MaxView

‹ 02_syntheticCase: 03_dropout04_int8 ›

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

MetricLinen  574ad3fb9NNX  574ad3fb9Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss6.98107.0160+0.035
TFLOP/s11.06811.252+0.184
Tok/s1769.01798.4+29.402
Avg s/step1.4281.460+0.032
Memory %1.311.310
JAX0.9.20.9.2

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

Linen  ·  574ad3fb9  ·  feat_nnx_post_train_fixes_20260418_235007  ·  full log
XPK Start: Sat Apr 18 23:55:39 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-18 23:56:03.102613: 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)
I0418 23:56:03.281724 140667799975744 max_utils.py:273] Attempting to initialize the jax distributed system...
I0418 23:56:12.322410 140667799975744 distributed.py:149] Starting JAX distributed service on [::]:8482
I0418 23:56:12.324642 140667799975744 distributed.py:172] Connecting to JAX distributed service on mt-03-dropout-1gq4a-slice-job-0-0.mt-03-dropout-1gq4a:8482
I0418 23:56:14.654660 140667799975744 max_utils.py:284] Jax distributed system initialized!
I0418 23:56:20.409855 140667799975744 max_utils.py:800] System Information: Jax Version: 0.9.2
I0418 23:56:20.409957 140667799975744 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0418 23:56:20.409998 140667799975744 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
I0418 23:56:20.410031 140667799975744 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0418 23:56:21.467888 140667799975744 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0418 23:56:21.468487 140667799975744 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0418 23:56:21.468680 140667799975744 checkpointing.py:688] Setting up checkpoint logger...
I0418 23:56:21.468733 140667799975744 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0418 23:56:21.468777 140667799975744 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0418 23:56:21.469115 140667799975744 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 0x7fef0e84c0b0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0418 23:56:24.724965 140667799975744 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0418 23:56:24.725198 140667799975744 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fee025a4d70>}, handler_registry=None
I0418 23:56:24.725447 140667799975744 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fee025a4d70>` 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`.
I0418 23:56:24.725495 140667799975744 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fee0222dac0>` 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`.
I0418 23:56:24.725530 140667799975744 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 0x7fee025a4d70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fee025a4d70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fee0222dac0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fee0222dac0>}).
I0418 23:56:24.725849 140667799975744 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0418 23:56:24.725919 140667799975744 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7fdadc4a2c00> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0418 23:56:26.781973 140667799975744 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints
I0418 23:56:27.654964 140667799975744 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7fee025a6990>
I0418 23:56:27.655135 140667799975744 checkpointing.py:302] Checkpoint manager created!
I0418 23:56:28.007631 140667799975744 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0418 23:56:28.007734 140667799975744 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0418 23:56:28.386887 140667799975744 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0418 23:56:28.386977 140667799975744 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0418 23:56:28.403480 140667799975744 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0418 23:56:28.403539 140667799975744 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0418 23:56:28.427059 140667799975744 attentions.py:1154] attentions/query Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0418 23:56:28.427124 140667799975744 attentions.py:1154] attentions/query Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0418 23:56:28.443512 140667799975744 attentions.py:1155] attentions/key Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0418 23:56:28.443571 140667799975744 attentions.py:1155] attentions/key Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0418 23:56:28.459959 140667799975744 attentions.py:1156] attentions/value Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0418 23:56:28.460016 140667799975744 attentions.py:1156] attentions/value Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0418 23:56:28.485483 140667799975744 attentions.py:1197] attentions/out Logical: bfloat16[32,128,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0418 23:56:28.485552 140667799975744 attentions.py:1197] attentions/out Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0418 23:56:28.511478 140667799975744 linears.py:525] linears/x Logical: bfloat16[32,128,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0418 23:56:28.511543 140667799975744 linears.py:525] linears/x Physical: bfloat16[32,128,7168]....................................... ('fsdp', None, None).
