MaxView

‹ 03_dropoutCase: 04_int805_fp8 ›

Metrics: Linen vs NNX  ·  feat/nnx-post-train-fixes

MetricLinen  574ad3fb9NNX  03aaf29baDiff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17508.1740-0.001
TFLOP/s59.05659.158+0.102
Tok/s8901.58916.9+15.347
Avg s/step3.6743.470-0.204
Memory %1.381.380
JAX0.9.20.9.2

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

Linen  ·  574ad3fb9  ·  feat_nnx_post_train_fixes_20260420_153543  ·  full log
XPK Start: Mon Apr 20 15:48:43 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-20 15:49:07.415468: 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)
I0420 15:49:07.593057 131960965498688 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 15:49:16.632573 131960965498688 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 15:49:16.635345 131960965498688 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-2ugmo-slice-job-0-0.mt-04-int8-2ugmo:8482
I0420 15:49:18.068390 131960965498688 max_utils.py:284] Jax distributed system initialized!
I0420 15:49:23.173704 131960965498688 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 15:49:23.173812 131960965498688 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 15:49:23.173851 131960965498688 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
I0420 15:49:23.173885 131960965498688 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 15:49:23.868773 131960965498688 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 15:49:23.869361 131960965498688 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 15:49:23.869665 131960965498688 checkpointing.py:688] Setting up checkpoint logger...
I0420 15:49:23.869725 131960965498688 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 15:49:23.869767 131960965498688 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 15:49:23.870091 131960965498688 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 0x7803d6a4c170>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 15:49:26.892600 131960965498688 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 15:49:26.892836 131960965498688 checkpoint_manager.py:708] [process=2][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x77f048204860>}, handler_registry=None
I0420 15:49:26.893076 131960965498688 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x77f048204860>` 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`.
I0420 15:49:26.893124 131960965498688 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x77f048206600>` 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`.
I0420 15:49:26.893159 131960965498688 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 0x77f048204860>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x77f048204860>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x77f048206600>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x77f048206600>}).
I0420 15:49:26.893477 131960965498688 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 15:49:26.893560 131960965498688 async_checkpointer.py:192] [process=2][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x77ef184ff740> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 15:49:28.498111 131960965498688 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints
I0420 15:49:28.519546 131960965498688 checkpoint_manager.py:929] [process=2][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_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x77f0482049e0>
I0420 15:49:28.519683 131960965498688 checkpointing.py:302] Checkpoint manager created!
I0420 15:49:29.680187 131960965498688 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 15:49:29.680296 131960965498688 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 15:49:30.070130 131960965498688 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 15:49:30.070221 131960965498688 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 15:49:30.086576 131960965498688 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 15:49:30.086634 131960965498688 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 15:49:30.151783 131960965498688 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 15:49:30.151867 131960965498688 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 15:49:30.168229 131960965498688 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 15:49:30.168292 131960965498688 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 15:49:30.184450 131960965498688 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 15:49:30.184512 131960965498688 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 15:49:30.209477 131960965498688 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 15:49:30.209559 131960965498688 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 15:49:30.275290 131960965498688 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 15:49:30.275374 131960965498688 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 15:49:30.789071 131960965498688 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 15:49:30.789181 131960965498688 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 15:49:33.254778 131960965498688 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 15:49:33.254902 131960965498688 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 15:49:33.254955 131960965498688 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 15:49:33.255014 131960965498688 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0420 15:49:33.255091 131960965498688 maxtext_utils.py:1827]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 15:49:33.255141 131960965498688 maxtext_utils.py:1827]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 15:49:33.255204 131960965498688 maxtext_utils.py:1827]  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)
I0420 15:49:33.255269 131960965498688 maxtext_utils.py:1827]  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')
I0420 15:49:33.255310 131960965498688 maxtext_utils.py:1827]  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)
I0420 15:49:33.255346 131960965498688 maxtext_utils.py:1827]  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)
I0420 15:49:33.255393 131960965498688 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0420 15:49:33.255440 131960965498688 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0420 15:49:34.650363 131960965498688 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 15:49:34.650461 131960965498688 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 15:49:34.665980 131960965498688 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 15:49:34.666048 131960965498688 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 15:49:47.929028 131960965498688 max_utils.py:791] Total memory size: 1.8 GB, Output size: 0.4 GB, Temp size: 1.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0420 15:49:47.929844 131960965498688 metric_logger.py:301] number parameters: 1.104 billion
I0420 15:50:03.166550 131960965498688 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 15:50:03.410194 131960965498688 checkpointing.py:798] Waited 0.24364686012268066 seconds for step 0 to finish before starting checkpointing.
