MaxView

‹ 13_scan_layers_falseCase: 14_async_ckpt_false_resume14_async_ckpt_false_save ›

Metrics: Linen vs NNX  ·  main

MetricLinen  b117f50cfNNX  b117f50cfDiff (NNX − Linen)
Parameters1.104 billion
Final loss5.3910
TFLOP/s92.088
Tok/s13880.5
Avg s/step13.793
Memory %1.38
JAX0.9.20.9.2

Diff = NNX value − Linen value. Green = NNX improved. Red = NNX regressed.

Linen  ·  b117f50cf  ·  main_20260424_070227  ·  full log
XPK Start: Fri Apr 24 07:53:57 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 07:54:22.284986: 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 07:54:22.490343 133420718110528 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 07:54:31.532402 133420718110528 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 07:54:31.534711 133420718110528 distributed.py:172] Connecting to JAX distributed service on mt-14-async-ckpt-false--ipqbh-slice-job-0-0.mt-14-async-ckpt-false--ipqbh:8482
I0424 07:54:32.710379 133420718110528 max_utils.py:284] Jax distributed system initialized!
I0424 07:54:38.846129 133420718110528 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 07:54:38.846233 133420718110528 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 07:54:38.846276 133420718110528 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 07:54:38.846310 133420718110528 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 07:54:39.573569 133420718110528 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 07:54:39.573847 133420718110528 checkpointing.py:677] Setting up checkpoint logger...
I0424 07:54:39.573901 133420718110528 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 07:54:39.573945 133420718110528 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 07:54:39.574305 133420718110528 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 0x7957da918dd0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 07:54:40.833438 133420718110528 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0424 07:54:40.833686 133420718110528 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 0x7943786a45f0>}, handler_registry=None
I0424 07:54:40.833960 133420718110528 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7943786a45f0>` 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 07:54:40.834016 133420718110528 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7943900d1eb0>` 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 07:54:40.834052 133420718110528 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 0x7943786a45f0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7943786a45f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7943900d1eb0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7943900d1eb0>}).
I0424 07:54:40.834432 133420718110528 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 07:54:42.351407 133299502556928 checkpoint.py:228] Read Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1777016990090517140, 'commit_timestamp_nsecs': 1777017033373552865, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/0/_CHECKPOINT_METADATA
I0424 07:54:42.595985 133299502556928 checkpoint_manager.py:1871] Missing metrics for step 0
I0424 07:54:42.609380 133420718110528 checkpoint_manager.py:1812] Found 3 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints
I0424 07:54:43.033157 133420718110528 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=False, 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_14_async_ckpt_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7943786a1d60>
I0424 07:54:43.033344 133420718110528 checkpointing.py:301] Checkpoint manager created!
I0424 07:54:43.969981 133420718110528 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 07:54:43.970118 133420718110528 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:54:44.348542 133420718110528 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 07:54:44.348635 133420718110528 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:54:44.365144 133420718110528 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 07:54:44.365201 133420718110528 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 07:54:44.394579 133420718110528 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 07:54:44.394676 133420718110528 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:54:44.412151 133420718110528 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 07:54:44.412217 133420718110528 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:54:44.428781 133420718110528 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 07:54:44.428844 133420718110528 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:54:44.453915 133420718110528 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 07:54:44.453987 133420718110528 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 07:54:44.475573 133420718110528 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 07:54:44.475637 133420718110528 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 07:54:44.686570 133420718110528 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 07:54:44.686682 133420718110528 checkpointing.py:581] restoring from this run's directory step 9
I0424 07:54:44.935317 133420718110528 event_tracking.py:70] [process=6] [sync] Started load checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/9.
I0424 07:54:45.180170 133420718110528 checkpointer.py:307] Restoring checkpoint from gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/9.
I0424 07:54:45.180611 133420718110528 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_14_async_ckpt_false/checkpoints/9.
