MaxView

‹ —Case: 01_base02_synthetic ›

Metrics: Linen vs NNX  ·  feat/nnx-trainstate-and-training-loop

MetricLinen  f093f3730NNX  f093f3730Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s92.26691.762-0.504
Tok/s13907.413831.3-76.079
Avg s/step2.6502.701+0.051
Memory %1.381.380
JAX0.9.20.9.2

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

XPK Start: Fri Apr 24 09:13:55 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 09:14:20.236331: 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 09:14:20.447774 134495874074432 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 09:14:29.488528 134495874074432 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 09:14:29.491110 134495874074432 distributed.py:172] Connecting to JAX distributed service on mt-01-base-xrp39-slice-job-0-0.mt-01-base-xrp39:8482
I0424 09:14:33.820415 134495874074432 max_utils.py:284] Jax distributed system initialized!
I0424 09:14:38.902365 134495874074432 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 09:14:38.902475 134495874074432 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 09:14:38.902516 134495874074432 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 09:14:38.902551 134495874074432 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 09:14:39.601745 134495874074432 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 09:14:39.602343 134495874074432 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 09:14:39.602527 134495874074432 checkpointing.py:688] Setting up checkpoint logger...
I0424 09:14:39.602578 134495874074432 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 09:14:39.602622 134495874074432 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 09:14:39.602977 134495874074432 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 0x7a520e417350>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 09:14:43.334277 134495874074432 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 09:14:43.334517 134495874074432 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 0x7a50fee40770>}, handler_registry=None
I0424 09:14:43.334759 134495874074432 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a50fee40770>` 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 09:14:43.334810 134495874074432 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a3e9445c980>` 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 09:14:43.334845 134495874074432 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 0x7a50fee40770>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a50fee40770>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a3e9445c980>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a3e9445c980>}).
I0424 09:14:43.335185 134495874074432 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 09:14:43.335255 134495874074432 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7a3e94285e40> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 09:14:44.295600 134495874074432 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints
I0424 09:14:44.755124 134495874074432 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=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_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7a3e9445b7a0>
I0424 09:14:44.755298 134495874074432 checkpointing.py:302] Checkpoint manager created!
I0424 09:14:45.707247 134495874074432 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 09:14:45.707354 134495874074432 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 09:14:46.090115 134495874074432 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 09:14:46.090209 134495874074432 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 09:14:46.106697 134495874074432 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 09:14:46.106761 134495874074432 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 09:14:46.130571 134495874074432 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 09:14:46.130642 134495874074432 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 09:14:46.147169 134495874074432 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 09:14:46.147229 134495874074432 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 09:14:46.163745 134495874074432 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 09:14:46.163805 134495874074432 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 09:14:46.189024 134495874074432 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 09:14:46.189092 134495874074432 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 09:14:46.210020 134495874074432 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 09:14:46.210087 134495874074432 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 09:14:46.428377 134495874074432 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 09:14:46.428490 134495874074432 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 09:14:47.864316 134495874074432 maxtext_utils.py:1874]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 09:14:47.864446 134495874074432 maxtext_utils.py:1874]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 09:14:47.864502 134495874074432 maxtext_utils.py:1874]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 09:14:47.864560 134495874074432 maxtext_utils.py:1874]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0424 09:14:47.864612 134495874074432 maxtext_utils.py:1874]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 09:14:47.864650 134495874074432 maxtext_utils.py:1874]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 09:14:47.864701 134495874074432 maxtext_utils.py:1874]  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 09:14:47.864754 134495874074432 maxtext_utils.py:1874]  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 09:14:47.864795 134495874074432 maxtext_utils.py:1874]  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 09:14:47.864834 134495874074432 maxtext_utils.py:1874]  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 09:14:47.864884 134495874074432 maxtext_utils.py:1874]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0424 09:14:47.864973 134495874074432 maxtext_utils.py:1874]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')

    Physical:  (None, 'fsdp')
I0424 09:14:48.366092 134495874074432 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 09:14:48.366184 134495874074432 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 09:14:48.381670 134495874074432 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 09:14:48.381740 134495874074432 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 09:14:59.419539 134495874074432 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 09:14:59.420279 134495874074432 metric_logger.py:301] number parameters: 1.104 billion
I0424 09:15:11.025703 134495874074432 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 09:15:11.179007 134495874074432 checkpointing.py:798] Waited 0.15328478813171387 seconds for step 0 to finish before starting checkpointing.
