MaxView

Timing Test Reference Updates
‹ 01_baseCase: 02_synthetic03_dropout ›

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

MetricLinen  574ad3fb9NNX  03aaf29baDiff (NNX − Linen)
Parameters1.104 billion1.104 billion—
Final loss8.17908.1750-0.004
TFLOP/s92.40991.488-0.921
Tok/s13928.813790.0-138.807
Avg s/step2.6922.790+0.098
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 15:41:17 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 15:41:41.552811: 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 15:41:41.733072 132654924457792 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 15:41:50.775427 132654924457792 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 15:41:50.778043 132654924457792 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-rntp2-slice-job-0-0.mt-02-synthetic-rntp2:8482
I0420 15:41:52.506756 132654924457792 max_utils.py:284] Jax distributed system initialized!
I0420 15:41:57.761940 132654924457792 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 15:41:57.762044 132654924457792 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 15:41:57.762101 132654924457792 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 15:41:57.762140 132654924457792 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 15:41:58.451492 132654924457792 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 15:41:58.452074 132654924457792 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 15:41:58.452268 132654924457792 checkpointing.py:688] Setting up checkpoint logger...
I0420 15:41:58.452318 132654924457792 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 15:41:58.452361 132654924457792 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 15:41:58.452701 132654924457792 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 0x78a59a75e6c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 15:42:01.395601 132654924457792 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 15:42:01.395838 132654924457792 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7891d424d5b0>}, handler_registry=None
I0420 15:42:01.396075 132654924457792 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7891d424d5b0>` 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 15:42:01.396138 132654924457792 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7891d4040080>` 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 15:42:01.396173 132654924457792 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 0x7891d424d5b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7891d424d5b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7891d4040080>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7891d4040080>}).
I0420 15:42:01.396501 132654924457792 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 15:42:01.396570 132654924457792 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7891bc5b7740> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 15:42:02.442820 132654924457792 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_02_synthetic/checkpoints
I0420 15:42:02.894504 132654924457792 checkpoint_manager.py:929] [process=0][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_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7891d403b680>
I0420 15:42:02.894687 132654924457792 checkpointing.py:302] Checkpoint manager created!
I0420 15:42:03.809712 132654924457792 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 15:42:03.809813 132654924457792 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 15:42:04.187891 132654924457792 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 15:42:04.187982 132654924457792 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 15:42:04.204294 132654924457792 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 15:42:04.204351 132654924457792 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 15:42:04.227684 132654924457792 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 15:42:04.227751 132654924457792 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 15:42:04.244203 132654924457792 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 15:42:04.244265 132654924457792 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 15:42:04.260536 132654924457792 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 15:42:04.260594 132654924457792 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 15:42:04.284951 132654924457792 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 15:42:04.285025 132654924457792 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 15:42:04.305693 132654924457792 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 15:42:04.305763 132654924457792 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 15:42:04.518685 132654924457792 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 15:42:04.518794 132654924457792 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 15:42:05.939427 132654924457792 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 15:42:05.939548 132654924457792 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 15:42:05.939612 132654924457792 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 15:42:05.939671 132654924457792 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 15:42:05.939723 132654924457792 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 15:42:05.939762 132654924457792 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 15:42:05.939814 132654924457792 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 15:42:05.939866 132654924457792 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 15:42:05.939912 132654924457792 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 15:42:05.939950 132654924457792 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 15:42:05.939999 132654924457792 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0420 15:42:05.940047 132654924457792 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0420 15:42:06.429277 132654924457792 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 15:42:06.429370 132654924457792 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 15:42:06.444647 132654924457792 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 15:42:06.444706 132654924457792 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 15:42:17.299239 132654924457792 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0420 15:42:17.306927 132654924457792 metric_logger.py:301] number parameters: 1.104 billion
I0420 15:42:28.799431 132654924457792 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 15:42:28.953581 132654924457792 checkpointing.py:798] Waited 0.1541283130645752 seconds for step 0 to finish before starting checkpointing.