I0420 15:50:03.412655 131960965498688 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 15:50:03.414615 131960965498688 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 0
I0420 15:50:03.416011 131960965498688 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/0.
I0420 15:50:04.149735 131960965498688 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 15:50:04.150701 131960965498688 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 15:50:04.150761 131960965498688 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
I0420 15:50:04.437921 131960965498688 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.288258s
I0420 15:50:04.438106 131960965498688 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 5.247 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.29396986961364746 s) (per-host)
I0420 15:50:04.438161 131960965498688 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.294035s (batch_requests_ready=0.002182s, total_serialization_initiated=0.291780s, others=0.000073s)
I0420 15:50:04.438254 131960965498688 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.298509s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.298491)
I0420 15:50:04.439032 131960965498688 event_tracking.py:125] [process=2] [async] Finished blocking save in 1.02 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/0.
I0420 15:50:04.439348 131830905132800 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 16:10:04.439310
I0420 15:50:04.441373 131960965498688 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 15:50:04.441679 131828744693504 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save.
I0420 15:50:04.441838 131960965498688 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776700203.412637, 'wait_for_prev_duration_secs': 6.4849853515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776700203.414654, 'checkpointer_blocking_duration_secs': 1.0248496532440186, 'get_old_steps_start_time': 1776700204.4395425, 'get_old_steps_duration_secs': 3.933906555175781e-05, 'checkpoint_manager_blocking_start_time': 1776700203.4107168, 'checkpoint_manager_blocking_duration_secs': 1.031080961227417}
I0420 15:50:04.441948 131960965498688 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 15:50:04.441996 131960965498688 max_utils.py:750] 
Memstats: After params initialized:
I0420 15:50:04.442046 131960965498688 max_utils.py:756] 	Using (GB) 0.44 / 31.25 (1.408000%) on TPU_8(process=2,(0,2,0,0))
I0420 15:50:04.442079 131960965498688 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_9(process=2,(1,2,0,0))
I0420 15:50:04.442106 131960965498688 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_12(process=2,(0,3,0,0))
I0420 15:50:04.442129 131960965498688 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_13(process=2,(1,3,0,0))
I0420 15:50:04.751501 131960965498688 metric_logger.py:196] completed step: 0, seconds: 15.237, TFLOP/s/device: 0.892, Tokens/s/device: 134.414, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277
I0420 15:50:05.021352 131960965498688 metric_logger.py:196] completed step: 1, seconds: 1.583, TFLOP/s/device: 8.581, Tokens/s/device: 1293.409, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52796.277
I0420 15:50:05.429816 131960965498688 metric_logger.py:196] completed step: 2, seconds: 0.041, TFLOP/s/device: 331.920, Tokens/s/device: 50030.536, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28634.820
I0420 15:50:05.659312 131960965498688 metric_logger.py:196] completed step: 3, seconds: 0.408, TFLOP/s/device: 33.263, Tokens/s/device: 5013.722, total_weights: 65536, loss: 9.731, lm_loss: 9.731, perplexity: 16827.199
I0420 15:50:06.118723 131960965498688 metric_logger.py:196] completed step: 4, seconds: 0.235, TFLOP/s/device: 57.749, Tokens/s/device: 8704.596, total_weights: 65536, loss: 9.272, lm_loss: 9.272, perplexity: 10638.531
I0420 15:50:06.124939 131960965498688 metric_logger.py:196] completed step: 5, seconds: 0.229, TFLOP/s/device: 59.209, Tokens/s/device: 8924.681, total_weights: 65536, loss: 8.887, lm_loss: 8.887, perplexity: 7233.941
I0420 15:50:13.035472    2902 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 15:50:15.446384 131828786657024 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/0/items/array_metadatas/process_2
I0420 15:50:35.994618 131960965498688 metric_logger.py:196] completed step: 6, seconds: 0.460, TFLOP/s/device: 29.548, Tokens/s/device: 4453.868, total_weights: 65536, loss: 8.588, lm_loss: 8.588, perplexity: 5365.468
I0420 15:50:36.224587 131960965498688 metric_logger.py:196] completed step: 7, seconds: 29.642, TFLOP/s/device: 0.458, Tokens/s/device: 69.092, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4358.779
I0420 15:50:36.455492 131960965498688 metric_logger.py:196] completed step: 8, seconds: 0.234, TFLOP/s/device: 58.105, Tokens/s/device: 8758.200, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.453
I0420 15:50:36.684340 131960965498688 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 15:50:36.685034 131960965498688 checkpointing.py:798] Waited 0.0007147789001464844 seconds for step 9 to finish before starting checkpointing.
