MaxView

‹ 06_grad_accumCase: 07_eval08_checkpoint_async_true_resume ›

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

MetricLinen  1abe20691NNX  1abe20691Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s92.32191.833-0.488
Tok/s13915.713842.0-73.643
Avg s/step2.7042.831+0.127
Memory %1.381.380
JAX0.9.20.9.2

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

XPK Start: Thu Apr 23 09:56:02 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 09:56:27.793875: 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 09:56:28.003720 132567259891520 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 09:56:37.046568 132567259891520 distributed.py:149] Starting JAX distributed service on [::]:8482
I0423 09:56:37.048929 132567259891520 distributed.py:172] Connecting to JAX distributed service on mt-07-eval-m5ghz-slice-job-0-0.mt-07-eval-m5ghz:8482
I0423 09:56:38.171184 132567259891520 max_utils.py:284] Jax distributed system initialized!
I0423 09:56:44.429900 132567259891520 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 09:56:44.430008 132567259891520 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 09:56:44.430047 132567259891520 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 09:56:44.430083 132567259891520 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 09:56:45.146083 132567259891520 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 09:56:45.146697 132567259891520 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 09:56:45.146877 132567259891520 checkpointing.py:688] Setting up checkpoint logger...
I0423 09:56:45.146928 132567259891520 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 09:56:45.146970 132567259891520 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0423 09:56:45.147327 132567259891520 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 0x789117714c50>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 09:56:48.062833 132567259891520 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0423 09:56:48.063063 132567259891520 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 0x787ccc4be480>}, handler_registry=None
I0423 09:56:48.063322 132567259891520 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x787ccc4be480>` 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 09:56:48.063372 132567259891520 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x787ccc31c200>` 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 09:56:48.063408 132567259891520 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 0x787ccc4be480>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x787ccc4be480>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x787ccc31c200>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x787ccc31c200>}).
I0423 09:56:48.063737 132567259891520 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0423 09:56:48.063806 132567259891520 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x787cb46cde40> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0423 09:56:49.216517 132567259891520 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints
I0423 09:56:49.629777 132567259891520 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/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x787ccc31ad20>
I0423 09:56:49.629955 132567259891520 checkpointing.py:302] Checkpoint manager created!
I0423 09:56:50.587338 132567259891520 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 09:56:50.587444 132567259891520 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 09:56:50.967272 132567259891520 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 09:56:50.967367 132567259891520 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 09:56:50.983977 132567259891520 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 09:56:50.984039 132567259891520 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 09:56:51.008278 132567259891520 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 09:56:51.008344 132567259891520 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 09:56:51.024932 132567259891520 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 09:56:51.024991 132567259891520 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 09:56:51.041557 132567259891520 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 09:56:51.041618 132567259891520 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 09:56:51.066543 132567259891520 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 09:56:51.066615 132567259891520 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 09:56:51.087526 132567259891520 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 09:56:51.087590 132567259891520 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 09:56:51.312278 132567259891520 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 09:56:51.312390 132567259891520 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0423 09:56:52.746107 132567259891520 maxtext_utils.py:1874]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0423 09:56:52.746235 132567259891520 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)
I0423 09:56:52.746293 132567259891520 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)
I0423 09:56:52.746350 132567259891520 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')
I0423 09:56:52.746407 132567259891520 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)
I0423 09:56:52.746448 132567259891520 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)
I0423 09:56:52.746500 132567259891520 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)
I0423 09:56:52.746551 132567259891520 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')
I0423 09:56:52.746591 132567259891520 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)
I0423 09:56:52.746627 132567259891520 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)
I0423 09:56:52.746675 132567259891520 maxtext_utils.py:1874]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]

    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0423 09:56:52.746725 132567259891520 maxtext_utils.py:1874]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0423 09:56:53.243504 132567259891520 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 09:56:53.243604 132567259891520 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 09:56:53.259251 132567259891520 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 09:56:53.259312 132567259891520 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 09:57:04.170618 132567259891520 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.
