AnalysisAI ModelsJuly 17, 2026

Byte-exact KV-cache grafting improves frozen Gemma 4

On Gemma 4 12B, the method improved AIME 2025 accuracy from 76.7% to 90.0% without weight changes. The approach stores verified knowledge as byte-exact KV state artifacts.

2 sources

Smarter and Cheaper at Once: Byte-Exact KV-Cache Grafting Turns a Frozen Small Model into a Verified-Knowledge Flywheelarxiv.org
Byte exact KV cache grafting on frozen Gemma 4reddit.com

Back to the feed