I0418 23:56:28.736428 140667799975744 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0418 23:56:28.736535 140667799975744 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0418 23:56:30.167156 140667799975744 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0418 23:56:30.167293 140667799975744 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)
I0418 23:56:30.167347 140667799975744 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)
I0418 23:56:30.167406 140667799975744 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')
I0418 23:56:30.167458 140667799975744 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)
I0418 23:56:30.167497 140667799975744 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)
I0418 23:56:30.167553 140667799975744 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)
I0418 23:56:30.167610 140667799975744 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')
I0418 23:56:30.167651 140667799975744 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)
I0418 23:56:30.167688 140667799975744 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)
I0418 23:56:30.167736 140667799975744 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0418 23:56:30.167784 140667799975744 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0418 23:56:30.695291 140667799975744 train.py:157] train/xent Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0418 23:56:30.695381 140667799975744 train.py:157] train/xent Physical: float32[32,128]............................................. ('fsdp', None).
I0418 23:56:30.710721 140667799975744 train.py:164] train/z_loss Logical: float32[32,128]............................................. ('activation_embed_and_logits_batch', 'activation_length').
I0418 23:56:30.710781 140667799975744 train.py:164] train/z_loss Physical: float32[32,128]............................................. ('fsdp', None).
I0418 23:56:34.332741 140667799975744 max_utils.py:791] Total memory size: 0.8 GB, Output size: 0.4 GB, Temp size: 0.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0418 23:56:34.333522 140667799975744 metric_logger.py:301] number parameters: 1.104 billion
I0418 23:56:38.513349 140667799975744 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0418 23:56:38.588053 140667799975744 checkpointing.py:798] Waited 0.0746917724609375 seconds for step 0 to finish before starting checkpointing.
I0418 23:56:38.590657 140667799975744 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0418 23:56:38.592659 140667799975744 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0418 23:56:38.594382 140667799975744 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/0.
I0418 23:56:39.309823 140667799975744 signaling_client.py:364] Using JaxDistributedSignalingClient
I0418 23:56:39.310783 140667799975744 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0418 23:56:39.310840 140667799975744 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
I0418 23:56:39.586446 140667799975744 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.276695s
I0418 23:56:39.586640 140667799975744 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.468 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28213024139404297 s) (per-host)
I0418 23:56:39.586708 140667799975744 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.282205s (batch_requests_ready=0.002156s, total_serialization_initiated=0.279963s, others=0.000086s)
I0418 23:56:39.586824 140667799975744 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.286725s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.286707)
I0418 23:56:39.587606 140667799975744 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.99 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/0.
I0418 23:56:39.587946 140543191406336 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:16:39.587908
I0418 23:56:39.604353 140667799975744 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0418 23:56:39.604622 140541005707008 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0418 23:56:39.604755 140667799975744 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776556598.5906343, 'wait_for_prev_duration_secs': 0.00010347366333007812, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776556598.592697, 'checkpointer_blocking_duration_secs': 0.9953563213348389, 'get_old_steps_start_time': 1776556599.5880806, 'get_old_steps_duration_secs': 2.7894973754882812e-05, 'checkpoint_manager_blocking_start_time': 1776556598.5887508, 'checkpoint_manager_blocking_duration_secs': 1.0159640312194824}
I0418 23:56:39.604861 140667799975744 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0418 23:56:39.604911 140667799975744 max_utils.py:750] 
Memstats: After params initialized:
I0418 23:56:39.604960 140667799975744 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_24(process=6,(0,6,0,0))