I0420 15:42:28.955917 132654924457792 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 15:42:28.957871 132654924457792 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0420 15:42:28.959168 132654924457792 event_tracking.py:70] [process=0] [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_02_synthetic/checkpoints/0.
I0420 15:42:29.282330 132654924457792 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 15:42:29.297339 132654924457792 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 15:42:29.297404 132654924457792 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 15:42:29.563781 132525905143552 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0
I0420 15:42:29.566275 132654924457792 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.269730s
I0420 15:42:29.567092 132654924457792 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.592 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27585768699645996 s) (per-host)
I0420 15:42:29.567157 132654924457792 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.275952s (batch_requests_ready=0.002016s, total_serialization_initiated=0.273210s, others=0.000726s)
I0420 15:42:29.567231 132654924457792 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.281106s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.281088)
I0420 15:42:29.568056 132654924457792 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.61 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_02_synthetic/checkpoints/0.
I0420 15:42:29.568395 132525913536256 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 16:02:29.568357
I0420 15:42:29.578134 132654924457792 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 15:42:29.578380 132524256773888 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 15:42:29.578535 132654924457792 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_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776699748.9558985, 'wait_for_prev_duration_secs': 5.91278076171875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776699748.9579082, 'checkpointer_blocking_duration_secs': 0.6106400489807129, 'get_old_steps_start_time': 1776699749.568579, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1776699748.95398, 'checkpoint_manager_blocking_duration_secs': 0.624518871307373}
I0420 15:42:29.578641 132654924457792 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 15:42:29.578693 132654924457792 max_utils.py:750] 
Memstats: After params initialized:
I0420 15:42:29.578740 132654924457792 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0420 15:42:29.578772 132654924457792 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0420 15:42:29.578799 132654924457792 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0420 15:42:29.578823 132654924457792 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0420 15:42:29.930589 132654924457792 metric_logger.py:196] completed step: 0, seconds: 11.187, TFLOP/s/device: 1.215, Tokens/s/device: 183.063, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 15:42:29.934043 132654924457792 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/tensorboard/'
I0420 15:42:30.364543 132654924457792 metric_logger.py:196] completed step: 1, seconds: 1.129, TFLOP/s/device: 12.032, Tokens/s/device: 1813.571, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0420 15:42:30.511052 132654924457792 metric_logger.py:196] completed step: 2, seconds: 0.435, TFLOP/s/device: 31.256, Tokens/s/device: 4711.306, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0420 15:42:30.658210 132654924457792 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1439.162, Tokens/s/device: 216926.173, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0420 15:42:32.322202 132525888358144 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0/items
I0420 15:42:32.517858 132519969220352 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776699752049116400, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776699752049116400, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA
I0420 15:42:33.911383    2905 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 15:42:36.138979 132524273559296 array_metadata_store.py:203] [process=0][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_02_synthetic/checkpoints/0/items/array_metadatas/process_0
I0420 15:42:52.718042 132654924457792 metric_logger.py:196] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.117, Tokens/s/device: 13884.840, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0420 15:42:52.727264 132654924457792 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.112, Tokens/s/device: 13884.181, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0420 15:42:52.873295 132654924457792 metric_logger.py:196] completed step: 6, seconds: 22.058, TFLOP/s/device: 0.616, Tokens/s/device: 92.845, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0420 15:42:53.020668 132654924457792 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 1876.675, Tokens/s/device: 282872.928, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0420 15:42:53.167898 132654924457792 metric_logger.py:196] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.379, Tokens/s/device: 13924.300, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0420 15:42:53.314375 132654924457792 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 15:42:53.315108 132654924457792 checkpointing.py:798] Waited 0.0007386207580566406 seconds for step 9 to finish before starting checkpointing.