I0424 07:54:46.585895    1453 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 07:54:53.614778 133420718110528 jax_array_handlers.py:843] [process=6] /jax/orbax/read/worker/io/requested throughput: 224.641 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 7.034670114517212 s) (per-host)
I0424 07:54:53.620788 133420718110528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/checkpoint/read/gbytes_per_sec: 212.831 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 7.435920000076294 s) (per-host)
I0424 07:54:53.622827 133420718110528 event_tracking.py:138] [process=6] [sync] Finished load in 8.69 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/9
I0424 07:54:53.622922 133420718110528 standard_logger.py:34] {'step': 9, 'event_type': 'restore', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints', 'checkpointer_start_time': 1777017284.9352345, 'checkpointer_duration_secs': 8.687658071517944, 'checkpoint_manager_start_time': 1777017284.686943, 'checkpoint_manager_duration_secs': 8.935950756072998}
fsdp: 32
I0424 07:54:54.046281 133420718110528 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 07:54:54.046419 133420718110528 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 07:54:54.046472 133420718110528 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 07:54:54.046531 133420718110528 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 07:54:54.046586 133420718110528 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 07:54:54.046626 133420718110528 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 07:54:54.046679 133420718110528 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 07:54:54.046731 133420718110528 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 07:54:54.046771 133420718110528 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 07:54:54.046809 133420718110528 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 07:54:54.046855 133420718110528 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0424 07:54:54.046901 133420718110528 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0424 07:54:54.535663 133420718110528 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 07:54:54.535756 133420718110528 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 07:54:54.551498 133420718110528 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 07:54:54.551559 133420718110528 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 07:55:05.493843 133420718110528 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 07:55:05.494627 133420718110528 metric_logger.py:301] number parameters: 1.104 billion
I0424 07:55:16.874961 133420718110528 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0424 07:55:17.215196 133420718110528 checkpointing.py:776] Waited 0.3402116298675537 seconds for step 10 to finish before starting checkpointing.
I0424 07:55:17.217718 133420718110528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:55:17.219681 133420718110528 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0424 07:55:17.221038 133420718110528 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/10.
I0424 07:55:19.856446 133420718110528 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 07:55:19.857808 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:55:21.238875 133420718110528 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 07:55:21.238968 133420718110528 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 07:55:21.522014 133420718110528 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.284412s
I0424 07:55:21.529307 133420718110528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.218 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2956211566925049 s) (per-host)
I0424 07:55:21.529379 133420718110528 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.295705s (batch_requests_ready=0.002306s, total_serialization_initiated=0.293304s, others=0.000096s)
I0424 07:55:21.529479 133420718110528 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.673122s (all_items=0.000038s, per_item={'items': '0.00003767'}, temp_paths=1.673085)
I0424 07:55:21.529533 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:55:21.529576 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=1] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:55:23.977466 133298906969856 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_14_async_ckpt_false/checkpoints/10/items/array_metadatas/process_6
I0424 07:55:59.743761 133420718110528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 41.018 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.51004219055176 s) (per-host)
I0424 07:56:05.459482 133420718110528 event_tracking.py:125] [process=6] [sync] Finished blocking save in 48.24 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/10.
I0424 07:56:09.450749 133420718110528 event_tracking.py:138] [process=6] [sync] Finished save in 52.23 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/10
I0424 07:56:09.452641 133420718110528 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0424 07:56:09.454177 133420718110528 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=10] CheckpointManager Save Finalize is done on all hosts.
I0424 07:56:09.454228 133420718110528 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=10] Finished synchronous save.
I0424 07:56:09.454279 133420718110528 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1777017317.2177, 'wait_for_prev_duration_secs': 6.103515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777017317.2197204, 'checkpointer_blocking_duration_secs': 52.231146812438965, 'get_old_steps_start_time': 1777017369.4508874, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1777017317.215722, 'checkpoint_manager_blocking_duration_secs': 52.2385311126709}
I0424 07:56:09.454365 133420718110528 checkpointing.py:410] Saved a checkpoint at step 10.
I0424 07:56:09.454410 133420718110528 max_utils.py:750] 
Memstats: After params initialized:
I0424 07:56:09.454456 133420718110528 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0424 07:56:09.454488 133420718110528 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0424 07:56:09.454522 133420718110528 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0424 07:56:09.454546 133420718110528 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0424 07:56:09.776058 133420718110528 metric_logger.py:196] completed step: 10, seconds: 11.380, TFLOP/s/device: 1.194, Tokens/s/device: 179.961, total_weights: 65536, loss: 6.833, lm_loss: 6.833, perplexity: 927.869
I0424 07:56:09.939977 133420718110528 metric_logger.py:196] completed step: 11, seconds: 52.899, TFLOP/s/device: 0.257, Tokens/s/device: 38.715, total_weights: 65536, loss: 6.527, lm_loss: 6.527, perplexity: 683.659
I0424 07:56:10.087514 133420718110528 metric_logger.py:196] completed step: 12, seconds: 0.016, TFLOP/s/device: 826.769, Tokens/s/device: 124619.691, total_weights: 65536, loss: 6.265, lm_loss: 6.265, perplexity: 526.068
I0424 07:56:10.234935 133420718110528 metric_logger.py:196] completed step: 13, seconds: 0.154, TFLOP/s/device: 88.416, Tokens/s/device: 13327.000, total_weights: 65536, loss: 6.045, lm_loss: 6.045, perplexity: 422.001
I0424 07:56:10.239413 133420718110528 checkpointing.py:772] Waiting for step 15 to finish before checkpoint...