I0418 23:56:39.604994 140667799975744 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_25(process=6,(1,6,0,0))
I0418 23:56:39.605021 140667799975744 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_28(process=6,(0,7,0,0))
I0418 23:56:39.605045 140667799975744 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_29(process=6,(1,7,0,0))
I0418 23:56:39.922759 140667799975744 metric_logger.py:196] completed step: 0, seconds: 4.180, TFLOP/s/device: 0.192, Tokens/s/device: 30.624, total_weights: 4096, loss: 10.889, lm_loss: 10.889, perplexity: 53560.699
I0418 23:56:40.011070 140667799975744 metric_logger.py:196] completed step: 1, seconds: 1.408, TFLOP/s/device: 0.569, Tokens/s/device: 90.916, total_weights: 4096, loss: 10.902, lm_loss: 10.902, perplexity: 54298.523
I0418 23:56:40.381994 140667799975744 metric_logger.py:196] completed step: 2, seconds: 0.018, TFLOP/s/device: 44.367, Tokens/s/device: 7091.020, total_weights: 4096, loss: 9.956, lm_loss: 9.956, perplexity: 21082.051
I0418 23:56:40.452730 140667799975744 metric_logger.py:196] completed step: 3, seconds: 0.371, TFLOP/s/device: 2.157, Tokens/s/device: 344.694, total_weights: 4096, loss: 9.139, lm_loss: 9.139, perplexity: 9314.691
I0418 23:56:40.594782 140667799975744 metric_logger.py:196] completed step: 4, seconds: 0.076, TFLOP/s/device: 10.548, Tokens/s/device: 1685.896, total_weights: 4096, loss: 8.459, lm_loss: 8.459, perplexity: 4715.493
I0418 23:56:40.600605 140667799975744 metric_logger.py:196] completed step: 5, seconds: 0.071, TFLOP/s/device: 11.337, Tokens/s/device: 1811.928, total_weights: 4096, loss: 7.924, lm_loss: 7.924, perplexity: 2762.373
I0418 23:56:42.835350    2756 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0418 23:56:46.097634 140541051295488 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/0/items/array_metadatas/process_6
I0418 23:56:51.288476 140667799975744 metric_logger.py:196] completed step: 6, seconds: 0.142, TFLOP/s/device: 5.625, Tokens/s/device: 899.053, total_weights: 4096, loss: 7.553, lm_loss: 7.553, perplexity: 1906.617
I0418 23:56:51.360664 140667799975744 metric_logger.py:196] completed step: 7, seconds: 10.618, TFLOP/s/device: 0.075, Tokens/s/device: 12.055, total_weights: 4096, loss: 7.267, lm_loss: 7.267, perplexity: 1431.653
I0418 23:56:51.431284 140667799975744 metric_logger.py:196] completed step: 8, seconds: 0.075, TFLOP/s/device: 10.671, Tokens/s/device: 1705.530, total_weights: 4096, loss: 7.111, lm_loss: 7.111, perplexity: 1225.983
I0418 23:56:51.501717 140667799975744 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0418 23:56:51.502377 140667799975744 checkpointing.py:798] Waited 0.0006783008575439453 seconds for step 9 to finish before starting checkpointing.
I0418 23:56:51.504440 140667799975744 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0418 23:57:12.113334 140543191406336 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 48.146 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 32.80880379676819 s) (per-host)
I0418 23:57:12.113440 140543191406336 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.524845s.
I0418 23:57:21.469576 140543191406336 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 41.880965s.
I0418 23:57:21.469849 140541005707008 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0418 23:57:21.469965 140541005707008 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0418 23:57:21.470017 140541005707008 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0418 23:57:21.471434 140541005707008 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0418 23:57:21.471608 140667799975744 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0418 23:57:21.471744 140667799975744 checkpoint_manager.py:1452] Waiting for previous save to complete took 29.967304 seconds. If this number is high, consider checkpointing less frequently.
I0418 23:57:21.473748 140667799975744 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0418 23:57:21.475887 140667799975744 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/9.
I0418 23:57:22.191451 140667799975744 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0418 23:57:22.191544 140667799975744 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
I0418 23:57:22.228230 140667799975744 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037933s
I0418 23:57:22.228390 140667799975744 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 37.343 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04130840301513672 s) (per-host)
I0418 23:57:22.228439 140667799975744 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.041366s (batch_requests_ready=0.001743s, total_serialization_initiated=0.039558s, others=0.000065s)
I0418 23:57:22.228521 140667799975744 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.045523s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.045508)
I0418 23:57:22.229253 140667799975744 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/9.
