MaxView

‹ 06_grad_accumCase: 07_eval08_checkpoint_async_true_resume ›

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/s92.45091.854-0.596
Tok/s13935.013845.2-89.778
Avg s/step2.6902.865+0.175
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 09:27:48 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 09:28:13.055145: 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 09:28:13.264616 138400798861120 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 09:28:22.305208 138400798861120 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 09:28:22.307647 138400798861120 distributed.py:172] Connecting to JAX distributed service on mt-07-eval-1oil7-slice-job-0-0.mt-07-eval-1oil7:8482
I0424 09:28:23.777455 138400798861120 max_utils.py:284] Jax distributed system initialized!
I0424 09:28:29.896358 138400798861120 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 09:28:29.896465 138400798861120 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 09:28:29.896506 138400798861120 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 09:28:29.896543 138400798861120 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 09:28:30.595534 138400798861120 maxtext_utils.py:1732] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 09:28:30.596131 138400798861120 maxtext_utils.py:1732] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 09:28:30.596316 138400798861120 checkpointing.py:688] Setting up checkpoint logger...
I0424 09:28:30.596366 138400798861120 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 09:28:30.596409 138400798861120 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 09:28:30.596740 138400798861120 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 0x7ddf5b7171d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 09:28:33.607093 138400798861120 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 09:28:33.607325 138400798861120 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7dcb040f66f0>}, handler_registry=None
I0424 09:28:33.607567 138400798861120 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7dcb040f66f0>` 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 09:28:33.607617 138400798861120 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dcb040f9f70>` 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 09:28:33.607656 138400798861120 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 0x7dcb040f66f0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7dcb040f66f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dcb040f9f70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dcb040f9f70>}).
I0424 09:28:33.607994 138400798861120 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 09:28:33.608064 138400798861120 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7dcae4291bc0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 09:28:34.620416 138400798861120 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_07_eval/checkpoints
I0424 09:28:34.966947 138400798861120 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/linen_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7dcb040f7dd0>
I0424 09:28:34.967117 138400798861120 checkpointing.py:302] Checkpoint manager created!
I0424 09:28:35.899006 138400798861120 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 09:28:35.899105 138400798861120 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 09:28:36.278013 138400798861120 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 09:28:36.278101 138400798861120 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 09:28:36.294461 138400798861120 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 09:28:36.294523 138400798861120 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 09:28:36.318169 138400798861120 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 09:28:36.318235 138400798861120 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 09:28:36.334603 138400798861120 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 09:28:36.334661 138400798861120 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 09:28:36.351058 138400798861120 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 09:28:36.351123 138400798861120 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 09:28:36.376324 138400798861120 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0424 09:28:36.376393 138400798861120 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 09:28:36.397033 138400798861120 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 09:28:36.397094 138400798861120 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 09:28:36.614061 138400798861120 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 09:28:36.614171 138400798861120 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 09:28:38.033782 138400798861120 maxtext_utils.py:1835]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 09:28:38.033935 138400798861120 maxtext_utils.py:1835]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 09:28:38.034005 138400798861120 maxtext_utils.py:1835]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 09:28:38.034079 138400798861120 maxtext_utils.py:1835]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0424 09:28:38.034145 138400798861120 maxtext_utils.py:1835]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 09:28:38.034195 138400798861120 maxtext_utils.py:1835]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 09:28:38.034265 138400798861120 maxtext_utils.py:1835]  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 09:28:38.034324 138400798861120 maxtext_utils.py:1835]  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 09:28:38.034362 138400798861120 maxtext_utils.py:1835]  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 09:28:38.034397 138400798861120 maxtext_utils.py:1835]  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 09:28:38.034441 138400798861120 maxtext_utils.py:1835]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0424 09:28:38.034490 138400798861120 maxtext_utils.py:1835]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0424 09:28:38.525618 138400798861120 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 09:28:38.525714 138400798861120 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 09:28:38.541313 138400798861120 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 09:28:38.541383 138400798861120 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 09:28:49.448736 138400798861120 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 09:28:49.449532 138400798861120 metric_logger.py:301] number parameters: 1.104 billion
I0424 09:29:00.661892 138400798861120 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 09:29:00.949544 138400798861120 checkpointing.py:798] Waited 0.2876429557800293 seconds for step 0 to finish before starting checkpointing.