I0424 07:56:10.529072 133420718110528 checkpointing.py:776] Waited 0.28962063789367676 seconds for step 15 to finish before starting checkpointing.
I0424 07:56:10.532142 133420718110528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:56:10.533776 133420718110528 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 15
I0424 07:56:10.535613 133420718110528 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/15.
I0424 07:56:13.243245 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:56:14.629536 133420718110528 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 07:56:14.629639 133420718110528 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 07:56:14.658259 133420718110528 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.029830s
I0424 07:56:14.658442 133420718110528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 45.277 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03407001495361328 s) (per-host)
I0424 07:56:14.658502 133420718110528 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.034139s (batch_requests_ready=0.001928s, total_serialization_initiated=0.032133s, others=0.000077s)
I0424 07:56:14.658604 133420718110528 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.416684s (all_items=0.000026s, per_item={'items': '0.00002551'}, temp_paths=1.416659)
I0424 07:56:14.658646 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:56:14.658686 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=2] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:56:16.687513 133291961218816 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_14_async_ckpt_false/checkpoints/15/items/array_metadatas/process_6
I0424 07:56:53.043599 133420718110528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 41.115 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.419188499450684 s) (per-host)
I0424 07:56:58.215442 133420718110528 event_tracking.py:125] [process=6] [sync] Finished blocking save in 47.68 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/15.
I0424 07:57:02.269626 133420718110528 event_tracking.py:138] [process=6] [sync] Finished save in 51.74 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/15
I0424 07:57:02.271617 133420718110528 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=15] CheckpointManager Save Finalize is syncing with other hosts...
I0424 07:57:02.273449 133420718110528 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=15] CheckpointManager Save Finalize is done on all hosts.
I0424 07:57:02.273497 133420718110528 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=15] Finished synchronous save.
I0424 07:57:02.273547 133420718110528 standard_logger.py:34] {'step': 15, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1777017370.53212, 'wait_for_prev_duration_secs': 6.031990051269531e-05, 'time_between_consecutive_saves_sec': 1.0779030323028564, 'checkpointer_blocking_start_time': 1777017370.5338147, 'checkpointer_blocking_duration_secs': 51.73592686653137, 'get_old_steps_start_time': 1777017422.2697625, 'get_old_steps_duration_secs': 2.4318695068359375e-05, 'checkpoint_manager_blocking_start_time': 1777017370.5297945, 'checkpoint_manager_blocking_duration_secs': 51.74372911453247}
I0424 07:57:02.273635 133420718110528 checkpointing.py:410] Saved a checkpoint at step 15.
I0424 07:57:02.274445 133420718110528 metric_logger.py:196] completed step: 14, seconds: 0.148, TFLOP/s/device: 92.061, Tokens/s/device: 13876.373, total_weights: 65536, loss: 5.864, lm_loss: 5.864, perplexity: 352.193
I0424 07:57:02.280513 133420718110528 metric_logger.py:196] completed step: 15, seconds: 0.147, TFLOP/s/device: 92.292, Tokens/s/device: 13911.248, total_weights: 65536, loss: 5.719, lm_loss: 5.719, perplexity: 304.661
I0424 07:57:23.266981 133420718110528 metric_logger.py:196] completed step: 16, seconds: 52.040, TFLOP/s/device: 0.261, Tokens/s/device: 39.355, total_weights: 65536, loss: 5.605, lm_loss: 5.605, perplexity: 271.872
I0424 07:57:23.414516 133420718110528 metric_logger.py:196] completed step: 17, seconds: 20.841, TFLOP/s/device: 0.652, Tokens/s/device: 98.270, total_weights: 65536, loss: 5.517, lm_loss: 5.517, perplexity: 248.975
I0424 07:57:23.562039 133420718110528 metric_logger.py:196] completed step: 18, seconds: 0.152, TFLOP/s/device: 89.311, Tokens/s/device: 13461.905, total_weights: 65536, loss: 5.448, lm_loss: 5.448, perplexity: 232.333
I0424 07:57:23.708681 133420718110528 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0424 07:57:23.709381 133420718110528 checkpointing.py:776] Waited 0.0007219314575195312 seconds for step 19 to finish before starting checkpointing.
I0424 07:57:23.711454 133420718110528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:57:23.713368 133420718110528 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0424 07:57:23.714695 133420718110528 event_tracking.py:70] [process=6] [sync] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/19.