I0418 23:57:22.229619 140541005707008 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:17:22.229583
I0418 23:57:22.234335 140667799975744 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0418 23:57:22.234630 140515743848192 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0418 23:57:22.234794 140667799975744 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776556611.50441, 'wait_for_prev_duration_secs': 29.96730375289917, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776556641.4737866, 'checkpointer_blocking_duration_secs': 0.7559816837310791, 'get_old_steps_start_time': 1776556642.229793, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1776556611.5026069, 'checkpoint_manager_blocking_duration_secs': 30.73215103149414}
I0418 23:57:22.234910 140667799975744 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0418 23:57:22.234954 140667799975744 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0418 23:57:29.854345 140515999676160 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/9/items/array_metadatas/process_6
I0418 23:58:02.687171 140541005707008 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 39.002 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.50005841255188 s) (per-host)
I0418 23:58:02.687295 140541005707008 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.457562s.
I0418 23:58:13.167758 140541005707008 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.938010s.
I0418 23:58:13.168051 140515743848192 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0418 23:58:13.168177 140515743848192 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0418 23:58:13.168226 140515743848192 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0418 23:58:13.169998 140515743848192 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0418 23:58:13.170179 140667799975744 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0418 23:58:13.170354 140667799975744 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0418 23:58:13.171342 140667799975744 metric_logger.py:196] completed step: 9, seconds: 0.072, TFLOP/s/device: 11.068, Tokens/s/device: 1768.982, total_weights: 4096, loss: 6.981, lm_loss: 6.981, perplexity: 1075.867
Per train step:
 Total TFLOPs: 0.80 
 split as 99.60% learnable weight flops and 0.40% attention flops
XPK End: Sat Apr 18 23:58:24 UTC 2026
EXIT_CODE=0
NNX  ·  574ad3fb9  ·  feat_nnx_post_train_fixes_20260418_235007  ·  full log
XPK Start: Sun Apr 19 01:29:43 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-19 01:30:07.727170: 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)
I0419 01:30:07.907216 135571520059200 max_utils.py:273] Attempting to initialize the jax distributed system...
I0419 01:30:16.946435 135571520059200 distributed.py:149] Starting JAX distributed service on [::]:8482
I0419 01:30:16.948685 135571520059200 distributed.py:172] Connecting to JAX distributed service on mt-03-dropout-nau1x-slice-job-0-0.mt-03-dropout-nau1x:8482
I0419 01:30:18.419432 135571520059200 max_utils.py:284] Jax distributed system initialized!
I0419 01:30:24.462013 135571520059200 max_utils.py:800] System Information: Jax Version: 0.9.2
I0419 01:30:24.462114 135571520059200 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0419 01:30:24.462153 135571520059200 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
I0419 01:30:24.462187 135571520059200 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0419 01:30:25.157864 135571520059200 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 01:30:25.269862 135571520059200 checkpointing.py:688] Setting up checkpoint logger...
I0419 01:30:25.269975 135571520059200 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0419 01:30:25.270018 135571520059200 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0419 01:30:25.270225 135571520059200 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 0x7b4c7cb4cfb0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0419 01:30:28.595527 135571520059200 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0419 01:30:28.595760 135571520059200 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b3710163770>}, handler_registry=None
I0419 01:30:28.596009 135571520059200 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b3710163770>` 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`.
I0419 01:30:28.596057 135571520059200 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b3710160440>` 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`.
I0419 01:30:28.596091 135571520059200 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 0x7b3710163770>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b3710163770>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b3710160440>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b3710160440>}).