I0420 15:42:53.317128 132654924457792 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 15:43:10.729991 132524265166592 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 36.825920s (commit=35.620865s, array_metadata_write=1.205055s)
I0420 15:43:10.731206 132525913536256 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 38.118 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.439953565597534 s) (per-host)
I0420 15:43:10.731332 132525913536256 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.162824s.
I0420 15:43:11.528187 132525913536256 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776699752049116400, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA
I0420 15:43:12.525484 132525913536256 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 15:43:12.657974 132525913536256 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000236s.
I0420 15:43:12.696193 132519969220352 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776699752049116400, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0/_CHECKPOINT_METADATA
I0420 15:43:13.185432 132525913536256 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.530038s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0/items
I0420 15:43:13.187264 132525913536256 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0/items
I0420 15:43:13.765015 132525913536256 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0
I0420 15:43:15.183960 132525913536256 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0`.
I0420 15:43:15.184796 132525913536256 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 46.23 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/0
I0420 15:43:15.186496 132525913536256 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 45.617985s.
I0420 15:43:15.186774 132524256773888 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 15:43:15.186889 132524256773888 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 15:43:15.186954 132524256773888 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 15:43:15.189622 132524256773888 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 15:43:15.189750 132654924457792 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 15:43:15.189843 132654924457792 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.872715 seconds. If this number is high, consider checkpointing less frequently.
I0420 15:43:15.191879 132654924457792 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0420 15:43:15.194025 132654924457792 event_tracking.py:70] [process=0] [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_02_synthetic/checkpoints/9.
I0420 15:43:15.922344 132654924457792 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0420 15:43:15.922437 132654924457792 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 15:43:15.948049 132654924457792 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026475s
I0420 15:43:15.950654 132654924457792 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 47.153 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.032714128494262695 s) (per-host)
I0420 15:43:15.950729 132654924457792 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.032807s (batch_requests_ready=0.001824s, total_serialization_initiated=0.028468s, others=0.002516s)
I0420 15:43:15.950811 132654924457792 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.036903s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.036888)
I0420 15:43:15.951615 132654924457792 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.76 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_02_synthetic/checkpoints/9.
I0420 15:43:15.951902 132524256773888 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 16:03:15.951879
I0420 15:43:15.958278 132654924457792 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 15:43:15.958515 132524831401728 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0420 15:43:15.958657 132654924457792 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_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776699773.3170962, 'wait_for_prev_duration_secs': 21.87271475791931, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776699795.1919177, 'checkpointer_blocking_duration_secs': 0.7600829601287842, 'get_old_steps_start_time': 1776699795.9520254, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1776699773.3153293, 'checkpoint_manager_blocking_duration_secs': 22.64329433441162}
I0420 15:43:15.958760 132654924457792 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 15:43:15.958803 132654924457792 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 15:43:16.167493 132525913536256 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/9
I0420 15:43:18.371569 132507542468352 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/9/items
I0420 15:43:21.331256 132525351597824 array_metadata_store.py:203] [process=0][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_02_synthetic/checkpoints/9/items/array_metadatas/process_0
I0420 15:44:00.908006 132525343205120 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.367862s (commit=40.114337s, array_metadata_write=1.253526s)
I0420 15:44:00.909072 132524256773888 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 35.109 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.99112248420715 s) (per-host)
I0420 15:44:00.909177 132524256773888 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.957200s.
I0420 15:44:02.966849 132524256773888 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000278s.
I0420 15:44:03.222475 132524256773888 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0420 15:44:03.867678 132524256773888 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.983117s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/9/items
I0420 15:44:03.869425 132524256773888 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/9/items
I0420 15:44:04.431577 132524256773888 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/9
I0420 15:44:06.369549 132524256773888 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/9`.
I0420 15:44:06.370395 132524256773888 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 51.18 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260420_153543/linen_xpk_feat_nnx_post_train_fixes_20260420_153543_02_synthetic/checkpoints/9
I0420 15:44:06.371843 132524256773888 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 50.419862s.