I0424 09:15:11.181435 134495874074432 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 09:15:11.183163 134495874074432 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0
I0424 09:15:11.184493 134495874074432 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/0.
I0424 09:15:11.511520 134495874074432 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 09:15:11.512453 134495874074432 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 09:15:11.512508 134495874074432 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 09:15:11.787460 134495874074432 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.276013s
I0424 09:15:11.787661 134495874074432 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 5.476 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2816798686981201 s) (per-host)
I0424 09:15:11.787740 134495874074432 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.281776s (batch_requests_ready=0.002283s, total_serialization_initiated=0.279386s, others=0.000107s)
I0424 09:15:11.787837 134495874074432 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.285856s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.285839)
I0424 09:15:11.788663 134495874074432 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/0.
I0424 09:15:11.788971 134369781405440 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 09:35:11.788936
I0424 09:15:11.799041 134495874074432 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 09:15:11.799290 134368137955072 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 09:15:11.799409 134495874074432 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777022111.181418, 'wait_for_prev_duration_secs': 6.008148193359375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777022111.1832025, 'checkpointer_blocking_duration_secs': 0.6058719158172607, 'get_old_steps_start_time': 1777022111.7890997, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1777022111.1795173, 'checkpoint_manager_blocking_duration_secs': 0.6198539733886719}
I0424 09:15:11.799512 134495874074432 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 09:15:11.799561 134495874074432 max_utils.py:750] 
Memstats: After params initialized:
I0424 09:15:11.799609 134495874074432 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0))
I0424 09:15:11.799642 134495874074432 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0))
I0424 09:15:11.799670 134495874074432 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0))
I0424 09:15:11.799694 134495874074432 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0))
I0424 09:15:12.113960 134495874074432 metric_logger.py:196] completed step: 0, seconds: 11.605, TFLOP/s/device: 1.171, Tokens/s/device: 176.471, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 09:15:12.291697 134495874074432 metric_logger.py:196] completed step: 1, seconds: 1.087, TFLOP/s/device: 12.503, Tokens/s/device: 1884.549, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 09:15:12.699068 134495874074432 metric_logger.py:196] completed step: 2, seconds: 0.030, TFLOP/s/device: 447.755, Tokens/s/device: 67490.526, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0424 09:15:12.846341 134495874074432 metric_logger.py:196] completed step: 3, seconds: 0.408, TFLOP/s/device: 33.313, Tokens/s/device: 5021.343, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0424 09:15:13.141757 134495874074432 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.936, Tokens/s/device: 13405.335, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0424 09:15:13.147676 134495874074432 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.182, Tokens/s/device: 13894.731, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0424 09:15:14.755407    2745 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 09:15:16.787822 134368186070784 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_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/0/items/array_metadatas/process_3
I0424 09:15:34.720522 134495874074432 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.931, Tokens/s/device: 6923.246, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0424 09:15:34.867888 134495874074432 metric_logger.py:196] completed step: 7, seconds: 21.426, TFLOP/s/device: 0.634, Tokens/s/device: 95.584, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0424 09:15:35.015164 134495874074432 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.225, Tokens/s/device: 13448.910, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0424 09:15:35.161767 134495874074432 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 09:15:35.162459 134495874074432 checkpointing.py:798] Waited 0.0007119178771972656 seconds for step 9 to finish before starting checkpointing.
I0424 09:15:35.164435 134495874074432 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 09:15:48.195439 134369781405440 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 43.053 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.689441204071045 s) (per-host)
I0424 09:15:48.195556 134369781405440 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.405964s.