I0424 09:29:00.952240 138400798861120 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 09:29:00.954107 138400798861120 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0
I0424 09:29:00.955606 138400798861120 event_tracking.py:70] [process=4] [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_07_eval/checkpoints/0.
I0424 09:29:01.314000 138400798861120 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 09:29:01.314843 138400798861120 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 09:29:01.314913 138400798861120 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 09:29:01.589625 138400798861120 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.275696s
I0424 09:29:01.589800 138400798861120 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.486 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2812082767486572 s) (per-host)
I0424 09:29:01.589852 138400798861120 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.281271s (batch_requests_ready=0.002165s, total_serialization_initiated=0.279035s, others=0.000071s)
I0424 09:29:01.589963 138400798861120 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.285441s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.285423)
I0424 09:29:01.590779 138400798861120 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.64 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_07_eval/checkpoints/0.
I0424 09:29:01.591140 138272858777344 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 09:49:01.591103
I0424 09:29:01.609101 138400798861120 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 09:29:01.609438 138270767838976 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0424 09:29:01.609609 138400798861120 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_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777022940.952222, 'wait_for_prev_duration_secs': 6.532669067382812e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777022940.9541461, 'checkpointer_blocking_duration_secs': 0.6371538639068604, 'get_old_steps_start_time': 1777022941.5913267, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1777022940.9500766, 'checkpoint_manager_blocking_duration_secs': 0.6594834327697754}
I0424 09:29:01.609735 138400798861120 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 09:29:01.609796 138400798861120 max_utils.py:750] 
Memstats: After params initialized:
I0424 09:29:01.609856 138400798861120 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0))
I0424 09:29:01.609911 138400798861120 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0))
I0424 09:29:01.609942 138400798861120 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0))
I0424 09:29:01.609966 138400798861120 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0))
I0424 09:29:01.924781 138400798861120 metric_logger.py:196] completed step: 0, seconds: 11.212, TFLOP/s/device: 1.212, Tokens/s/device: 182.657, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 09:29:02.104967 138400798861120 metric_logger.py:196] completed step: 1, seconds: 1.261, TFLOP/s/device: 10.772, Tokens/s/device: 1623.736, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 09:29:02.546332 138400798861120 metric_logger.py:196] completed step: 2, seconds: 0.033, TFLOP/s/device: 410.611, Tokens/s/device: 61891.810, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0424 09:29:02.693911 138400798861120 metric_logger.py:196] completed step: 3, seconds: 0.441, TFLOP/s/device: 30.779, Tokens/s/device: 4639.310, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0424 09:29:02.989085 138400798861120 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.731, Tokens/s/device: 13374.432, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0424 09:29:02.995001 138400798861120 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.057, Tokens/s/device: 13875.809, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0424 09:29:05.357980    2803 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 09:29:07.541814 138271249745664 array_metadata_store.py:203] [process=4][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_07_eval/checkpoints/0/items/array_metadatas/process_4
I0424 09:29:24.722748 138400798861120 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.971, Tokens/s/device: 6929.219, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0424 09:29:24.869746 138400798861120 metric_logger.py:196] completed step: 7, seconds: 21.580, TFLOP/s/device: 0.630, Tokens/s/device: 94.902, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0424 09:29:25.017195 138400798861120 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.693, Tokens/s/device: 13368.757, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0424 09:29:25.163735 138400798861120 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 09:29:25.164405 138400798861120 checkpointing.py:798] Waited 0.0006895065307617188 seconds for step 9 to finish before starting checkpointing.
I0424 09:29:25.166470 138400798861120 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 09:29:38.649037 138272858777344 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 42.303 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.34040856361389 s) (per-host)
I0424 09:29:38.649154 138272858777344 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.057901s.
I0424 09:29:46.978422 138272858777344 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 45.387151s.
