 to finish before checkpoint...
I0424 13:03:05.404832 133269405091648 checkpointing.py:798] Waited 0.15361547470092773 seconds for step 10 to finish before starting checkpointing.
I0424 13:03:05.407589 133269405091648 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 13:03:05.409557 133269405091648 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 10
I0424 13:03:05.411076 133269405091648 event_tracking.py:70] [process=3] [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_15_ocdbt_true/checkpoints/10.
I0424 13:03:05.935394 133269405091648 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 13:03:05.936461 133269405091648 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:03:05.936520 133269405091648 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 13:03:06.215112 133269405091648 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279796s
I0424 13:03:06.215281 133269405091648 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 5.439 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2836008071899414 s) (per-host)
I0424 13:03:06.215333 133269405091648 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.283664s (batch_requests_ready=0.002132s, total_serialization_initiated=0.281461s, others=0.000071s)
I0424 13:03:06.215426 133269405091648 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287873s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.287856)
I0424 13:03:06.216062 133269405091648 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.81 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_15_ocdbt_true/checkpoints/10.
I0424 13:03:06.216359 133146723133184 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:23:06.216319
I0424 13:03:06.228805 133269405091648 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:03:06.229090 133150487340800 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:03:06.229240 133269405091648 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_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777035785.4075694, 'wait_for_prev_duration_secs': 6.270408630371094e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777035785.4095957, 'checkpointer_blocking_duration_secs': 0.8068661689758301, 'get_old_steps_start_time': 1777035786.2164817, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1777035785.4055314, 'checkpoint_manager_blocking_duration_secs': 0.823674201965332}
I0424 13:03:06.229336 133269405091648 checkpointing.py:409] Started an asynchronous checkpoint save for step 10
I0424 13:03:06.229383 133269405091648 max_utils.py:750] 
Memstats: After params initialized:
I0424 13:03:06.229430 133269405091648 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0))
I0424 13:03:06.229460 133269405091648 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0))
I0424 13:03:06.229485 133269405091648 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0))
I0424 13:03:06.229510 133269405091648 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0))
I0424 13:03:06.984195 133269405091648 metric_logger.py:196] completed step: 10, seconds: 11.406, TFLOP/s/device: 1.191, Tokens/s/device: 179.561, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0424 13:03:07.138676 133269405091648 metric_logger.py:196] completed step: 11, seconds: 1.731, TFLOP/s/device: 7.848, Tokens/s/device: 1182.867, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0424 13:03:07.286116 133269405091648 metric_logger.py:196] completed step: 12, seconds: 0.007, TFLOP/s/device: 1925.341, Tokens/s/device: 290208.304, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0424 13:03:07.433459 133269405091648 metric_logger.py:196] completed step: 13, seconds: 0.154, TFLOP/s/device: 88.500, Tokens/s/device: 13339.673, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0424 13:03:07.437899 133269405091648 checkpointing.py:794] Waiting for step 15 to finish before checkpoint...
I0424 13:03:07.727523 133269405091648 checkpointing.py:798] Waited 0.28958821296691895 seconds for step 15 to finish before starting checkpointing.
I0424 13:03:07.730510 133269405091648 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:03:11.507753 133146731525888 array_metadata_store.py:203] [process=3][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_15_ocdbt_true/checkpoints/10/items/array_metadatas/process_3
I0424 13:03:49.254775 133146723133184 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 36.461 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.32305645942688 s) (per-host)
I0424 13:03:49.254904 133146723133184 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.038001s.
I0424 13:03:55.839541 133146723133184 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 49.622621s.
I0424 13:03:55.839847 133150487340800 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:03:55.839983 133150487340800 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:03:55.840036 133150487340800 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:03:55.841721 133150487340800 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0424 13:03:55.841903 133269405091648 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.
W0424 13:03:55.842055 133269405091648 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.111541 seconds. If this number is high, consider checkpointing less frequently.
I0424 13:03:55.843961 133269405091648 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 15
I0424 13:03:55.846387 133269405091648 event_tracking.py:70] [process=3] [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_15_ocdbt_true/checkpoints/15.
I0424 13:03:56.155987 133269405091648 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:03:56.156081 133269405091648 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 13:03:56.181422 133269405091648 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026512s
I0424 13:03:56.181560 133269405091648 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 51.546 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.029926061630249023 s) (per-host)
I0424 13:03:56.181610 133269405091648 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.029983s (batch_requests_ready=0.001783s, total_serialization_initiated=0.028138s, others=0.000062s)
I0424 13:03:56.181691 133269405091648 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035131s (all_items=0.000013s, per_item={'items': '0.00001335'}, temp_paths=0.035118)
I0424 13:03:56.182420 133269405091648 event_tracking.py:125] [process=3] [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_15_ocdbt_true/checkpoints/15.
