MaxView

‹ 10_shardy_falseCase: 11_optimizer_offload_true13_scan_layers_false ›

Metrics: Linen vs NNX  ·  test/pipeline-scan-nnx

MetricLinen  5b21b66fcNNX  5b21b66fcDiff (NNX − Linen)
Parameters1.104 billion
Final loss8.1790
TFLOP/s83.447
Tok/s12578.0
Avg s/step2.741
Memory %0.54
JAX0.9.20.9.2

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

Linen  ·  5b21b66fc  ·  test_pipeline_scan_nnx_20260423_211617  ·  full log
XPK Start: Thu Apr 23 21:51:51 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-23 21:52:16.715052: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0423 21:52:16.925341 136509572691776 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 21:52:25.968143 136509572691776 distributed.py:149] Starting JAX distributed service on [::]:8482
I0423 21:52:25.970394 136509572691776 distributed.py:172] Connecting to JAX distributed service on mt-11-optimizer-offload-wwzq4-slice-job-0-0.mt-11-optimizer-offload-wwzq4:8482
I0423 21:52:26.900855 136509572691776 max_utils.py:284] Jax distributed system initialized!
I0423 21:52:33.140812 136509572691776 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 21:52:33.140937 136509572691776 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 21:52:33.140979 136509572691776 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0423 21:52:33.141014 136509572691776 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 21:52:33.841624 136509572691776 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 21:52:33.841927 136509572691776 checkpointing.py:677] Setting up checkpoint logger...
I0423 21:52:33.841999 136509572691776 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 21:52:33.842055 136509572691776 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0423 21:52:33.842433 136509572691776 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 0x7c26e503f290>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 21:52:36.706849 136509572691776 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0423 21:52:36.707121 136509572691776 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c12c0757980>}, handler_registry=None
I0423 21:52:36.707390 136509572691776 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c12c0757980>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0423 21:52:36.707446 136509572691776 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c12c0564f80>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0423 21:52:36.707496 136509572691776 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 0x7c12c0757980>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c12c0757980>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c12c0564f80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c12c0564f80>}).
I0423 21:52:36.707847 136509572691776 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0423 21:52:36.707940 136509572691776 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c12c0383560> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0423 21:52:37.471064 136509572691776 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints
I0423 21:52:37.919344 136509572691776 checkpoint_manager.py:929] [process=4][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=10), 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_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c12c0564680>
I0423 21:52:37.919518 136509572691776 checkpointing.py:301] Checkpoint manager created!
I0423 21:52:38.848312 136509572691776 nnx_wrappers.py:453] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 21:52:38.848433 136509572691776 nnx_wrappers.py:453] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 21:52:39.228363 136509572691776 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 21:52:39.228452 136509572691776 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 21:52:39.244797 136509572691776 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 21:52:39.244857 136509572691776 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 21:52:39.268456 136509572691776 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 21:52:39.268527 136509572691776 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 21:52:39.285010 136509572691776 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 21:52:39.285070 136509572691776 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 21:52:39.301495 136509572691776 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 21:52:39.301554 136509572691776 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 21:52:39.326039 136509572691776 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 21:52:39.326108 136509572691776 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 21:52:39.347254 136509572691776 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 21:52:39.347323 136509572691776 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 21:52:39.555404 136509572691776 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 21:52:39.555510 136509572691776 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0423 21:52:41.147478 136509572691776 maxtext_utils.py:1668]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0423 21:52:41.147612 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 21:52:41.147665 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 21:52:41.147723 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 21:52:41.147775 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 21:52:41.147812 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 21:52:41.147863 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 21:52:41.147931 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0423 21:52:41.147973 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0423 21:52:41.148009 136509572691776 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 21:52:41.148055 136509572691776 maxtext_utils.py:1668]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0423 21:52:41.148101 136509572691776 maxtext_utils.py:1668]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]