I0419 01:30:28.596423 135571520059200 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0419 01:30:28.596495 135571520059200 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b37101f0fe0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0419 01:30:32.707976 135571520059200 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints
I0419 01:30:33.718605 135571520059200 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b37101ae0c0>
I0419 01:30:33.718771 135571520059200 checkpointing.py:302] Checkpoint manager created!
I0419 01:30:34.095014 135571520059200 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0419 01:30:34.095127 135571520059200 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0419 01:30:35.464098 135571520059200 maxtext_utils.py:1836]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0419 01:30:35.464229 135571520059200 maxtext_utils.py:1836]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0419 01:30:35.464322 135571520059200 maxtext_utils.py:1836]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0419 01:30:35.464386 135571520059200 maxtext_utils.py:1836]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0419 01:30:35.464440 135571520059200 maxtext_utils.py:1836]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0419 01:30:35.464491 135571520059200 maxtext_utils.py:1836]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0419 01:30:35.464550 135571520059200 maxtext_utils.py:1836]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 01:30:35.464611 135571520059200 maxtext_utils.py:1836]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0419 01:30:35.464673 135571520059200 maxtext_utils.py:1836]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 01:30:35.464737 135571520059200 maxtext_utils.py:1836]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 01:30:35.464797 135571520059200 maxtext_utils.py:1836]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0419 01:30:35.464859 135571520059200 maxtext_utils.py:1836]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')

I0419 01:30:35.873610 135571520059200 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 01:30:35.873702 135571520059200 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0419 01:30:35.879926 135571520059200 nnx_decoders.py:465] Unknown Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 01:30:35.879984 135571520059200 nnx_decoders.py:465] Unknown Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0419 01:30:35.896501 135571520059200 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 01:30:35.896567 135571520059200 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0419 01:30:35.912483 135571520059200 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,128,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 01:30:35.912549 135571520059200 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,128,2048]....................................... ('fsdp', None, None).
I0419 01:30:35.936646 135571520059200 attentions.py:1154] attentions/query Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 01:30:35.936713 135571520059200 attentions.py:1154] attentions/query Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0419 01:30:35.952602 135571520059200 attentions.py:1155] attentions/key Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 01:30:35.952662 135571520059200 attentions.py:1155] attentions/key Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0419 01:30:35.968569 135571520059200 attentions.py:1156] attentions/value Logical: bfloat16[32,128,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0419 01:30:35.968634 135571520059200 attentions.py:1156] attentions/value Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0419 01:30:35.999061 135571520059200 attentions.py:1197] attentions/out Logical: bfloat16[32,128,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0419 01:30:35.999136 135571520059200 attentions.py:1197] attentions/out Physical: bfloat16[32,128,16,128]..................................... ('fsdp', None, None, None).
I0419 01:30:36.025734 135571520059200 linears.py:525] linears/x Logical: bfloat16[32,128,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0419 01:30:36.025802 135571520059200 linears.py:525] linears/x Physical: bfloat16[32,128,7168]....................................... ('fsdp', None, None).
I0419 01:30:39.818457 135571520059200 max_utils.py:791] Total memory size: 0.8 GB, Output size: 0.4 GB, Temp size: 0.4 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0419 01:30:39.820922 135571520059200 metric_logger.py:301] number parameters: 1.104 billion
I0419 01:30:43.865880 135571520059200 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0419 01:30:43.938865 135571520059200 checkpointing.py:798] Waited 0.07297563552856445 seconds for step 0 to finish before starting checkpointing.
I0419 01:30:43.941437 135571520059200 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 01:30:43.943417 135571520059200 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0419 01:30:43.944945 135571520059200 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/0.