I0424 13:03:56.182691 133150487340800 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:23:56.182668
I0424 13:03:56.195871 133269405091648 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=15] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:03:56.196132 133144005940992 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:03:56.196265 133269405091648 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_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777035787.7304819, 'wait_for_prev_duration_secs': 48.11154079437256, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777035835.8439987, 'checkpointer_blocking_duration_secs': 0.338808536529541, 'get_old_steps_start_time': 1777035836.1828322, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1777035787.7281141, 'checkpoint_manager_blocking_duration_secs': 48.46811628341675}
I0424 13:03:56.196372 133269405091648 checkpointing.py:409] Started an asynchronous checkpoint save for step 15
I0424 13:03:56.197209 133269405091648 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 92.106, Tokens/s/device: 13883.146, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0424 13:03:56.203466 133269405091648 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.363, Tokens/s/device: 13921.934, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0424 13:04:01.300155 133146731525888 array_metadata_store.py:203] [process=3][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_15_ocdbt_true/checkpoints/15/items/array_metadatas/process_3
I0424 13:04:17.151607 133269405091648 metric_logger.py:196] completed step: 16, seconds: 48.764, TFLOP/s/device: 0.279, Tokens/s/device: 41.998, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0424 13:04:17.298970 133269405091648 metric_logger.py:196] completed step: 17, seconds: 20.802, TFLOP/s/device: 0.653, Tokens/s/device: 98.451, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0424 13:04:17.446120 133269405091648 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.212, Tokens/s/device: 13447.056, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0424 13:04:17.592847 133269405091648 checkpointing.py:794] Waiting for step 19 to finish before checkpoint...
I0424 13:04:17.593576 133269405091648 checkpointing.py:798] Waited 0.0007467269897460938 seconds for step 19 to finish before starting checkpointing.
I0424 13:04:17.595607 133269405091648 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=15][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:04:37.745433 133150487340800 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.977 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.593761920928955 s) (per-host)
I0424 13:04:37.745551 133150487340800 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.562767s.
I0424 13:04:46.196956 133150487340800 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 50.014137s.
I0424 13:04:46.197173 133144005940992 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:04:46.197238 133144005940992 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:04:46.197288 133144005940992 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:04:46.198954 133144005940992 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=15] CheckpointManager Save Finalize is done on all hosts.
I0424 13:04:46.199134 133269405091648 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.
W0424 13:04:46.199267 133269405091648 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.603662 seconds. If this number is high, consider checkpointing less frequently.
I0424 13:04:46.201509 133269405091648 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 19
I0424 13:04:46.203488 133269405091648 event_tracking.py:70] [process=3] [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_15_ocdbt_true/checkpoints/19.
I0424 13:04:46.515672 133269405091648 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 13:04:46.515768 133269405091648 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 13:04:46.541869 133269405091648 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027137s
I0424 13:04:46.542053 133269405091648 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 50.269 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.030686616897583008 s) (per-host)
I0424 13:04:46.542104 133269405091648 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.030749s (batch_requests_ready=0.001680s, total_serialization_initiated=0.029002s, others=0.000067s)
I0424 13:04:46.542200 133269405091648 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035060s (all_items=0.000010s, per_item={'items': '0.00001049'}, temp_paths=0.035049)
I0424 13:04:46.542885 133269405091648 event_tracking.py:125] [process=3] [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_15_ocdbt_true/checkpoints/19.
I0424 13:04:46.543216 133144005940992 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:24:46.543180
I0424 13:04:46.546980 133269405091648 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:04:46.547235 133146731525888 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:04:46.547362 133269405091648 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_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777035857.5955796, 'wait_for_prev_duration_secs': 28.603662252426147, 'time_between_consecutive_saves_sec': 21.75381636619568, 'checkpointer_blocking_start_time': 1777035886.20155, 'checkpointer_blocking_duration_secs': 0.3418090343475342, 'get_old_steps_start_time': 1777035886.5433753, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1777035857.5938148, 'checkpoint_manager_blocking_duration_secs': 28.95351505279541}
I0424 13:04:46.547470 133269405091648 checkpointing.py:409] Started an asynchronous checkpoint save for step 19
I0424 13:04:46.547519 133269405091648 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:04:51.811593 133146686506752 array_metadata_store.py:203] [process=3][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_15_ocdbt_true/checkpoints/19/items/array_metadatas/process_3
I0424 13:05:28.938584 133144005940992 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.231 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.42717432975769 s) (per-host)
I0424 13:05:28.938706 133144005940992 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.395376s.
I0424 13:05:35.857604 133144005940992 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 49.314258s.
I0424 13:05:35.857841 133146731525888 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:05:35.857901 133146731525888 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:05:35.857953 133146731525888 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:05:35.859605 133146731525888 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0424 13:05:35.859824 133269405091648 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.
I0424 13:05:35.860008 133269405091648 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 13:05:35.860916 133269405091648 metric_logger.py:196] completed step: 19, seconds: 0.147, TFLOP/s/device: 92.356, Tokens/s/device: 13920.893, 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 13:05:45 UTC 2026
EXIT_CODE=0