I0420 15:44:06.371974 132524831401728 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 15:44:06.372037 132524831401728 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0420 15:44:06.372106 132524831401728 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 15:44:06.374602 132524831401728 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 15:44:06.374761 132654924457792 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0420 15:44:06.374926 132654924457792 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 15:44:06.375978 132654924457792 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.409, Tokens/s/device: 13928.846, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Mon Apr 20 15:44:17 UTC 2026
EXIT_CODE=0
NNX  ·  03aaf29ba  ·  feat_nnx_post_train_fixes_20260420_153543  ·  full log 
XPK Start: Mon Apr 20 17:36:52 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 17:37:31.416577: 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 17:37:31.595714 137679678887744 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 17:37:40.635774 137679678887744 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 17:37:40.638062 137679678887744 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-qw6t5-slice-job-0-0.mt-02-synthetic-qw6t5:8482
I0420 17:37:42.503784 137679678887744 max_utils.py:284] Jax distributed system initialized!
I0420 17:37:48.956257 137679678887744 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 17:37:48.956358 137679678887744 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 17:37:48.956397 137679678887744 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 17:37:48.956431 137679678887744 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 17:37:49.647768 137679678887744 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 17:37:49.759949 137679678887744 checkpointing.py:688] Setting up checkpoint logger...
I0420 17:37:49.760062 137679678887744 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 17:37:49.760108 137679678887744 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 17:37:49.760299 137679678887744 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 0x7d375494c560>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 17:37:53.072271 137679678887744 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 17:37:53.072500 137679678887744 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 0x7d230c46d940>}, handler_registry=None
I0420 17:37:53.072757 137679678887744 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d230c46d940>` 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 17:37:53.072810 137679678887744 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d230c13ddc0>` 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 17:37:53.072846 137679678887744 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 0x7d230c46d940>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d230c46d940>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d230c13ddc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d230c13ddc0>}).
I0420 17:37:53.073164 137679678887744 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 17:37:53.073235 137679678887744 async_checkpointer.py:192] [process=1][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d230c1d8fe0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 17:37:53.860889 137679678887744 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_02_synthetic/checkpoints
I0420 17:37:54.315461 137679678887744 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_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d230c727980>
I0420 17:37:54.315641 137679678887744 checkpointing.py:302] Checkpoint manager created!
I0420 17:37:55.286300 137679678887744 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 17:37:55.286418 137679678887744 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 17:37:56.675576 137679678887744 maxtext_utils.py:1836]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0420 17:37:56.675701 137679678887744 maxtext_utils.py:1836]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0420 17:37:56.675755 137679678887744 maxtext_utils.py:1836]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0420 17:37:56.675798 137679678887744 maxtext_utils.py:1836]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0420 17:37:56.675835 137679678887744 maxtext_utils.py:1836]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0420 17:37:56.675875 137679678887744 maxtext_utils.py:1836]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0420 17:37:56.675911 137679678887744 maxtext_utils.py:1836]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:37:56.675948 137679678887744 maxtext_utils.py:1836]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0420 17:37:56.675983 137679678887744 maxtext_utils.py:1836]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:37:56.676016 137679678887744 maxtext_utils.py:1836]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0420 17:37:56.676049 137679678887744 maxtext_utils.py:1836]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0420 17:37:56.676080 137679678887744 maxtext_utils.py:1836]  token_embedder/embedding/value
    Shape:     float32[32000,2048]

    Physical:  (None, 'fsdp')
I0420 17:37:57.076543 137679678887744 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 17:37:57.076648 137679678887744 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:37:57.081985 137679678887744 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 17:37:57.082042 137679678887744 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:37:57.098121 137679678887744 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 17:37:57.098182 137679678887744 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:37:57.113780 137679678887744 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 17:37:57.113851 137679678887744 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 17:37:57.138555 137679678887744 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 17:37:57.138628 137679678887744 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:37:57.154267 137679678887744 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 17:37:57.154327 137679678887744 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:37:57.169749 137679678887744 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 17:37:57.169810 137679678887744 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:37:57.199874 137679678887744 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 17:37:57.199948 137679678887744 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 17:37:57.219552 137679678887744 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 17:37:57.219626 137679678887744 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 17:38:08.620042 137679678887744 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0420 17:38:08.622422 137679678887744 metric_logger.py:301] number parameters: 1.104 billion
I0420 17:38:19.778101 137679678887744 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 17:38:19.931385 137679678887744 checkpointing.py:798] Waited 0.15326905250549316 seconds for step 0 to finish before starting checkpointing.