I0424 09:29:46.978714 138270767838976 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 09:29:46.978841 138270767838976 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0424 09:29:46.978924 138270767838976 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 09:29:46.980664 138270767838976 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 09:29:46.980840 138400798861120 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0424 09:29:46.980985 138400798861120 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.814515 seconds. If this number is high, consider checkpointing less frequently.
I0424 09:29:46.982861 138400798861120 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9
I0424 09:29:46.984767 138400798861120 event_tracking.py:70] [process=4] [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_07_eval/checkpoints/9.
I0424 09:29:47.294425 138400798861120 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 09:29:47.294517 138400798861120 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 09:29:47.331036 138400798861120 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037706s
I0424 09:29:47.331215 138400798861120 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 37.529 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04110312461853027 s) (per-host)
I0424 09:29:47.331270 138400798861120 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.041167s (batch_requests_ready=0.001758s, total_serialization_initiated=0.039334s, others=0.000074s)
I0424 09:29:47.331370 138400798861120 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.045183s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.045168)
I0424 09:29:47.332119 138400798861120 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.35 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_07_eval/checkpoints/9.
I0424 09:29:47.332478 138270767838976 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 09:49:47.332443
I0424 09:29:47.334434 138400798861120 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 09:29:47.334674 138250349958912 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0424 09:29:47.334831 138400798861120 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_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777022965.1664395, 'wait_for_prev_duration_secs': 21.814514636993408, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777022986.982914, 'checkpointer_blocking_duration_secs': 0.34970855712890625, 'get_old_steps_start_time': 1777022987.3326468, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1777022965.164631, 'checkpoint_manager_blocking_duration_secs': 22.170166969299316}
I0424 09:29:47.334947 138400798861120 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 09:29:47.334991 138400798861120 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 09:29:52.746397 138271249745664 array_metadata_store.py:203] [process=4][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_07_eval/checkpoints/9/items/array_metadatas/process_4
I0424 09:30:28.903041 138270767838976 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.959 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.612890005111694 s) (per-host)
I0424 09:30:28.903170 138270767838976 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.570579s.
I0424 09:30:36.998041 138270767838976 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 49.665434s.
I0424 09:30:36.998325 138250349958912 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 09:30:36.998443 138250349958912 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0424 09:30:36.998492 138250349958912 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 09:30:37.000899 138250349958912 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 09:30:37.001069 138400798861120 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 09:30:37.001232 138400798861120 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 09:30:37.002179 138400798861120 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.450, Tokens/s/device: 13935.006, 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 09:30:48 UTC 2026
EXIT_CODE=0
XPK Start: Fri Apr 24 10:54:30 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 10:54:55.086713: 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 10:54:55.300499 135602131814208 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 10:55:04.341750 135602131814208 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 10:55:04.344215 135602131814208 distributed.py:172] Connecting to JAX distributed service on mt-07-eval-gq16l-slice-job-0-0.mt-07-eval-gq16l:8482
I0424 10:55:05.464244 135602131814208 max_utils.py:284] Jax distributed system initialized!
I0424 10:55:11.639654 135602131814208 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 10:55:11.639764 135602131814208 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 10:55:11.639806 135602131814208 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 10:55:11.639842 135602131814208 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 10:55:12.356728 135602131814208 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 10:55:12.471829 135602131814208 checkpointing.py:688] Setting up checkpoint logger...
I0424 10:55:12.471958 135602131814208 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 10:55:12.472004 135602131814208 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 10:55:12.472209 135602131814208 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 0x7b53c8718110>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 10:55:15.345908 135602131814208 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 10:55:15.346166 135602131814208 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 0x7b3f4c0f76b0>}, handler_registry=None
I0424 10:55:15.346404 135602131814208 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b3f4c0f76b0>` 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 10:55:15.346454 135602131814208 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b3f34392960>` 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 10:55:15.346494 135602131814208 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 0x7b3f4c0f76b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b3f4c0f76b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b3f34392960>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b3f34392960>}).