I0419 01:30:45.083772 135571520059200 signaling_client.py:364] Using JaxDistributedSignalingClient
I0419 01:30:45.084791 135571520059200 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0419 01:30:45.084856 135571520059200 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
I0419 01:30:45.367695 135571520059200 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.284725s
I0419 01:30:45.367864 135571520059200 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.249 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2938823699951172 s) (per-host)
I0419 01:30:45.367917 135571520059200 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.293943s (batch_requests_ready=0.003035s, total_serialization_initiated=0.290839s, others=0.000069s)
I0419 01:30:45.368031 135571520059200 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.298181s (all_items=0.000019s, per_item={'items': '0.00001860'}, temp_paths=0.298162)
I0419 01:30:45.368860 135571520059200 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.43 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/0.
I0419 01:30:45.369220 135441827215104 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 01:50:45.369181
I0419 01:30:45.382586 135571520059200 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 01:30:45.382879 135440757679872 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0419 01:30:45.383034 135571520059200 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776562243.941418, 'wait_for_prev_duration_secs': 6.246566772460938e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776562243.9434555, 'checkpointer_blocking_duration_secs': 1.4259402751922607, 'get_old_steps_start_time': 1776562245.3694217, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776562243.9392276, 'checkpoint_manager_blocking_duration_secs': 1.4437689781188965}
I0419 01:30:45.383137 135571520059200 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0419 01:30:45.383202 135571520059200 max_utils.py:750] 
Memstats: After params initialized:
I0419 01:30:45.383266 135571520059200 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_24(process=6,(0,6,0,0))
I0419 01:30:45.383301 135571520059200 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_25(process=6,(1,6,0,0))
I0419 01:30:45.383328 135571520059200 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_28(process=6,(0,7,0,0))
I0419 01:30:45.383353 135571520059200 max_utils.py:756] 	Using (GB) 0.41 / 31.25 (1.312000%) on TPU_29(process=6,(1,7,0,0))
I0419 01:30:45.700817 135571520059200 metric_logger.py:196] completed step: 0, seconds: 4.044, TFLOP/s/device: 0.198, Tokens/s/device: 31.653, total_weights: 4096, loss: 10.874, lm_loss: 10.874, perplexity: 52805.141
I0419 01:30:45.795851 135571520059200 metric_logger.py:196] completed step: 1, seconds: 1.833, TFLOP/s/device: 0.437, Tokens/s/device: 69.834, total_weights: 4096, loss: 10.892, lm_loss: 10.892, perplexity: 53764.488
I0419 01:30:46.208224 135571520059200 metric_logger.py:196] completed step: 2, seconds: 0.025, TFLOP/s/device: 31.518, Tokens/s/device: 5037.387, total_weights: 4096, loss: 9.969, lm_loss: 9.969, perplexity: 21350.100
I0419 01:30:46.278855 135571520059200 metric_logger.py:196] completed step: 3, seconds: 0.403, TFLOP/s/device: 1.989, Tokens/s/device: 317.834, total_weights: 4096, loss: 9.125, lm_loss: 9.125, perplexity: 9186.262
I0419 01:30:46.420688 135571520059200 metric_logger.py:196] completed step: 4, seconds: 0.087, TFLOP/s/device: 9.159, Tokens/s/device: 1463.811, total_weights: 4096, loss: 8.469, lm_loss: 8.469, perplexity: 4765.735
I0419 01:30:46.428298 135571520059200 metric_logger.py:196] completed step: 5, seconds: 0.070, TFLOP/s/device: 11.429, Tokens/s/device: 1826.614, total_weights: 4096, loss: 7.908, lm_loss: 7.908, perplexity: 2719.938
I0419 01:30:48.242372    2807 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0419 01:30:50.186921 135441810429696 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/0/items/array_metadatas/process_6
I0419 01:30:56.931004 135571520059200 metric_logger.py:196] completed step: 6, seconds: 0.143, TFLOP/s/device: 5.616, Tokens/s/device: 897.534, total_weights: 4096, loss: 7.554, lm_loss: 7.554, perplexity: 1909.063
I0419 01:30:57.001282 135571520059200 metric_logger.py:196] completed step: 7, seconds: 10.433, TFLOP/s/device: 0.077, Tokens/s/device: 12.269, total_weights: 4096, loss: 7.313, lm_loss: 7.313, perplexity: 1500.206
I0419 01:30:57.071956 135571520059200 metric_logger.py:196] completed step: 8, seconds: 0.077, TFLOP/s/device: 10.428, Tokens/s/device: 1666.580, total_weights: 4096, loss: 7.091, lm_loss: 7.091, perplexity: 1200.912
I0419 01:30:57.143104 135571520059200 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0419 01:30:57.143797 135571520059200 checkpointing.py:798] Waited 0.0007028579711914062 seconds for step 9 to finish before starting checkpointing.