I0420 15:50:36.687137 131960965498688 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 15:50:44.220617 131830905132800 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 39.415 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.07644081115723 s) (per-host)
I0420 15:50:44.220739 131830905132800 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 39.781280s.
I0420 15:50:51.981571 131830905132800 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 47.542096s.
I0420 15:50:51.981883 131828744693504 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 15:50:51.982024 131828744693504 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save.
I0420 15:50:51.982091 131828744693504 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 15:50:51.984553 131828744693504 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 15:50:51.984732 131960965498688 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 15:50:51.984874 131960965498688 checkpoint_manager.py:1452] Waiting for previous save to complete took 15.297731 seconds. If this number is high, consider checkpointing less frequently.
I0420 15:50:51.986700 131960965498688 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 9
I0420 15:50:51.988928 131960965498688 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/9.
I0420 15:50:52.729727 131960965498688 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 15:50:52.729818 131960965498688 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
I0420 15:50:52.763177 131960965498688 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034534s
I0420 15:50:52.763314 131960965498688 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 40.656 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.037941932678222656 s) (per-host)
I0420 15:50:52.763361 131960965498688 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.037998s (batch_requests_ready=0.001774s, total_serialization_initiated=0.036161s, others=0.000063s)
I0420 15:50:52.763442 131960965498688 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042169s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.042153)
I0420 15:50:52.764212 131960965498688 event_tracking.py:125] [process=2] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/9.
I0420 15:50:52.764508 131828744693504 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 16:10:52.764476
I0420 15:50:52.766432 131960965498688 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 15:50:52.766706 131828786657024 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save.
I0420 15:50:52.766842 131960965498688 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776700236.6871064, 'wait_for_prev_duration_secs': 15.297730684280396, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776700251.9867408, 'checkpointer_blocking_duration_secs': 0.7779183387756348, 'get_old_steps_start_time': 1776700252.7646828, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776700236.6852584, 'checkpoint_manager_blocking_duration_secs': 16.08155083656311}
I0420 15:50:52.767003 131960965498688 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 15:50:52.767048 131960965498688 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 15:50:58.414455 131805197829888 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/9/items/array_metadatas/process_2
I0420 15:51:35.045594 131828744693504 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 37.325 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.3201789855957 s) (per-host)
I0420 15:51:35.045720 131828744693504 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.281087s.
I0420 15:51:45.164677 131828744693504 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 52.400029s.
I0420 15:51:45.164971 131828786657024 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 15:51:45.165106 131828786657024 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save.
I0420 15:51:45.165157 131828786657024 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 15:51:45.167230 131828786657024 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 15:51:45.167410 131960965498688 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0420 15:51:45.167570 131960965498688 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 15:51:45.168587 131960965498688 metric_logger.py:196] completed step: 9, seconds: 0.230, TFLOP/s/device: 59.056, Tokens/s/device: 8901.523, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3552.333
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Mon Apr 20 15:51:53 UTC 2026
EXIT_CODE=0
NNX  ·  03aaf29ba  ·  feat_nnx_post_train_fixes_20260420_153543  ·  full log
XPK Start: Mon Apr 20 17:41:15 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-20 17:41:39.812939: 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)
I0420 17:41:39.994723 138993608660800 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 17:41:49.037034 138993608660800 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 17:41:49.039343 138993608660800 distributed.py:172] Connecting to JAX distributed service on mt-04-int8-1q65q-slice-job-0-0.mt-04-int8-1q65q:8482
I0420 17:41:50.709883 138993608660800 max_utils.py:284] Jax distributed system initialized!