I0424 10:55:15.346818 135602131814208 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 10:55:15.346891 135602131814208 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b3f343ef7e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 10:55:17.109344 135602131814208 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_07_eval/checkpoints
I0424 10:55:17.111644 135602131814208 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/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260424_091312_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b3f4c1ea1b0>
I0424 10:55:17.111757 135602131814208 checkpointing.py:302] Checkpoint manager created!
I0424 10:55:18.082682 135602131814208 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 10:55:18.082809 135602131814208 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32

I0424 10:55:19.882992 135602131814208 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 10:55:19.883104 135602131814208 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 10:55:19.888736 135602131814208 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 10:55:19.888790 135602131814208 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 10:55:19.905476 135602131814208 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 10:55:19.905538 135602131814208 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 10:55:19.921644 135602131814208 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0424 10:55:19.921706 135602131814208 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 10:55:19.947093 135602131814208 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 10:55:19.947159 135602131814208 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 10:55:19.963211 135602131814208 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 10:55:19.963270 135602131814208 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 10:55:19.979376 135602131814208 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 10:55:19.979434 135602131814208 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 10:55:20.010219 135602131814208 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0424 10:55:20.010296 135602131814208 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 10:55:20.030442 135602131814208 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 10:55:20.030510 135602131814208 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 10:55:31.441043 135602131814208 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 10:55:31.443426 135602131814208 metric_logger.py:301] number parameters: 1.104 billion
I0424 10:55:42.572703 135602131814208 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 10:55:42.725373 135602131814208 checkpointing.py:798] Waited 0.15266180038452148 seconds for step 0 to finish before starting checkpointing.
I0424 10:55:42.727787 135602131814208 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 10:55:42.729367 135602131814208 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0
I0424 10:55:42.730759 135602131814208 event_tracking.py:70] [process=3] [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_07_eval/checkpoints/0.
I0424 10:55:43.075584 135602131814208 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 10:55:43.076526 135602131814208 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0424 10:55:43.076585 135602131814208 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 10:55:43.355140 135602131814208 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.280331s
I0424 10:55:43.355316 135602131814208 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 5.326 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2896237373352051 s) (per-host)
I0424 10:55:43.355370 135602131814208 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.289689s (batch_requests_ready=0.003222s, total_serialization_initiated=0.286394s, others=0.000073s)
I0424 10:55:43.355477 135602131814208 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.294056s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.294038)
I0424 10:55:43.356287 135602131814208 event_tracking.py:125] [process=3] [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_07_eval/checkpoints/0.
I0424 10:55:43.356606 135473066407680 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 11:15:43.356571
I0424 10:55:43.359634 135602131814208 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 10:55:43.359917 135470382053120 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 10:55:43.360087 135602131814208 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_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777028142.7277696, 'wait_for_prev_duration_secs': 6.29425048828125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777028142.729408, 'checkpointer_blocking_duration_secs': 0.6273565292358398, 'get_old_steps_start_time': 1777028143.3567908, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1777028142.725756, 'checkpoint_manager_blocking_duration_secs': 0.6342940330505371}
I0424 10:55:43.360198 135602131814208 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 10:55:43.360269 135602131814208 max_utils.py:750] 
Memstats: After params initialized:
I0424 10:55:43.360316 135602131814208 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0))
I0424 10:55:43.360347 135602131814208 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0))
I0424 10:55:43.360374 135602131814208 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0))
I0424 10:55:43.360399 135602131814208 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0))
I0424 10:55:43.682891 135602131814208 metric_logger.py:196] completed step: 0, seconds: 11.128, TFLOP/s/device: 1.221, Tokens/s/device: 184.037, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0424 10:55:43.854526 135602131814208 metric_logger.py:196] completed step: 1, seconds: 1.108, TFLOP/s/device: 12.258, Tokens/s/device: 1847.708, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0424 10:55:44.291531 135602131814208 metric_logger.py:196] completed step: 2, seconds: 0.025, TFLOP/s/device: 538.339, Tokens/s/device: 81144.261, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28632.133
I0424 10:55:44.439167 135602131814208 metric_logger.py:196] completed step: 3, seconds: 0.427, TFLOP/s/device: 31.811, Tokens/s/device: 4794.883, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16855.789
I0424 10:55:44.734729 135602131814208 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.693, Tokens/s/device: 12464.396, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10661.594
I0424 10:55:44.742439 135602131814208 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.279, Tokens/s/device: 13909.264, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7245.978
I0424 10:55:46.591172    2804 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 10:55:48.661007 135471402137344 array_metadata_store.py:203] [process=3][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_07_eval/checkpoints/0/items/array_metadatas/process_3
I0424 10:56:08.209130 135602131814208 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.874, Tokens/s/device: 6914.644, total_weights: 65536, loss: 8.589, lm_loss: 8.589, perplexity: 5371.156
I0424 10:56:08.356704 135602131814208 metric_logger.py:196] completed step: 7, seconds: 23.321, TFLOP/s/device: 0.583, Tokens/s/device: 87.819, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4360.465
I0424 10:56:08.504199 135602131814208 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.569, Tokens/s/device: 13350.021, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.256
I0424 10:56:08.652336 135602131814208 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 10:56:08.653022 135602131814208 checkpointing.py:798] Waited 0.0006973743438720703 seconds for step 9 to finish before starting checkpointing.