I0424 09:15:56.736551 134369781405440 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 44.946935s.
I0424 09:15:56.736833 134368137955072 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 09:15:56.736964 134368137955072 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 09:15:56.737016 134368137955072 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 09:15:56.739216 134368137955072 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 09:15:56.739362 134495874074432 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0424 09:15:56.739454 134495874074432 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.575022 seconds. If this number is high, consider checkpointing less frequently.
I0424 09:15:56.741368 134495874074432 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 9
I0424 09:15:56.743476 134495874074432 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/9.
I0424 09:15:57.043457 134495874074432 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 09:15:57.043558 134495874074432 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 09:15:57.083745 134495874074432 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041357s
I0424 09:15:57.083907 134495874074432 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 34.369 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.044882774353027344 s) (per-host)
I0424 09:15:57.083970 134495874074432 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044954s (batch_requests_ready=0.001888s, total_serialization_initiated=0.042988s, others=0.000078s)
I0424 09:15:57.084061 134495874074432 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048925s (all_items=0.000035s, per_item={'items': '0.00003529'}, temp_paths=0.048890)
I0424 09:15:57.084780 134495874074432 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/9.
I0424 09:15:57.085080 134368137955072 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 09:35:57.085047
I0424 09:15:57.087651 134495874074432 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 09:15:57.087962 134368186070784 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 09:15:57.088125 134495874074432 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777022135.1644056, 'wait_for_prev_duration_secs': 21.575022220611572, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777022156.7414157, 'checkpointer_blocking_duration_secs': 0.3438072204589844, 'get_old_steps_start_time': 1777022157.0852468, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1777022135.1627042, 'checkpoint_manager_blocking_duration_secs': 21.92538571357727}
I0424 09:15:57.088235 134495874074432 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 09:15:57.088279 134495874074432 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 09:16:02.835616 134342474983168 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_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/9/items/array_metadatas/process_3
I0424 09:16:38.946375 134368137955072 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.693 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.90731120109558 s) (per-host)
I0424 09:16:38.946501 134368137955072 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.861306s.
I0424 09:16:48.061455 134368137955072 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 50.976246s.
I0424 09:16:48.061759 134368186070784 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 09:16:48.061880 134368186070784 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 09:16:48.061947 134368186070784 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 09:16:48.064251 134368186070784 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 09:16:48.064434 134495874074432 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 09:16:48.064587 134495874074432 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 09:16:48.065562 134495874074432 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.266, Tokens/s/device: 13907.375, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 24 09:17:00 UTC 2026
EXIT_CODE=0
XPK Start: Fri Apr 24 10:42:55 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 10:43:19.968780: 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 10:43:20.179847 140590941796160 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 10:43:29.222017 140590941796160 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 10:43:29.224335 140590941796160 distributed.py:172] Connecting to JAX distributed service on mt-01-base-q6ypl-slice-job-0-0.mt-01-base-q6ypl:8482
I0424 10:43:30.221644 140590941796160 max_utils.py:284] Jax distributed system initialized!
I0424 10:43:36.349261 140590941796160 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 10:43:36.349365 140590941796160 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 10:43:36.349414 140590941796160 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 10:43:36.349452 140590941796160 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 10:43:37.048753 140590941796160 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 10:43:37.162192 140590941796160 checkpointing.py:688] Setting up checkpoint logger...
I0424 10:43:37.162313 140590941796160 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 10:43:37.162359 140590941796160 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 10:43:37.162556 140590941796160 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 0x7fdd29845850>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 10:43:41.272790 140590941796160 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 10:43:41.273023 140590941796160 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 0x7fc8b04ae4b0>}, handler_registry=None
I0424 10:43:41.273277 140590941796160 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fc8b04ae4b0>` 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 10:43:41.273327 140590941796160 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fc970253fb0>` 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 10:43:41.273363 140590941796160 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 0x7fc8b04ae4b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fc8b04ae4b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fc970253fb0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fc970253fb0>}).