I0423 09:57:04.171516 132567259891520 metric_logger.py:301] number parameters: 1.104 billion
I0423 09:57:15.378968 132567259891520 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0423 09:57:15.533647 132567259891520 checkpointing.py:798] Waited 0.15465593338012695 seconds for step 0 to finish before starting checkpointing.
I0423 09:57:15.536245 132567259891520 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 09:57:15.537847 132567259891520 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0423 09:57:15.539399 132567259891520 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/0.
I0423 09:57:15.903788 132567259891520 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 09:57:15.904892 132567259891520 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 09:57:15.904950 132567259891520 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 09:57:16.179876 132567259891520 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.276162s
I0423 09:57:16.180048 132567259891520 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.474 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2818169593811035 s) (per-host)
I0423 09:57:16.180114 132567259891520 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.281882s (batch_requests_ready=0.002200s, total_serialization_initiated=0.279609s, others=0.000072s)
I0423 09:57:16.180206 132567259891520 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.286192s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=0.286172)
I0423 09:57:16.181009 132567259891520 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.64 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/0.
I0423 09:57:16.181347 132438509561600 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 10:17:16.181310
I0423 09:57:16.193655 132567259891520 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 09:57:16.193903 132436885219072 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 09:57:16.194032 132567259891520 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776938235.5362256, 'wait_for_prev_duration_secs': 6.29425048828125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776938235.5378861, 'checkpointer_blocking_duration_secs': 0.6436097621917725, 'get_old_steps_start_time': 1776938236.1815221, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776938235.5341995, 'checkpoint_manager_blocking_duration_secs': 0.6597943305969238}
I0423 09:57:16.194145 132567259891520 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0423 09:57:16.194193 132567259891520 max_utils.py:750] 
Memstats: After params initialized:
I0423 09:57:16.194241 132567259891520 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0423 09:57:16.194273 132567259891520 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0423 09:57:16.194301 132567259891520 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0423 09:57:16.194327 132567259891520 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0423 09:57:16.503793 132567259891520 metric_logger.py:196] completed step: 0, seconds: 11.207, TFLOP/s/device: 1.212, Tokens/s/device: 182.737, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0423 09:57:16.678889 132567259891520 metric_logger.py:196] completed step: 1, seconds: 1.123, TFLOP/s/device: 12.095, Tokens/s/device: 1823.128, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0423 09:57:17.096744 132567259891520 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 486.523, Tokens/s/device: 73334.049, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0423 09:57:17.244142 132567259891520 metric_logger.py:196] completed step: 3, seconds: 0.418, TFLOP/s/device: 32.501, Tokens/s/device: 4898.912, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0423 09:57:17.538947 132567259891520 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
I0423 09:57:17.544770 132567259891520 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.208, Tokens/s/device: 13898.597, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0423 09:57:20.181627    2845 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 09:57:22.212970 132437405996800 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/0/items/array_metadatas/process_6
I0423 09:57:39.561756 132567259891520 metric_logger.py:196] completed step: 6, seconds: 0.295, TFLOP/s/device: 46.001, Tokens/s/device: 6933.747, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0423 09:57:39.708902 132567259891520 metric_logger.py:196] completed step: 7, seconds: 21.870, TFLOP/s/device: 0.621, Tokens/s/device: 93.645, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0423 09:57:39.856243 132567259891520 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.172, Tokens/s/device: 13440.966, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0423 09:57:40.002888 132567259891520 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0423 09:57:40.003574 132567259891520 checkpointing.py:798] Waited 0.0007073879241943359 seconds for step 9 to finish before starting checkpointing.
I0423 09:57:40.005643 132567259891520 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 09:57:54.416140 132438509561600 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 41.009 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.517842054367065 s) (per-host)
I0423 09:57:54.416265 132438509561600 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.234808s.
I0423 09:58:02.992750 132438509561600 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 46.811276s.