I0420 17:41:57.000005 138993608660800 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 17:41:57.000105 138993608660800 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 17:41:57.000146 138993608660800 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
I0420 17:41:57.000180 138993608660800 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 17:41:57.694519 138993608660800 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 17:41:58.244336 138993608660800 checkpointing.py:688] Setting up checkpoint logger...
I0420 17:41:58.244453 138993608660800 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 17:41:58.244498 138993608660800 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 17:41:58.244727 138993608660800 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 0x7e6940e50410>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 17:42:01.084997 138993608660800 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 17:42:01.085223 138993608660800 checkpoint_manager.py:708] [process=1][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e540c686000>}, handler_registry=None
I0420 17:42:01.085459 138993608660800 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e540c686000>` 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`.
I0420 17:42:01.085508 138993608660800 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e540c6a9970>` 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`.
I0420 17:42:01.085543 138993608660800 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 0x7e540c686000>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e540c686000>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e540c6a9970>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e540c6a9970>}).
I0420 17:42:01.085904 138993608660800 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 17:42:01.085976 138993608660800 async_checkpointer.py:192] [process=1][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e540c1151c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 17:42:02.267416 138993608660800 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints
I0420 17:42:02.754380 138993608660800 checkpoint_manager.py:929] [process=1][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_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e540c69c2f0>
I0420 17:42:02.754553 138993608660800 checkpointing.py:302] Checkpoint manager created!
I0420 17:42:04.127649 138993608660800 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 17:42:04.127777 138993608660800 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 17:42:06.465894 138993608660800 maxtext_utils.py:1836]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0420 17:42:06.466007 138993608660800 maxtext_utils.py:1836]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0420 17:42:06.466058 138993608660800 maxtext_utils.py:1836]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0420 17:42:06.466109 138993608660800 maxtext_utils.py:1836]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0420 17:42:06.466148 138993608660800 maxtext_utils.py:1836]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0420 17:42:06.466183 138993608660800 maxtext_utils.py:1836]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0420 17:42:06.466219 138993608660800 maxtext_utils.py:1836]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:42:06.466272 138993608660800 maxtext_utils.py:1836]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0420 17:42:06.466322 138993608660800 maxtext_utils.py:1836]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:42:06.466359 138993608660800 maxtext_utils.py:1836]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:42:06.466392 138993608660800 maxtext_utils.py:1836]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0420 17:42:06.466425 138993608660800 maxtext_utils.py:1836]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')

I0420 17:42:06.878256 138993608660800 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 17:42:06.878347 138993608660800 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:42:06.883439 138993608660800 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 17:42:06.883495 138993608660800 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:42:06.899819 138993608660800 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 17:42:06.899877 138993608660800 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:42:06.915447 138993608660800 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 17:42:06.915508 138993608660800 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:42:06.996439 138993608660800 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:42:06.996534 138993608660800 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:42:07.012346 138993608660800 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:42:07.012410 138993608660800 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:42:07.028081 138993608660800 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 17:42:07.028142 138993608660800 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:42:07.057914 138993608660800 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 17:42:07.057985 138993608660800 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:42:07.543812 138993608660800 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 17:42:07.543903 138993608660800 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 17:42:21.612354 138993608660800 max_utils.py:791] Total memory size: 1.8 GB, Output size: 0.4 GB, Temp size: 1.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0420 17:42:21.616015 138993608660800 metric_logger.py:301] number parameters: 1.104 billion
I0420 17:42:36.341220 138993608660800 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 17:42:36.580647 138993608660800 checkpointing.py:798] Waited 0.2394113540649414 seconds for step 0 to finish before starting checkpointing.