I0424 10:43:41.273694 140590941796160 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 10:43:41.273765 140590941796160 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7fc97028b7e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 10:43:41.958788 140590941796160 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints
I0424 10:43:42.003268 140590941796160 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=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/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7fc97814df10>
I0424 10:43:42.003415 140590941796160 checkpointing.py:302] Checkpoint manager created!
I0424 10:43:42.994120 140590941796160 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 10:43:42.994244 140590941796160 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32

I0424 10:43:44.803066 140590941796160 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 10:43:44.803180 140590941796160 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 10:43:44.808798 140590941796160 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 10:43:44.808855 140590941796160 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 10:43:44.825249 140590941796160 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 10:43:44.825308 140590941796160 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 10:43:44.840973 140590941796160 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 10:43:44.841034 140590941796160 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 10:43:44.866147 140590941796160 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 10:43:44.866216 140590941796160 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 10:43:44.882167 140590941796160 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 10:43:44.882229 140590941796160 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 10:43:44.898037 140590941796160 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 10:43:44.898107 140590941796160 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 10:43:44.928498 140590941796160 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 10:43:44.928581 140590941796160 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 10:43:44.948589 140590941796160 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 10:43:44.948655 140590941796160 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 10:43:56.322782 140590941796160 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 10:43:56.325266 140590941796160 metric_logger.py:301] number parameters: 1.104 billion
I0424 10:44:07.799040 140590941796160 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 10:44:07.951155 140590941796160 checkpointing.py:798] Waited 0.1520841121673584 seconds for step 0 to finish before starting checkpointing.
I0424 10:44:07.953865 140590941796160 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 10:44:07.955613 140590941796160 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0424 10:44:07.956992 140590941796160 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/0.
I0424 10:44:08.299124 140590941796160 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 10:44:08.300159 140590941796160 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0424 10:44:08.300215 140590941796160 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 10:44:08.579666 140590941796160 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.281321s
I0424 10:44:08.579849 140590941796160 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.310 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.29051804542541504 s) (per-host)
I0424 10:44:08.579919 140590941796160 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.290596s (batch_requests_ready=0.003163s, total_serialization_initiated=0.287346s, others=0.000087s)
I0424 10:44:08.580066 140590941796160 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.295337s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=0.295318)
I0424 10:44:08.580884 140590941796160 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.63 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/0.
I0424 10:44:08.581178 140465758451456 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 11:04:08.581144
I0424 10:44:08.585477 140590941796160 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 10:44:08.585707 140463071815424 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 10:44:08.585829 140590941796160 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777027447.9538465, 'wait_for_prev_duration_secs': 6.079673767089844e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777027447.955651, 'checkpointer_blocking_duration_secs': 0.6256909370422363, 'get_old_steps_start_time': 1777027448.5813673, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1777027447.9515662, 'checkpoint_manager_blocking_duration_secs': 0.634225606918335}
I0424 10:44:08.585938 140590941796160 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 10:44:08.586004 140590941796160 max_utils.py:750] 
Memstats: After params initialized:
I0424 10:44:08.586049 140590941796160 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0424 10:44:08.586078 140590941796160 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0424 10:44:08.586123 140590941796160 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0424 10:44:08.586153 140590941796160 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0424 10:44:08.903601 140590941796160 metric_logger.py:196] completed step: 0, seconds: 11.473, TFLOP/s/device: 1.184, Tokens/s/device: 178.511, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0424 10:44:09.087255 140590941796160 metric_logger.py:196] completed step: 1, seconds: 1.103, TFLOP/s/device: 12.321, Tokens/s/device: 1857.137, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0424 10:44:09.521375 140590941796160 metric_logger.py:196] completed step: 2, seconds: 0.037, TFLOP/s/device: 364.530, Tokens/s/device: 54945.939, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28632.133