I0420 17:38:19.933832 137679678887744 checkpoint_manager.py:2009] [process=1][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 17:38:19.935671 137679678887744 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 0
I0420 17:38:19.937063 137679678887744 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_02_synthetic/checkpoints/0.
I0420 17:38:20.645012 137679678887744 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 17:38:20.646038 137679678887744 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0420 17:38:20.646096 137679678887744 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 17:38:20.930843 137679678887744 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.286612s
I0420 17:38:20.931015 137679678887744 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 5.217 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.29567956924438477 s) (per-host)
I0420 17:38:20.931070 137679678887744 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.295745s (batch_requests_ready=0.003033s, total_serialization_initiated=0.292638s, others=0.000074s)
I0420 17:38:20.931169 137679678887744 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.299920s (all_items=0.000019s, per_item={'items': '0.00001860'}, temp_paths=0.299901)
I0420 17:38:20.931987 137679678887744 event_tracking.py:125] [process=1] [async] Finished blocking save in 1.00 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_02_synthetic/checkpoints/0.
I0420 17:38:20.932340 137550622619392 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:58:20.932302
I0420 17:38:20.934381 137679678887744 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:38:20.934632 137550085748480 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:38:20.934800 137679678887744 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_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776706699.9338133, 'wait_for_prev_duration_secs': 6.580352783203125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776706699.93571, 'checkpointer_blocking_duration_secs': 0.9967379570007324, 'get_old_steps_start_time': 1776706700.9324741, 'get_old_steps_duration_secs': 2.86102294921875e-05, 'checkpoint_manager_blocking_start_time': 1776706699.9317844, 'checkpoint_manager_blocking_duration_secs': 1.002978801727295}
I0420 17:38:20.934912 137679678887744 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 17:38:20.934977 137679678887744 max_utils.py:750] 
Memstats: After params initialized:
I0420 17:38:20.935026 137679678887744 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_2(process=1,(2,0,0,0))
I0420 17:38:20.935060 137679678887744 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_3(process=1,(3,0,0,0))
I0420 17:38:20.935088 137679678887744 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_6(process=1,(2,1,0,0))
I0420 17:38:20.935113 137679678887744 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_7(process=1,(3,1,0,0))
I0420 17:38:21.252880 137679678887744 metric_logger.py:196] completed step: 0, seconds: 11.155, TFLOP/s/device: 1.218, Tokens/s/device: 183.601, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0420 17:38:21.423143 137679678887744 metric_logger.py:196] completed step: 1, seconds: 1.473, TFLOP/s/device: 9.225, Tokens/s/device: 1390.560, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0420 17:38:21.819572 137679678887744 metric_logger.py:196] completed step: 2, seconds: 0.024, TFLOP/s/device: 562.684, Tokens/s/device: 84813.849, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28632.133
I0420 17:38:21.967241 137679678887744 metric_logger.py:196] completed step: 3, seconds: 0.387, TFLOP/s/device: 35.150, Tokens/s/device: 5298.191, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16855.789
I0420 17:38:22.263450 137679678887744 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.699, Tokens/s/device: 12465.307, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10661.594
I0420 17:38:22.271437 137679678887744 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.320, Tokens/s/device: 13915.501, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7245.978
I0420 17:38:24.464921    2826 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 17:38:26.938127 137550568613632 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 69 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_02_synthetic/checkpoints/0/items/array_metadatas/process_1
I0420 17:38:44.737954 137679678887744 metric_logger.py:196] completed step: 6, seconds: 0.297, TFLOP/s/device: 45.719, Tokens/s/device: 6891.238, total_weights: 65536, loss: 8.589, lm_loss: 8.589, perplexity: 5371.156
I0420 17:38:44.885646 137679678887744 metric_logger.py:196] completed step: 7, seconds: 22.320, TFLOP/s/device: 0.609, Tokens/s/device: 91.755, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4360.465
I0420 17:38:45.033461 137679678887744 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.657, Tokens/s/device: 13363.349, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.256
I0420 17:38:45.181485 137679678887744 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0420 17:38:45.182179 137679678887744 checkpointing.py:798] Waited 0.0007033348083496094 seconds for step 9 to finish before starting checkpointing.