I0424 07:57:26.160840 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:57:27.604125 133420718110528 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 07:57:27.604230 133420718110528 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 07:57:27.632116 133420718110528 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.029129s
I0424 07:57:27.632274 133420718110528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 47.012 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03281235694885254 s) (per-host)
I0424 07:57:27.632328 133420718110528 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.032876s (batch_requests_ready=0.001808s, total_serialization_initiated=0.030999s, others=0.000068s)
I0424 07:57:27.632414 133420718110528 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.472930s (all_items=0.000021s, per_item={'items': '0.00002050'}, temp_paths=1.472910)
I0424 07:57:27.632452 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:57:27.632489 133420718110528 future.py:372] [process=6][thread=MainThread][operation_id=3] _SignalingThread.join() waiting for signals ([]) blocking the main thread will slow down blocking save times. This is likely due to main thread calling result() on a CommitFuture.
I0424 07:57:29.678317 133298906969856 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_14_async_ckpt_false/checkpoints/19/items/array_metadatas/process_6
I0424 07:58:06.094718 133420718110528 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 41.034 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.4952175617218 s) (per-host)
I0424 07:58:11.648401 133420718110528 event_tracking.py:125] [process=6] [sync] Finished blocking save in 47.93 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/19.
I0424 07:58:15.402336 133420718110528 event_tracking.py:138] [process=6] [sync] Finished save in 51.69 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints/19
I0424 07:58:15.405076 133420718110528 checkpoint_manager.py:2137] [process=6][thread=MainThread][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0424 07:58:15.406939 133420718110528 checkpoint_manager.py:2146] [process=6][thread=MainThread][step=19] CheckpointManager Save Finalize is done on all hosts.
I0424 07:58:15.406989 133420718110528 checkpoint_manager.py:1581] [process=6][thread=MainThread][step=19] Finished synchronous save.
I0424 07:58:15.407039 133420718110528 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260424_070227/linen_xpk_main_20260424_070227_14_async_ckpt_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': True, 'wait_for_prev_start_time': 1777017443.7114382, 'wait_for_prev_duration_secs': 6.103515625e-05, 'time_between_consecutive_saves_sec': 21.43795108795166, 'checkpointer_blocking_start_time': 1777017443.7134054, 'checkpointer_blocking_duration_secs': 51.689046144485474, 'get_old_steps_start_time': 1777017495.4024718, 'get_old_steps_duration_secs': 2.0742416381835938e-05, 'checkpoint_manager_blocking_start_time': 1777017443.709618, 'checkpoint_manager_blocking_duration_secs': 51.69739747047424}
I0424 07:58:15.407137 133420718110528 checkpointing.py:410] Saved a checkpoint at step 19.
I0424 07:58:15.407169 133420718110528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:58:15.407200 133420718110528 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 07:58:15.407829 133420718110528 metric_logger.py:196] completed step: 19, seconds: 0.148, TFLOP/s/device: 92.088, Tokens/s/device: 13880.511, 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 07:58:27 UTC 2026
EXIT_CODE=0
NNX  ·  b117f50cf  ·  main_20260424_070227  ·  full log
XPK Start: Fri Apr 24 08:55:31 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 08:55:56.491215: 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 08:55:56.697008 133550322825024 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 08:56:05.738738 133550322825024 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 08:56:05.741216 133550322825024 distributed.py:172] Connecting to JAX distributed service on mt-14-async-ckpt-false--l90sx-slice-job-0-0.mt-14-async-ckpt-false--l90sx:8482
I0424 08:56:06.802973 133550322825024 max_utils.py:284] Jax distributed system initialized!
I0424 08:56:12.964069 133550322825024 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 08:56:12.964174 133550322825024 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 08:56:12.964214 133550322825024 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 08:56:12.964249 133550322825024 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/deps/src/maxtext/trainers/pre_train/train.py", line 744, in <module>
    app.run(main)
  File "/usr/local/lib/python3.12/site-packages/absl/app.py", line 367, in run
    _run_main(main, args)
  File "/usr/local/lib/python3.12/site-packages/absl/app.py", line 312, in _run_main
    sys.exit(main(argv))
             ^^^^^^^^^^
  File "/deps/src/maxtext/trainers/pre_train/train.py", line 740, in main
    train_func()
  File "/deps/src/maxtext/trainers/pre_train/train.py", line 730, in train_func
    run(config, recorder, diagnostic_config)
  File "/deps/src/maxtext/trainers/pre_train/train.py", line 709, in run
    train_loop(config, recorder)
  File "/deps/src/maxtext/trainers/pre_train/train.py", line 536, in train_loop
    ) = train_utils.setup_train_loop(config, recorder)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/deps/src/maxtext/utils/train_utils.py", line 218, in setup_train_loop
    raise NotImplementedError("Pure NNX support has not been implemented yet.")
NotImplementedError: Pure NNX support has not been implemented yet.
XPK End: Fri Apr 24 08:56:21 UTC 2026
EXIT_CODE=1