I0420 17:42:36.583701 138993608660800 checkpoint_manager.py:2009] [process=1][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 17:42:36.585907 138993608660800 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 0
I0420 17:42:36.589258 138993608660800 event_tracking.py:70] [process=1] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/0.
I0420 17:42:36.971962 138993608660800 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 17:42:36.972990 138993608660800 jax_array_handlers.py:360] Scheduling D2H of 111 prioritized jax.Array.
I0420 17:42:36.973051 138993608660800 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
I0420 17:42:37.252180 138993608660800 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.281882s
I0420 17:42:37.252354 138993608660800 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 5.202 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.29651498794555664 s) (per-host)
I0420 17:42:37.252408 138993608660800 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.296578s (batch_requests_ready=0.004608s, total_serialization_initiated=0.291900s, others=0.000070s)
I0420 17:42:37.252524 138993608660800 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.300728s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.300710)
I0420 17:42:37.253358 138993608660800 event_tracking.py:125] [process=1] [async] Finished blocking save in 0.67 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/0.
I0420 17:42:37.253646 138860957460224 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 18:02:37.253623
I0420 17:42:37.259906 138993608660800 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:42:37.260174 138860453426944 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:42:37.260297 138993608660800 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776706956.5836828, 'wait_for_prev_duration_secs': 5.9604644775390625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776706956.5859473, 'checkpointer_blocking_duration_secs': 0.6678225994110107, 'get_old_steps_start_time': 1776706957.2537959, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1776706956.5811102, 'checkpoint_manager_blocking_duration_secs': 0.6791467666625977}
I0420 17:42:37.260412 138993608660800 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 17:42:37.260493 138993608660800 max_utils.py:750] 
Memstats: After params initialized:
I0420 17:42:37.260543 138993608660800 max_utils.py:756] 	Using (GB) 0.44 / 31.25 (1.408000%) on TPU_2(process=1,(2,0,0,0))
I0420 17:42:37.260577 138993608660800 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_3(process=1,(3,0,0,0))
I0420 17:42:37.260611 138993608660800 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_6(process=1,(2,1,0,0))
I0420 17:42:37.260637 138993608660800 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_7(process=1,(3,1,0,0))
I0420 17:42:37.591830 138993608660800 metric_logger.py:196] completed step: 0, seconds: 14.723, TFLOP/s/device: 0.923, Tokens/s/device: 139.099, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52668.902
I0420 17:42:37.836906 138993608660800 metric_logger.py:196] completed step: 1, seconds: 1.248, TFLOP/s/device: 10.885, Tokens/s/device: 1640.721, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52668.902
I0420 17:42:38.294351 138993608660800 metric_logger.py:196] completed step: 2, seconds: 0.019, TFLOP/s/device: 733.211, Tokens/s/device: 110517.511, total_weights: 65536, loss: 10.260, lm_loss: 10.260, perplexity: 28557.658
I0420 17:42:38.523890 138993608660800 metric_logger.py:196] completed step: 3, seconds: 0.437, TFLOP/s/device: 31.110, Tokens/s/device: 4689.246, total_weights: 65536, loss: 9.728, lm_loss: 9.728, perplexity: 16789.145
I0420 17:42:38.984357 138993608660800 metric_logger.py:196] completed step: 4, seconds: 0.257, TFLOP/s/device: 52.842, Tokens/s/device: 7964.905, total_weights: 65536, loss: 9.271, lm_loss: 9.271, perplexity: 10625.563
I0420 17:42:38.993219 138993608660800 metric_logger.py:196] completed step: 5, seconds: 0.229, TFLOP/s/device: 59.297, Tokens/s/device: 8937.845, total_weights: 65536, loss: 8.885, lm_loss: 8.885, perplexity: 7225.917
I0420 17:42:40.969293    2987 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 17:42:43.019068 138860935063296 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 111 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/0/items/array_metadatas/process_1
I0420 17:43:07.332400 138993608660800 metric_logger.py:196] completed step: 6, seconds: 0.461, TFLOP/s/device: 29.484, Tokens/s/device: 4444.203, total_weights: 65536, loss: 8.586, lm_loss: 8.586, perplexity: 5354.806
I0420 17:43:07.562688 138993608660800 metric_logger.py:196] completed step: 7, seconds: 28.112, TFLOP/s/device: 0.483, Tokens/s/device: 72.851, total_weights: 65536, loss: 8.378, lm_loss: 8.378, perplexity: 4350.756
I0420 17:43:07.792060 138993608660800 metric_logger.py:196] completed step: 8, seconds: 0.236, TFLOP/s/device: 57.520, Tokens/s/device: 8669.994, total_weights: 65536, loss: 8.249, lm_loss: 8.249, perplexity: 3824.619
I0420 17:43:08.023131 138993608660800 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 17:43:08.024085 138993608660800 checkpointing.py:798] Waited 0.0009648799896240234 seconds for step 9 to finish before starting checkpointing.
