MaxView

‹ 01_baseCase: 02_synthetic03_dropout ›

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

MetricLinen  28b1e4a16NNX  28b1e4a16Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s92.27091.809-0.461
Tok/s13907.913838.5-69.449
Avg s/step2.6742.789+0.115
Memory %1.381.380
JAX0.9.20.9.2

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

Linen  ·  28b1e4a16  ·  feat_nnx_post_train_fixes_20260424_120657  ·  full log
XPK Start: Fri Apr 24 12:09:41 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-24 12:10:06.012659: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0424 12:10:06.224476 132344083617600 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 12:10:15.265642 132344083617600 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 12:10:15.267908 132344083617600 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-8ikpn-slice-job-0-0.mt-02-synthetic-8ikpn:8482
I0424 12:10:16.540572 132344083617600 max_utils.py:284] Jax distributed system initialized!
I0424 12:10:22.552057 132344083617600 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 12:10:22.552164 132344083617600 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 12:10:22.552205 132344083617600 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0424 12:10:22.552242 132344083617600 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 12:10:23.254651 132344083617600 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 12:10:23.255228 132344083617600 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 12:10:23.255408 132344083617600 checkpointing.py:688] Setting up checkpoint logger...
I0424 12:10:23.255470 132344083617600 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 12:10:23.255513 132344083617600 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 12:10:23.255856 132344083617600 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 0x785d2111c1d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 12:10:26.222076 132344083617600 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 12:10:26.222310 132344083617600 checkpoint_manager.py:708] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7849043e6900>}, handler_registry=None
I0424 12:10:26.222557 132344083617600 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7849043e6900>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0424 12:10:26.222607 132344083617600 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7849043ea450>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0424 12:10:26.222648 132344083617600 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 0x7849043e6900>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7849043e6900>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7849043ea450>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7849043ea450>}).
I0424 12:10:26.222983 132344083617600 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 12:10:26.223054 132344083617600 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x784904161e40> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 12:10:26.938407 132344083617600 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints
I0424 12:10:27.344661 132344083617600 checkpoint_manager.py:929] [process=3][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7849043e7470>
I0424 12:10:27.344830 132344083617600 checkpointing.py:302] Checkpoint manager created!
I0424 12:10:28.306096 132344083617600 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 12:10:28.306206 132344083617600 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 12:10:28.686337 132344083617600 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 12:10:28.686445 132344083617600 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 12:10:28.703117 132344083617600 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 12:10:28.703181 132344083617600 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 12:10:28.727239 132344083617600 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 12:10:28.727310 132344083617600 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 12:10:28.743999 132344083617600 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 12:10:28.744060 132344083617600 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 12:10:28.760713 132344083617600 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 12:10:28.760773 132344083617600 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 12:10:28.785713 132344083617600 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0424 12:10:28.785784 132344083617600 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 12:10:28.806864 132344083617600 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 12:10:28.806935 132344083617600 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 12:10:29.028918 132344083617600 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 12:10:29.029029 132344083617600 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 12:10:30.479886 132344083617600 maxtext_utils.py:1880]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0424 12:10:30.480020 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 12:10:30.480073 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0424 12:10:30.480131 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0424 12:10:30.480183 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 12:10:30.480226 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0424 12:10:30.480278 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0424 12:10:30.480329 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0424 12:10:30.480368 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0424 12:10:30.480404 132344083617600 maxtext_utils.py:1880]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0424 12:10:30.480487 132344083617600 maxtext_utils.py:1880]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0424 12:10:30.480542 132344083617600 maxtext_utils.py:1880]  params/params/token_embedder/embedding

    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0424 12:10:30.980171 132344083617600 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 12:10:30.980266 132344083617600 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 12:10:30.995836 132344083617600 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0424 12:10:30.995897 132344083617600 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0424 12:10:41.729207 132344083617600 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0424 12:10:41.730056 132344083617600 metric_logger.py:301] number parameters: 1.104 billion
I0424 12:10:52.986894 132344083617600 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 12:10:53.186244 132344083617600 checkpointing.py:798] Waited 0.19933462142944336 seconds for step 0 to finish before starting checkpointing.
I0424 12:10:53.189004 132344083617600 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 12:10:53.190798 132344083617600 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0
I0424 12:10:53.192335 132344083617600 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/0.