    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0423 21:52:41.637727 136509572691776 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 21:52:41.637820 136509572691776 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 21:52:41.653441 136509572691776 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 21:52:41.653500 136509572691776 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 21:52:52.734708 136509572691776 max_utils.py:791] Total memory size: 1.3 GB, Output size: 0.1 GB, Temp size: 1.1 GB, Argument size: 0.1 GB, Host temp size: 0.0 GB.
I0423 21:52:52.735536 136509572691776 metric_logger.py:301] number parameters: 1.104 billion
I0423 21:53:04.397493 136509572691776 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0423 21:53:04.565139 136509572691776 checkpointing.py:776] Waited 0.16762614250183105 seconds for step 0 to finish before starting checkpointing.
I0423 21:53:04.567772 136509572691776 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 21:53:04.569335 136509572691776 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0
I0423 21:53:04.570777 136509572691776 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints/0.
I0423 21:53:04.897676 136509572691776 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 21:53:04.898473 136509572691776 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 21:53:04.898539 136509572691776 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0423 21:53:05.173907 136509572691776 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.276310s
I0423 21:53:05.174086 136509572691776 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.448 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2831306457519531 s) (per-host)
I0423 21:53:05.174141 136509572691776 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.283196s (batch_requests_ready=0.002308s, total_serialization_initiated=0.280815s, others=0.000073s)
I0423 21:53:05.174240 136509572691776 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287953s (all_items=0.000024s, per_item={'items': '0.00002384'}, temp_paths=0.287929)
I0423 21:53:05.175126 136509572691776 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints/0.
I0423 21:53:05.175446 136362880534272 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 22:13:05.175409
I0423 21:53:05.177548 136509572691776 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 21:53:05.177865 136362674042624 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0423 21:53:05.178038 136509572691776 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776981184.5677547, 'wait_for_prev_duration_secs': 6.103515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776981184.569372, 'checkpointer_blocking_duration_secs': 0.606231689453125, 'get_old_steps_start_time': 1776981185.1756306, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776981184.565703, 'checkpoint_manager_blocking_duration_secs': 0.6122968196868896}
I0423 21:53:05.178146 136509572691776 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0423 21:53:05.178194 136509572691776 max_utils.py:750] 
Memstats: After params initialized:
I0423 21:53:05.178243 136509572691776 max_utils.py:756] 	Using (GB) 0.17 / 31.25 (0.544000%) on TPU_16(process=4,(0,4,0,0))
I0423 21:53:05.178276 136509572691776 max_utils.py:756] 	Using (GB) 0.17 / 31.25 (0.544000%) on TPU_17(process=4,(1,4,0,0))
I0423 21:53:05.178307 136509572691776 max_utils.py:756] 	Using (GB) 0.17 / 31.25 (0.544000%) on TPU_20(process=4,(0,5,0,0))
I0423 21:53:05.178333 136509572691776 max_utils.py:756] 	Using (GB) 0.17 / 31.25 (0.544000%) on TPU_21(process=4,(1,5,0,0))
I0423 21:53:05.491998 136509572691776 metric_logger.py:196] completed step: 0, seconds: 11.662, TFLOP/s/device: 1.165, Tokens/s/device: 175.615, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0423 21:53:05.693042 136509572691776 metric_logger.py:196] completed step: 1, seconds: 1.093, TFLOP/s/device: 12.432, Tokens/s/device: 1873.823, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0423 21:53:06.497130 136509572691776 metric_logger.py:196] completed step: 2, seconds: 0.038, TFLOP/s/device: 354.931, Tokens/s/device: 53499.125, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0423 21:53:06.659751 136509572691776 metric_logger.py:196] completed step: 3, seconds: 0.804, TFLOP/s/device: 16.899, Tokens/s/device: 2547.232, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.047
I0423 21:53:06.986422 136509572691776 metric_logger.py:196] completed step: 4, seconds: 0.169, TFLOP/s/device: 80.524, Tokens/s/device: 12137.447, total_weights: 65536, loss: 9.278, lm_loss: 9.278, perplexity: 10694.968
I0423 21:53:06.992186 136509572691776 metric_logger.py:196] completed step: 5, seconds: 0.163, TFLOP/s/device: 83.511, Tokens/s/device: 12587.662, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.520
I0423 21:53:08.666215    2764 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 21:53:10.689501 136362682435328 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints/0/items/array_metadatas/process_4
I0423 21:53:28.897926 136509572691776 metric_logger.py:196] completed step: 6, seconds: 0.327, TFLOP/s/device: 41.543, Tokens/s/device: 6261.886, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.688
I0423 21:53:29.060802 136509572691776 metric_logger.py:196] completed step: 7, seconds: 21.743, TFLOP/s/device: 0.625, Tokens/s/device: 94.190, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4377.010
I0423 21:53:29.223469 136509572691776 metric_logger.py:196] completed step: 8, seconds: 0.168, TFLOP/s/device: 80.904, Tokens/s/device: 12194.686, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3845.943