I0420 17:38:45.184169 137679678887744 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:38:57.995611 137550622619392 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 42.280 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.360238790512085 s) (per-host)
I0420 17:38:57.995729 137550622619392 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.062747s.
I0420 17:39:07.455445 137550622619392 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 46.522447s.
I0420 17:39:07.455760 137550085748480 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:39:07.455923 137550085748480 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:39:07.455975 137550085748480 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:39:07.458691 137550085748480 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 17:39:07.458861 137679678887744 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 17:39:07.458989 137679678887744 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.274822 seconds. If this number is high, consider checkpointing less frequently.
I0420 17:39:07.460899 137679678887744 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 9
I0420 17:39:07.463046 137679678887744 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_02_synthetic/checkpoints/9.
I0420 17:39:08.183176 137679678887744 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0420 17:39:08.183271 137679678887744 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 17:39:08.218147 137679678887744 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036627s
I0420 17:39:08.218302 137679678887744 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 35.453 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04351019859313965 s) (per-host)
I0420 17:39:08.218350 137679678887744 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043567s (batch_requests_ready=0.002595s, total_serialization_initiated=0.040909s, others=0.000064s)
I0420 17:39:08.218438 137679678887744 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.047717s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.047702)
I0420 17:39:08.219154 137679678887744 event_tracking.py:125] [process=1] [async] Finished blocking save in 0.76 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_02_synthetic/checkpoints/9.
I0420 17:39:08.219504 137550085748480 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 17:59:08.219470
I0420 17:39:08.221512 137679678887744 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 17:39:08.221751 137550614226688 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save.
I0420 17:39:08.221873 137679678887744 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_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776706725.1841395, 'wait_for_prev_duration_secs': 22.274821758270264, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776706747.4609382, 'checkpointer_blocking_duration_secs': 0.7587125301361084, 'get_old_steps_start_time': 1776706748.2196848, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776706725.1824303, 'checkpoint_manager_blocking_duration_secs': 23.03940987586975}
I0420 17:39:08.221983 137679678887744 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0420 17:39:08.222029 137679678887744 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 17:39:13.715210 137550568613632 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 69 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_02_synthetic/checkpoints/9/items/array_metadatas/process_1
I0420 17:39:50.394508 137550085748480 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 37.414 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.21967625617981 s) (per-host)
I0420 17:39:50.394631 137550085748480 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.175015s.
I0420 17:40:01.003180 137550085748480 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 52.783549s.
I0420 17:40:01.003391 137550614226688 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 17:40:01.003448 137550614226688 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save.
I0420 17:40:01.003499 137550614226688 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0420 17:40:01.005271 137550614226688 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0420 17:40:01.005458 137679678887744 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 17:40:01.005628 137679678887744 checkpoint_manager.py:2009] [process=1][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 17:40:01.006548 137679678887744 metric_logger.py:196] completed step: 9, seconds: 0.149, TFLOP/s/device: 91.488, Tokens/s/device: 13790.039, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3551.438
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Mon Apr 20 17:40:13 UTC 2026
EXIT_CODE=0