I0424 12:10:53.484116 132344083617600 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 12:10:53.485121 132344083617600 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 12:10:53.485179 132344083617600 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 12:10:53.762128 132344083617600 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278086s
I0424 12:10:53.762301 132344083617600 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 5.434 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28388547897338867 s) (per-host)
I0424 12:10:53.762351 132344083617600 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.283946s (batch_requests_ready=0.002627s, total_serialization_initiated=0.281250s, others=0.000069s)
I0424 12:10:53.762454 132344083617600 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288443s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.288425)
I0424 12:10:53.763289 132344083617600 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.57 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/0.
I0424 12:10:53.763655 132215724144384 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 12:30:53.763618
I0424 12:10:53.778823 132344083617600 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 12:10:53.779137 132215153346304 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 12:10:53.779294 132344083617600 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777032653.1889858, 'wait_for_prev_duration_secs': 6.127357482910156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777032653.190837, 'checkpointer_blocking_duration_secs': 0.5729751586914062, 'get_old_steps_start_time': 1777032653.7638395, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1777032653.1868088, 'checkpoint_manager_blocking_duration_secs': 0.5924475193023682}
I0424 12:10:53.779398 132344083617600 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 12:10:53.779463 132344083617600 max_utils.py:750] 
Memstats: After params initialized:
I0424 12:10:53.779510 132344083617600 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_10(process=3,(2,2,0,0))
I0424 12:10:53.779543 132344083617600 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_11(process=3,(3,2,0,0))
I0424 12:10:53.779572 132344083617600 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_14(process=3,(2,3,0,0))
I0424 12:10:53.779595 132344083617600 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_15(process=3,(3,3,0,0))
I0424 12:10:54.092505 132344083617600 metric_logger.py:196] completed step: 0, seconds: 11.257, TFLOP/s/device: 1.207, Tokens/s/device: 181.936, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 12:10:54.263592 132344083617600 metric_logger.py:196] completed step: 1, seconds: 1.104, TFLOP/s/device: 12.306, Tokens/s/device: 1854.822, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0424 12:10:54.662268 132344083617600 metric_logger.py:196] completed step: 2, seconds: 0.023, TFLOP/s/device: 578.644, Tokens/s/device: 87219.454, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0424 12:10:54.809525 132344083617600 metric_logger.py:196] completed step: 3, seconds: 0.399, TFLOP/s/device: 34.049, Tokens/s/device: 5132.189, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0424 12:10:55.104877 132344083617600 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.781, Tokens/s/device: 13382.035, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0424 12:10:55.110613 132344083617600 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.174, Tokens/s/device: 13893.506, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0424 12:10:57.122870    2804 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 12:10:59.353412 132215161739008 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/0/items/array_metadatas/process_3
I0424 12:11:16.902386 132344083617600 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.964, Tokens/s/device: 6928.188, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0424 12:11:17.049710 132344083617600 metric_logger.py:196] completed step: 7, seconds: 21.644, TFLOP/s/device: 0.628, Tokens/s/device: 94.621, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0424 12:11:17.197095 132344083617600 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.874, Tokens/s/device: 13396.128, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0424 12:11:17.343823 132344083617600 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 12:11:17.344488 132344083617600 checkpointing.py:798] Waited 0.0006859302520751953 seconds for step 9 to finish before starting checkpointing.
I0424 12:11:17.346527 132344083617600 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 12:11:31.400663 132215724144384 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 41.654 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.922210693359375 s) (per-host)
I0424 12:11:31.400785 132215724144384 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.637016s.
I0424 12:11:38.726879 132215724144384 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 44.963080s.
I0424 12:11:38.727205 132215153346304 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 12:11:38.727325 132215153346304 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 12:11:38.727383 132215153346304 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 12:11:38.728920 132215153346304 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 12:11:38.729127 132344083617600 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0424 12:11:38.729268 132344083617600 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.382741 seconds. If this number is high, consider checkpointing less frequently.
I0424 12:11:38.731104 132344083617600 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 9
I0424 12:11:38.733381 132344083617600 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/9.