I0419 01:30:57.145737 135571520059200 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 01:31:17.483382 135441827215104 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 48.739 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 32.40934419631958 s) (per-host)
I0419 01:31:17.483513 135441827215104 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.114167s.
I0419 01:31:28.016375 135441827215104 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 42.647005s.
I0419 01:31:28.016663 135440757679872 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 01:31:28.016781 135440757679872 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0419 01:31:28.016833 135440757679872 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0419 01:31:28.019175 135440757679872 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0419 01:31:28.019434 135571520059200 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0419 01:31:28.019570 135571520059200 checkpoint_manager.py:1452] Waiting for previous save to complete took 30.873832 seconds. If this number is high, consider checkpointing less frequently.
I0419 01:31:28.021366 135571520059200 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0419 01:31:28.023421 135571520059200 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/9.
I0419 01:31:29.169365 135571520059200 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0419 01:31:29.169465 135571520059200 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
I0419 01:31:29.204490 135571520059200 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036744s
I0419 01:31:29.204660 135571520059200 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 35.133 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.043906450271606445 s) (per-host)
I0419 01:31:29.204724 135571520059200 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043979s (batch_requests_ready=0.002708s, total_serialization_initiated=0.041188s, others=0.000082s)
I0419 01:31:29.204840 135571520059200 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048009s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.047993)
I0419 01:31:29.205552 135571520059200 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.18 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/9.
I0419 01:31:29.205841 135440757679872 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 01:51:29.205808
I0419 01:31:29.207907 135571520059200 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 01:31:29.208137 135441810429696 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0419 01:31:29.208283 135571520059200 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776562257.1457074, 'wait_for_prev_duration_secs': 30.87383198738098, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776562288.0214055, 'checkpointer_blocking_duration_secs': 1.1845755577087402, 'get_old_steps_start_time': 1776562289.2060058, 'get_old_steps_duration_secs': 2.86102294921875e-05, 'checkpoint_manager_blocking_start_time': 1776562257.1440263, 'checkpoint_manager_blocking_duration_secs': 32.064202070236206}
I0419 01:31:29.208461 135571520059200 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0419 01:31:29.208518 135571520059200 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 01:31:35.910939 135418943006464 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_03_dropout/checkpoints/9/items/array_metadatas/process_6
I0419 01:32:10.826186 135440757679872 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.911 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.665395975112915 s) (per-host)
I0419 01:32:10.826328 135440757679872 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.620380s.
I0419 01:32:19.475131 135440757679872 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 50.269167s.
I0419 01:32:19.475364 135441810429696 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 01:32:19.475429 135441810429696 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0419 01:32:19.475476 135441810429696 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0419 01:32:19.478378 135441810429696 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0419 01:32:19.478627 135571520059200 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0419 01:32:19.478808 135571520059200 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 01:32:19.479818 135571520059200 metric_logger.py:196] completed step: 9, seconds: 0.071, TFLOP/s/device: 11.252, Tokens/s/device: 1798.384, total_weights: 4096, loss: 7.016, lm_loss: 7.016, perplexity: 1113.843
Per train step:
 Total TFLOPs: 0.80 
 split as 99.60% learnable weight flops and 0.40% attention flops
XPK End: Sun Apr 19 01:32:30 UTC 2026
EXIT_CODE=0