List your company

Paged Decode Attention Serving on Uniform B8 trace

1.053GPU RatioPersistentKV
1.000351.0266751.0531.079325Jun 25, 2026

Evaluation results

MethodLinks
PersistentKVSelected route=PersistentKV workqueue, split 24, Page size=16, Hq=32, Hkv=8, d=128, Precision=FP162026.061.0531.0823.0522.609