I0424 12:11:39.283082 132344083617600 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0424 12:11:39.283174 132344083617600 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 12:11:39.323714 132344083617600 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041506s
I0424 12:11:39.323877 132344083617600 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 34.155 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04516458511352539 s) (per-host)
I0424 12:11:39.323932 132344083617600 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045222s (batch_requests_ready=0.001923s, total_serialization_initiated=0.043234s, others=0.000065s)
I0424 12:11:39.324019 132344083617600 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049594s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.049579)
I0424 12:11:39.324710 132344083617600 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.59 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/9.
I0424 12:11:39.325063 132215153346304 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 12:31:39.325026
I0424 12:11:39.326996 132344083617600 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 12:11:39.327194 132193229719296 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save.
I0424 12:11:39.327305 132344083617600 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777032677.3464973, 'wait_for_prev_duration_secs': 21.382741451263428, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777032698.7311428, 'checkpointer_blocking_duration_secs': 0.5940582752227783, 'get_old_steps_start_time': 1777032699.3252242, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1777032677.344712, 'checkpoint_manager_blocking_duration_secs': 21.982558488845825}
I0424 12:11:39.327410 132344083617600 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 12:11:39.327471 132344083617600 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 12:11:44.513097 132215161739008 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/9/items/array_metadatas/process_3
I0424 12:12:19.889492 132215153346304 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 38.896 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.610745906829834 s) (per-host)
I0424 12:12:19.889599 132215153346304 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.564426s.
I0424 12:12:30.091538 132215153346304 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 50.766350s.
I0424 12:12:30.091893 132193229719296 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 12:12:30.091948 132193229719296 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save.
I0424 12:12:30.091991 132193229719296 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 12:12:30.093444 132193229719296 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 12:12:30.093600 132344083617600 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 12:12:30.093760 132344083617600 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 12:12:30.094700 132344083617600 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.270, Tokens/s/device: 13907.941, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 24 12:12:41 UTC 2026
EXIT_CODE=0
NNX  ·  28b1e4a16  ·  feat_nnx_post_train_fixes_20260424_120657  ·  full log
XPK Start: Fri Apr 24 13:34:19 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-24 13:34:44.217810: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0424 13:34:44.429071 140054653589312 max_utils.py:273] Attempting to initialize the jax distributed system...
I0424 13:34:53.467699 140054653589312 distributed.py:149] Starting JAX distributed service on [::]:8482
I0424 13:34:53.470033 140054653589312 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-jroc6-slice-job-0-0.mt-02-synthetic-jroc6:8482
I0424 13:34:54.816418 140054653589312 max_utils.py:284] Jax distributed system initialized!
I0424 13:35:00.886041 140054653589312 max_utils.py:800] System Information: Jax Version: 0.9.2
I0424 13:35:00.886145 140054653589312 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0424 13:35:00.886189 140054653589312 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0424 13:35:00.886225 140054653589312 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0424 13:35:01.579359 140054653589312 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0424 13:35:01.692224 140054653589312 checkpointing.py:688] Setting up checkpoint logger...
I0424 13:35:01.692341 140054653589312 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0424 13:35:01.692385 140054653589312 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0424 13:35:01.692577 140054653589312 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 0x7f604c63c4d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0424 13:35:04.717318 140054653589312 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0424 13:35:04.717552 140054653589312 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f4c40567980>}, handler_registry=None
I0424 13:35:04.717797 140054653589312 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f4c40567980>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0424 13:35:04.717844 140054653589312 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f4c40553800>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0424 13:35:04.717878 140054653589312 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 0x7f4c40567980>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f4c40567980>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f4c40553800>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f4c40553800>}).
I0424 13:35:04.718206 140054653589312 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0424 13:35:04.718276 140054653589312 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f4c40543600> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0424 13:35:05.886916 140054653589312 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints
I0424 13:35:06.171783 140054653589312 checkpoint_manager.py:929] [process=5][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_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f4c583c0440>
I0424 13:35:06.171953 140054653589312 checkpointing.py:302] Checkpoint manager created!