I0423 09:58:02.993042 132436885219072 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 09:58:02.993180 132436885219072 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 09:58:02.993234 132436885219072 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 09:58:02.995810 132436885219072 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 09:58:02.996005 132567259891520 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.
W0423 09:58:02.996152 132567259891520 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.990509 seconds. If this number is high, consider checkpointing less frequently.
I0423 09:58:02.998076 132567259891520 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0423 09:58:03.000091 132567259891520 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/9.
I0423 09:58:03.697371 132567259891520 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0423 09:58:03.697463 132567259891520 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 09:58:03.739657 132567259891520 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.043483s
I0423 09:58:03.739822 132567259891520 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 32.875 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04692196846008301 s) (per-host)
I0423 09:58:03.739871 132567259891520 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046979s (batch_requests_ready=0.001784s, total_serialization_initiated=0.045131s, others=0.000064s)
I0423 09:58:03.739956 132567259891520 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.051793s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.051777)
I0423 09:58:03.740606 132567259891520 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/9.
I0423 09:58:03.740906 132436885219072 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 10:18:03.740872
I0423 09:58:03.742903 132567259891520 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 09:58:03.743196 132416760919808 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 09:58:03.743360 132567259891520 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776938260.0056129, 'wait_for_prev_duration_secs': 22.990508556365967, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776938282.998128, 'checkpointer_blocking_duration_secs': 0.7429113388061523, 'get_old_steps_start_time': 1776938283.7410605, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1776938260.0038023, 'checkpoint_manager_blocking_duration_secs': 23.739522457122803}
I0423 09:58:03.743469 132567259891520 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0423 09:58:03.743513 132567259891520 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 09:58:09.292570 132437405996800 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/linen_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/9/items/array_metadatas/process_6
I0423 09:58:46.603816 132436885219072 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.811 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.91087555885315 s) (per-host)
I0423 09:58:46.603944 132436885219072 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.862930s.
I0423 09:58:55.044474 132436885219072 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 51.303446s.
I0423 09:58:55.044752 132416760919808 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 09:58:55.044873 132416760919808 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 09:58:55.044919 132416760919808 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 09:58:55.046063 132416760919808 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 09:58:55.046245 132567259891520 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.
I0423 09:58:55.046391 132567259891520 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 09:58:55.047388 132567259891520 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.321, Tokens/s/device: 13915.690, 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: Thu Apr 23 09:59:07 UTC 2026
EXIT_CODE=0
XPK Start: Thu Apr 23 11:26:52 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 11:27:17.486634: 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 11:27:17.697203 132804084930368 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 11:27:26.739409 132804084930368 distributed.py:149] Starting JAX distributed service on [::]:8482
I0423 11:27:26.741745 132804084930368 distributed.py:172] Connecting to JAX distributed service on mt-07-eval-gjbdn-slice-job-0-0.mt-07-eval-gjbdn:8482
I0423 11:27:27.646245 132804084930368 max_utils.py:284] Jax distributed system initialized!
I0423 11:27:33.889496 132804084930368 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 11:27:33.889605 132804084930368 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 11:27:33.889645 132804084930368 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 11:27:33.889682 132804084930368 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 11:27:34.587753 132804084930368 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 11:27:34.701056 132804084930368 checkpointing.py:688] Setting up checkpoint logger...
I0423 11:27:34.701186 132804084930368 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 11:27:34.701230 132804084930368 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0423 11:27:34.701421 132804084930368 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 0x78c824b277d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 11:27:37.529969 132804084930368 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0423 11:27:37.530225 132804084930368 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 0x78b41432d790>}, handler_registry=None
I0423 11:27:37.530466 132804084930368 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78b41432d790>` 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 11:27:37.530520 132804084930368 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b3ec46fe30>` 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 11:27:37.530555 132804084930368 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 0x78b41432d790>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78b41432d790>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b3ec46fe30>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78b3ec46fe30>}).
