Preprint
Aug 2026
Unified Lookup-Table Inference with Signed-Digit K/V Caches for Ternary LLMs
This work develops a unified lookup-table inference approach for ternary LLMs that stores runtime K/V states as scaled multi-plane signed digits organized around the reduction structure of attention.
Ziang Duan, Jia-Jun Wu, Ze-Tian Chen et al.
· 0 citations