I0423 21:53:29.385724 136509572691776 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0423 21:53:29.386410 136509572691776 checkpointing.py:776] Waited 0.0007078647613525391 seconds for step 9 to finish before starting checkpointing.
I0423 21:53:29.388500 136509572691776 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 21:53:42.309012 136362880534272 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 42.213 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.418001651763916 s) (per-host)
I0423 21:53:42.309135 136362880534272 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.133575s.
I0423 21:53:51.194273 136362880534272 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 46.018697s.
I0423 21:53:51.194592 136362674042624 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 21:53:51.194738 136362674042624 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0423 21:53:51.194792 136362674042624 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 21:53:51.196194 136362674042624 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 21:53:51.196371 136509572691776 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0423 21:53:51.196509 136509572691776 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.808011 seconds. If this number is high, consider checkpointing less frequently.
I0423 21:53:51.198247 136509572691776 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9
I0423 21:53:51.200302 136509572691776 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints/9.
I0423 21:53:51.851405 136509572691776 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 21:53:51.851500 136509572691776 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0423 21:53:51.887134 136509572691776 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036648s
I0423 21:53:51.887300 136509572691776 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 38.559 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04000377655029297 s) (per-host)
I0423 21:53:51.887355 136509572691776 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040068s (batch_requests_ready=0.001787s, total_serialization_initiated=0.038210s, others=0.000072s)
I0423 21:53:51.887443 136509572691776 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043989s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.043975)
I0423 21:53:51.888109 136509572691776 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.69 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints/9.
I0423 21:53:51.888463 136362674042624 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 22:13:51.888426
I0423 21:53:51.899283 136509572691776 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 21:53:51.899579 136362754709248 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0423 21:53:51.899747 136509572691776 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776981209.3884687, 'wait_for_prev_duration_secs': 21.808010816574097, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776981231.1982853, 'checkpointer_blocking_duration_secs': 0.6903238296508789, 'get_old_steps_start_time': 1776981231.8886316, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776981209.3866534, 'checkpoint_manager_blocking_duration_secs': 22.513060808181763}
I0423 21:53:51.899853 136509572691776 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0423 21:53:51.899912 136509572691776 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 21:53:57.785613 136362682435328 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260423_211617/linen_xpk_test_pipeline_scan_nnx_20260423_211617_11_optimizer_offload_true/checkpoints/9/items/array_metadatas/process_4
I0423 21:54:33.911680 136362674042624 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.550 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.064348459243774 s) (per-host)
I0423 21:54:33.911801 136362674042624 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.023220s.
I0423 21:54:42.694315 136362674042624 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 50.805718s.
I0423 21:54:42.694606 136362754709248 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 21:54:42.694724 136362754709248 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0423 21:54:42.694774 136362754709248 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 21:54:42.696403 136362754709248 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 21:54:42.696592 136509572691776 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0423 21:54:42.696768 136509572691776 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 21:54:42.697773 136509572691776 metric_logger.py:196] completed step: 9, seconds: 0.163, TFLOP/s/device: 83.447, Tokens/s/device: 12577.998, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.739
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Thu Apr 23 21:54:51 UTC 2026
EXIT_CODE=0
NNX  ·  5b21b66fc  ·  test_pipeline_scan_nnx_20260423_211617  ·  full log
XPK Start: Thu Apr 23 23:02:46 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-23 23:03:11.152325: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0423 23:03:11.367714 137675962189632 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 23:03:20.409936 137675962189632 distributed.py:172] Connecting to JAX distributed service on mt-11-optimizer-offload-e91ik-slice-job-0-0.mt-11-optimizer-offload-e91ik:8482
I0423 23:03:20.454643 137675962189632 max_utils.py:284] Jax distributed system initialized!
I0423 23:03:25.439433 137675962189632 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 23:03:25.439545 137675962189632 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 23:03:25.439586 137675962189632 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0423 23:03:25.439618 137675962189632 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: Thu Apr 23 23:03:31 UTC 2026
EXIT_CODE=1