I0420 17:43:08.026078 138993608660800 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:43:11.064019 138860957460224 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 46.311 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 34.10814189910889 s) (per-host)
I0420 17:43:11.064145 138860957460224 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 33.810406s.
I0420 17:43:20.473430 138860957460224 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 43.219675s.
I0420 17:43:20.473651 138860453426944 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:43:20.473733 138860453426944 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:43:20.473795 138860453426944 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:43:20.475547 138860453426944 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 17:43:20.475751 138993608660800 checkpoint_manager.py:2032] [process=1][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 17:43:20.475894 138993608660800 checkpoint_manager.py:1452] Waiting for previous save to complete took 12.449813 seconds. If this number is high, consider checkpointing less frequently.
I0420 17:43:20.477818 138993608660800 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 9
I0420 17:43:20.480010 138993608660800 event_tracking.py:70] [process=1] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/9.
I0420 17:43:21.212307 138993608660800 jax_array_handlers.py:360] Scheduling D2H of 111 prioritized jax.Array.
I0420 17:43:21.212409 138993608660800 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
I0420 17:43:21.245274 138993608660800 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035482s
I0420 17:43:21.245447 138993608660800 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 32.498 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04746723175048828 s) (per-host)
I0420 17:43:21.245499 138993608660800 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.047529s (batch_requests_ready=0.004232s, total_serialization_initiated=0.043230s, others=0.000068s)
I0420 17:43:21.245616 138993608660800 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.051926s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.051912)
I0420 17:43:21.246367 138993608660800 event_tracking.py:125] [process=1] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/9.
I0420 17:43:21.246721 138860453426944 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 18:03:21.246684
I0420 17:43:21.250063 138993608660800 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:43:21.250320 138860935063296 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:43:21.250488 138993608660800 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776706988.02605, 'wait_for_prev_duration_secs': 12.44981336593628, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776707000.4778664, 'checkpointer_blocking_duration_secs': 0.7689986228942871, 'get_old_steps_start_time': 1776707001.2468905, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776706988.0243587, 'checkpoint_manager_blocking_duration_secs': 13.226092338562012}
I0420 17:43:21.250631 138993608660800 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 17:43:21.250706 138993608660800 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:43:27.311068 138837560878848 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 111 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/nnx_xpk_feat_nnx_post_train_fixes_20260420_153543_04_int8/checkpoints/9/items/array_metadatas/process_1
I0420 17:44:03.472497 138860453426944 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 37.365 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.27447271347046 s) (per-host)
I0420 17:44:03.472624 138860453426944 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.225791s.
I0420 17:44:12.763243 138860453426944 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 51.516395s.
I0420 17:44:12.763518 138860935063296 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:44:12.763635 138860935063296 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:44:12.763705 138860935063296 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:44:12.765337 138860935063296 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 17:44:12.765533 138993608660800 checkpoint_manager.py:2032] [process=1][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0420 17:44:12.765744 138993608660800 checkpoint_manager.py:2009] [process=1][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 17:44:12.766853 138993608660800 metric_logger.py:196] completed step: 9, seconds: 0.230, TFLOP/s/device: 59.158, Tokens/s/device: 8916.870, total_weights: 65536, loss: 8.174, lm_loss: 8.174, perplexity: 3546.104
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Mon Apr 20 17:44:25 UTC 2026
EXIT_CODE=0