I0424 10:56:08.655047 135602131814208 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 10:56:20.224800 135473066407680 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 42.509 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.1590621471405 s) (per-host)
I0424 10:56:20.224911 135473066407680 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.868196s.
I0424 10:56:30.497262 135473066407680 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 47.140530s.
I0424 10:56:30.497552 135470382053120 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 10:56:30.497708 135470382053120 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 10:56:30.497764 135470382053120 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 10:56:30.500319 135470382053120 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 10:56:30.500500 135602131814208 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 10:56:30.500653 135602131814208 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.845606 seconds. If this number is high, consider checkpointing less frequently.
I0424 10:56:30.502516 135602131814208 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 9
I0424 10:56:30.504587 135602131814208 event_tracking.py:70] [process=3] [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_07_eval/checkpoints/9.
I0424 10:56:30.833127 135602131814208 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0424 10:56:30.833224 135602131814208 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 10:56:30.868024 135602131814208 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036537s
I0424 10:56:30.868191 135602131814208 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 35.111 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04393410682678223 s) (per-host)
I0424 10:56:30.868241 135602131814208 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043993s (batch_requests_ready=0.003005s, total_serialization_initiated=0.040922s, others=0.000066s)
I0424 10:56:30.868339 135602131814208 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048351s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.048336)
I0424 10:56:30.869050 135602131814208 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.37 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_07_eval/checkpoints/9.
I0424 10:56:30.869326 135470382053120 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 11:16:30.869296
I0424 10:56:30.871293 135602131814208 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 10:56:30.871582 135450610050816 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 10:56:30.871716 135602131814208 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_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777028168.6550171, 'wait_for_prev_duration_secs': 21.845606088638306, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777028190.5025542, 'checkpointer_blocking_duration_secs': 0.36690855026245117, 'get_old_steps_start_time': 1777028190.8694887, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1777028168.653249, 'checkpoint_manager_blocking_duration_secs': 22.218433380126953}
I0424 10:56:30.871887 135602131814208 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 10:56:30.871947 135602131814208 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 10:56:36.855575 135471402137344 array_metadata_store.py:203] [process=3][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_07_eval/checkpoints/9/items/array_metadatas/process_3
I0424 10:57:13.770711 135470382053120 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 36.781 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.946414947509766 s) (per-host)
I0424 10:57:13.770838 135470382053120 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.901403s.
I0424 10:57:22.403559 135470382053120 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 51.534107s.
I0424 10:57:22.403839 135450610050816 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 10:57:22.403993 135450610050816 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 10:57:22.404042 135450610050816 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 10:57:22.405392 135450610050816 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 10:57:22.405577 135602131814208 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 10:57:22.405750 135602131814208 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 10:57:22.406792 135602131814208 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.854, Tokens/s/device: 13845.228, 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 10:57:35 UTC 2026
EXIT_CODE=0