I0424 13:35:07.142918 140054653589312 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0424 13:35:07.143044 140054653589312 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0424 13:35:08.529224 140054653589312 maxtext_utils.py:1889]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0424 13:35:08.529341 140054653589312 maxtext_utils.py:1889]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0424 13:35:08.529392 140054653589312 maxtext_utils.py:1889]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0424 13:35:08.529435 140054653589312 maxtext_utils.py:1889]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0424 13:35:08.529469 140054653589312 maxtext_utils.py:1889]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0424 13:35:08.529503 140054653589312 maxtext_utils.py:1889]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0424 13:35:08.529539 140054653589312 maxtext_utils.py:1889]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0424 13:35:08.529573 140054653589312 maxtext_utils.py:1889]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0424 13:35:08.529606 140054653589312 maxtext_utils.py:1889]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0424 13:35:08.529639 140054653589312 maxtext_utils.py:1889]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0424 13:35:08.529686 140054653589312 maxtext_utils.py:1889]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0424 13:35:08.529721 140054653589312 maxtext_utils.py:1889]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')

I0424 13:35:08.937376 140054653589312 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 13:35:08.937483 140054653589312 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:35:08.943015 140054653589312 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0424 13:35:08.943074 140054653589312 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:35:08.960014 140054653589312 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 13:35:08.960083 140054653589312 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:35:08.976174 140054653589312 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0424 13:35:08.976234 140054653589312 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0424 13:35:09.000497 140054653589312 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:35:09.000571 140054653589312 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:35:09.016922 140054653589312 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:35:09.016989 140054653589312 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:35:09.033062 140054653589312 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0424 13:35:09.033123 140054653589312 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:35:09.064369 140054653589312 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0424 13:35:09.064441 140054653589312 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0424 13:35:09.084805 140054653589312 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0424 13:35:09.084867 140054653589312 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0424 13:35:20.535492 140054653589312 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0424 13:35:20.537880 140054653589312 metric_logger.py:301] number parameters: 1.104 billion
I0424 13:35:31.599198 140054653589312 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0424 13:35:31.751304 140054653589312 checkpointing.py:798] Waited 0.1520988941192627 seconds for step 0 to finish before starting checkpointing.
I0424 13:35:31.753701 140054653589312 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 13:35:31.755568 140054653589312 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0
I0424 13:35:31.757156 140054653589312 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/0.
I0424 13:35:32.178600 140054653589312 signaling_client.py:364] Using JaxDistributedSignalingClient
I0424 13:35:32.179510 140054653589312 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0424 13:35:32.179566 140054653589312 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 13:35:32.456567 140054653589312 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278736s
I0424 13:35:32.456749 140054653589312 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.353 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2881467342376709 s) (per-host)
I0424 13:35:32.456805 140054653589312 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.288214s (batch_requests_ready=0.003369s, total_serialization_initiated=0.284771s, others=0.000075s)
I0424 13:35:32.456908 140054653589312 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.292315s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.292298)
I0424 13:35:32.457721 140054653589312 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.70 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/0.
I0424 13:35:32.458046 139926672946944 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:55:32.458007
I0424 13:35:32.465192 140054653589312 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:35:32.465502 139922316982016 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:35:32.465673 140054653589312 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777037731.753677, 'wait_for_prev_duration_secs': 0.00010824203491210938, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777037731.755606, 'checkpointer_blocking_duration_secs': 0.702603816986084, 'get_old_steps_start_time': 1777037732.458236, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1777037731.751726, 'checkpoint_manager_blocking_duration_secs': 0.713895320892334}
I0424 13:35:32.465782 140054653589312 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0424 13:35:32.465847 140054653589312 max_utils.py:750] 
Memstats: After params initialized:
I0424 13:35:32.465897 140054653589312 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0))
I0424 13:35:32.465930 140054653589312 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0))
I0424 13:35:32.465957 140054653589312 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0))
I0424 13:35:32.465981 140054653589312 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0))
I0424 13:35:32.784738 140054653589312 metric_logger.py:196] completed step: 0, seconds: 11.060, TFLOP/s/device: 1.228, Tokens/s/device: 185.171, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0424 13:35:33.055392 140054653589312 metric_logger.py:196] completed step: 1, seconds: 1.184, TFLOP/s/device: 11.478, Tokens/s/device: 1730.058, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0424 13:35:33.500567 140054653589312 metric_logger.py:196] completed step: 2, seconds: 0.125, TFLOP/s/device: 108.768, Tokens/s/device: 16394.755, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28632.133