I0424 10:44:09.668960 140590941796160 metric_logger.py:196] completed step: 3, seconds: 0.424, TFLOP/s/device: 32.060, Tokens/s/device: 4832.468, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16855.789
I0424 10:44:09.964209 140590941796160 metric_logger.py:196] completed step: 4, seconds: 0.165, TFLOP/s/device: 82.531, Tokens/s/device: 12440.017, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10661.594
I0424 10:44:09.972251 140590941796160 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.265, Tokens/s/device: 13907.186, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7245.978
I0424 10:44:11.715452    2763 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 10:44:13.734730 140463080208128 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/0/items/array_metadatas/process_6
I0424 10:44:31.956062 140590941796160 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.871, Tokens/s/device: 6914.247, total_weights: 65536, loss: 8.589, lm_loss: 8.589, perplexity: 5371.156
I0424 10:44:32.103615 140590941796160 metric_logger.py:196] completed step: 7, seconds: 21.838, TFLOP/s/device: 0.622, Tokens/s/device: 93.784, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4360.465
I0424 10:44:32.251006 140590941796160 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.572, Tokens/s/device: 13350.543, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.256
I0424 10:44:32.399228 140590941796160 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 10:44:32.399894 140590941796160 checkpointing.py:798] Waited 0.0006766319274902344 seconds for step 9 to finish before starting checkpointing.
I0424 10:44:32.401988 140590941796160 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 10:44:45.580511 140465758451456 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 42.358 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.29114294052124 s) (per-host)
I0424 10:44:45.580649 140465758451456 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.999350s.
I0424 10:44:51.710770 140465758451456 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 43.129448s.
I0424 10:44:51.711022 140463071815424 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 10:44:51.711166 140463071815424 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 10:44:51.711220 140463071815424 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 10:44:51.713424 140463071815424 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 10:44:51.713609 140590941796160 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0424 10:44:51.713745 140590941796160 checkpoint_manager.py:1452] Waiting for previous save to complete took 19.311759 seconds. If this number is high, consider checkpointing less frequently.
I0424 10:44:51.715607 140590941796160 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0424 10:44:51.717702 140590941796160 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/9.
I0424 10:44:52.358730 140590941796160 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0424 10:44:52.358826 140590941796160 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 10:44:52.394758 140590941796160 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037638s
I0424 10:44:52.394897 140590941796160 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 34.342 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.044918060302734375 s) (per-host)
I0424 10:44:52.394945 140590941796160 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044975s (batch_requests_ready=0.002955s, total_serialization_initiated=0.041957s, others=0.000063s)
I0424 10:44:52.395022 140590941796160 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049023s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.049006)
I0424 10:44:52.395681 140590941796160 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.68 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/9.
I0424 10:44:52.396018 140463071815424 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 11:04:52.395981
I0424 10:44:52.398054 140590941796160 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 10:44:52.398378 140437486552832 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0424 10:44:52.398528 140590941796160 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777027472.4019568, 'wait_for_prev_duration_secs': 19.311758518218994, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777027491.7156446, 'checkpointer_blocking_duration_secs': 0.6804900169372559, 'get_old_steps_start_time': 1777027492.396158, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1777027472.4001362, 'checkpoint_manager_blocking_duration_secs': 19.998356103897095}
I0424 10:44:52.398708 140590941796160 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 10:44:52.398756 140590941796160 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 10:44:58.038227 140463080208128 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_01_base/checkpoints/9/items/array_metadatas/process_6
I0424 10:45:34.334982 140463071815424 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.623 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.98497271537781 s) (per-host)
I0424 10:45:34.335088 140463071815424 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.938418s.
I0424 10:45:41.405643 140463071815424 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.008955s.
I0424 10:45:41.405955 140437486552832 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 10:45:41.406081 140437486552832 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0424 10:45:41.406151 140437486552832 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 10:45:41.408088 140437486552832 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 10:45:41.408272 140590941796160 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 10:45:41.408455 140590941796160 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 10:45:41.409478 140590941796160 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.762, Tokens/s/device: 13831.296, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3551.438
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 24 10:45:50 UTC 2026
EXIT_CODE=0