I0423 11:27:37.530879 132804084930368 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0423 11:27:37.530951 132804084930368 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x78b3ec4277e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0423 11:27:38.694375 132804084930368 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints
I0423 11:27:38.814414 132804084930368 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_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x78b414327e90>
I0423 11:27:38.814599 132804084930368 checkpointing.py:302] Checkpoint manager created!
I0423 11:27:39.781468 132804084930368 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 11:27:39.781592 132804084930368 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32

I0423 11:27:41.553405 132804084930368 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 11:27:41.553514 132804084930368 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 11:27:41.559163 132804084930368 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 11:27:41.559220 132804084930368 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 11:27:41.575875 132804084930368 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 11:27:41.575937 132804084930368 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 11:27:41.591830 132804084930368 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 11:27:41.591891 132804084930368 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 11:27:41.616956 132804084930368 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 11:27:41.617022 132804084930368 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 11:27:41.633036 132804084930368 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 11:27:41.633108 132804084930368 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 11:27:41.648965 132804084930368 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 11:27:41.649026 132804084930368 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 11:27:41.679142 132804084930368 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 11:27:41.679215 132804084930368 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 11:27:41.699237 132804084930368 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 11:27:41.699309 132804084930368 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 11:27:52.578177 132804084930368 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.
I0423 11:27:52.580513 132804084930368 metric_logger.py:301] number parameters: 1.104 billion
I0423 11:28:03.619909 132804084930368 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0423 11:28:04.119283 132804084930368 checkpointing.py:798] Waited 0.49934816360473633 seconds for step 0 to finish before starting checkpointing.
I0423 11:28:04.121456 132804084930368 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 11:28:04.123231 132804084930368 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0423 11:28:04.124658 132804084930368 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/0.
I0423 11:28:04.852519 132804084930368 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 11:28:04.853540 132804084930368 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0423 11:28:04.853604 132804084930368 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 11:28:05.131728 132804084930368 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279989s
I0423 11:28:05.131899 132804084930368 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.337 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.289031982421875 s) (per-host)
I0423 11:28:05.131955 132804084930368 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.289098s (batch_requests_ready=0.002996s, total_serialization_initiated=0.286027s, others=0.000074s)
I0423 11:28:05.132052 132804084930368 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.293335s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.293318)
I0423 11:28:05.132808 132804084930368 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.01 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/0.
I0423 11:28:05.133078 132676381964032 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 11:48:05.133053
I0423 11:28:05.143973 132804084930368 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 11:28:05.144233 132674828105472 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 11:28:05.144359 132804084930368 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776943684.121437, 'wait_for_prev_duration_secs': 6.341934204101562e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776943684.1232743, 'checkpointer_blocking_duration_secs': 1.0099289417266846, 'get_old_steps_start_time': 1776943685.1332273, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776943684.1196716, 'checkpoint_manager_blocking_duration_secs': 1.024646282196045}
I0423 11:28:05.144473 132804084930368 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0423 11:28:05.144541 132804084930368 max_utils.py:750] 
Memstats: After params initialized:
I0423 11:28:05.144591 132804084930368 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0423 11:28:05.144623 132804084930368 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0423 11:28:05.144650 132804084930368 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0423 11:28:05.144674 132804084930368 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0423 11:28:05.458692 132804084930368 metric_logger.py:196] completed step: 0, seconds: 11.038, TFLOP/s/device: 1.231, Tokens/s/device: 185.536, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0423 11:28:05.640048 132804084930368 metric_logger.py:196] completed step: 1, seconds: 1.837, TFLOP/s/device: 7.396, Tokens/s/device: 1114.828, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0423 11:28:06.073942 132804084930368 metric_logger.py:196] completed step: 2, seconds: 0.035, TFLOP/s/device: 388.948, Tokens/s/device: 58626.514, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28632.133
I0423 11:28:06.221339 132804084930368 metric_logger.py:196] completed step: 3, seconds: 0.424, TFLOP/s/device: 32.057, Tokens/s/device: 4831.978, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16855.789
I0423 11:28:06.516878 132804084930368 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.822, Tokens/s/device: 12483.847, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10661.594
I0423 11:28:06.524064 132804084930368 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.365, Tokens/s/device: 13922.312, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7245.978
I0423 11:28:08.212733    2793 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 11:28:10.690824 132675885598464 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_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/0/items/array_metadatas/process_6
I0423 11:28:28.948151 132804084930368 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.893, Tokens/s/device: 6917.517, total_weights: 65536, loss: 8.589, lm_loss: 8.589, perplexity: 5371.156
I0423 11:28:29.095849 132804084930368 metric_logger.py:196] completed step: 7, seconds: 22.278, TFLOP/s/device: 0.610, Tokens/s/device: 91.930, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4360.465
I0423 11:28:29.243337 132804084930368 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.737, Tokens/s/device: 13375.393, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.256
I0423 11:28:29.391641 132804084930368 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0423 11:28:29.392391 132804084930368 checkpointing.py:798] Waited 0.0007581710815429688 seconds for step 9 to finish before starting checkpointing.