I0424 13:35:33.648313 140054653589312 metric_logger.py:196] completed step: 3, seconds: 0.435, TFLOP/s/device: 31.267, Tokens/s/device: 4712.911, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16855.789
I0424 13:35:33.944327 140054653589312 metric_logger.py:196] completed step: 4, seconds: 0.166, TFLOP/s/device: 82.026, Tokens/s/device: 12363.790, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10661.594
I0424 13:35:33.951785 140054653589312 metric_logger.py:196] completed step: 5, seconds: 0.146, TFLOP/s/device: 92.857, Tokens/s/device: 13996.433, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7245.978
I0424 13:35:35.763952    2798 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0424 13:35:37.789326 139923898230528 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/0/items/array_metadatas/process_5
I0424 13:35:56.550802 140054653589312 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.834, Tokens/s/device: 6908.673, total_weights: 65536, loss: 8.589, lm_loss: 8.589, perplexity: 5371.156
I0424 13:35:56.698445 140054653589312 metric_logger.py:196] completed step: 7, seconds: 22.452, TFLOP/s/device: 0.605, Tokens/s/device: 91.215, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4360.465
I0424 13:35:56.846187 140054653589312 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.520, Tokens/s/device: 13342.628, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.256
I0424 13:35:56.994526 140054653589312 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0424 13:35:56.995490 140054653589312 checkpointing.py:798] Waited 0.0009403228759765625 seconds for step 9 to finish before starting checkpointing.
I0424 13:35:56.997686 140054653589312 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:36:09.593748 139926672946944 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 42.207 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.425121784210205 s) (per-host)
I0424 13:36:09.593853 139926672946944 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.135695s.
I0424 13:36:17.711134 139926672946944 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 45.252959s.
I0424 13:36:17.711426 139922316982016 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:36:17.711544 139922316982016 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:36:17.711594 139922316982016 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:36:17.713306 139922316982016 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0424 13:36:17.713465 140054653589312 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0424 13:36:17.713592 140054653589312 checkpoint_manager.py:1452] Waiting for previous save to complete took 20.715922 seconds. If this number is high, consider checkpointing less frequently.
I0424 13:36:17.715553 140054653589312 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9
I0424 13:36:17.717785 140054653589312 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/9.
I0424 13:36:18.293064 140054653589312 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0424 13:36:18.293154 140054653589312 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0424 13:36:18.327512 140054653589312 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036028s
I0424 13:36:18.327680 140054653589312 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 35.832 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.043050289154052734 s) (per-host)
I0424 13:36:18.327730 140054653589312 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043123s (batch_requests_ready=0.002645s, total_serialization_initiated=0.040397s, others=0.000080s)
I0424 13:36:18.327819 140054653589312 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.047204s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.047190)
I0424 13:36:18.328499 140054653589312 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.61 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/9.
I0424 13:36:18.328800 139922316982016 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 13:56:18.328767
I0424 13:36:18.330682 140054653589312 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0424 13:36:18.330957 139904021407488 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0424 13:36:18.331090 140054653589312 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777037756.9976418, 'wait_for_prev_duration_secs': 20.715921878814697, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777037777.715591, 'checkpointer_blocking_duration_secs': 0.6133556365966797, 'get_old_steps_start_time': 1777037778.3289695, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1777037756.9958918, 'checkpoint_manager_blocking_duration_secs': 21.335166454315186}
I0424 13:36:18.331254 140054653589312 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0424 13:36:18.331302 140054653589312 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0424 13:36:23.581378 139923898230528 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_02_synthetic/checkpoints/9/items/array_metadatas/process_5
I0424 13:36:59.949302 139922316982016 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.912 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.66464614868164 s) (per-host)
I0424 13:36:59.949424 139922316982016 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.620512s.
I0424 13:37:08.954156 139922316982016 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.625229s.
I0424 13:37:08.954450 139904021407488 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0424 13:37:08.954567 139904021407488 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0424 13:37:08.954614 139904021407488 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0424 13:37:08.957318 139904021407488 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0424 13:37:08.957493 140054653589312 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0424 13:37:08.957682 140054653589312 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0424 13:37:08.958744 140054653589312 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.809, Tokens/s/device: 13838.492, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3551.438
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 24 13:37:20 UTC 2026
EXIT_CODE=0