I0423 11:28:29.394310 132804084930368 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 11:28:41.410246 132676381964032 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 43.197 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.567328453063965 s) (per-host)
I0423 11:28:41.410381 132676381964032 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.277210s.
I0423 11:28:48.710339 132676381964032 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 43.577140s.
I0423 11:28:48.710581 132674828105472 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 11:28:48.710640 132674828105472 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 11:28:48.710695 132674828105472 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 11:28:48.712728 132674828105472 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 11:28:48.712892 132804084930368 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.
W0423 11:28:48.713026 132804084930368 checkpoint_manager.py:1452] Waiting for previous save to complete took 19.318718 seconds. If this number is high, consider checkpointing less frequently.
I0423 11:28:48.714983 132804084930368 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0423 11:28:48.717050 132804084930368 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/9.
I0423 11:28:49.195151 132804084930368 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0423 11:28:49.195244 132804084930368 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 11:28:49.236972 132804084930368 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.043531s
I0423 11:28:49.237153 132804084930368 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 30.461 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.050641775131225586 s) (per-host)
I0423 11:28:49.237204 132804084930368 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.050702s (batch_requests_ready=0.002724s, total_serialization_initiated=0.047910s, others=0.000068s)
I0423 11:28:49.237303 132804084930368 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.054766s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.054752)
I0423 11:28:49.238041 132804084930368 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.52 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/9.
I0423 11:28:49.238346 132674828105472 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 11:48:49.238325
I0423 11:28:49.240260 132804084930368 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 11:28:49.240559 132675885598464 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 11:28:49.240723 132804084930368 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776943709.3942795, 'wait_for_prev_duration_secs': 19.31871771812439, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776943728.7150257, 'checkpointer_blocking_duration_secs': 0.523411750793457, 'get_old_steps_start_time': 1776943729.2384617, 'get_old_steps_duration_secs': 3.600120544433594e-05, 'checkpoint_manager_blocking_start_time': 1776943709.3926682, 'checkpoint_manager_blocking_duration_secs': 19.84801697731018}
I0423 11:28:49.240835 132804084930368 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0423 11:28:49.240882 132804084930368 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 11:28:54.562017 132653143549696 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_20260423_093806/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260423_093806_07_eval/checkpoints/9/items/array_metadatas/process_6
I0423 11:29:30.851034 132674828105472 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.912 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.6644823551178 s) (per-host)
I0423 11:29:30.851186 132674828105472 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.612770s.
I0423 11:29:37.518676 132674828105472 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 48.280246s.
I0423 11:29:37.518974 132675885598464 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 11:29:37.519091 132675885598464 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 11:29:37.519172 132675885598464 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0423 11:29:37.523395 132675885598464 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0423 11:29:37.523590 132804084930368 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.
I0423 11:29:37.523761 132804084930368 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 11:29:37.524718 132804084930368 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.833, Tokens/s/device: 13842.047, 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: Thu Apr 23 11:29:47 UTC 2026
EXIT_CODE=0