Offer a half-precision KV cache behind a flag (default off) - #109

Merged

bsbodden merged 2 commits into

mainfrom

fix/f16-kv-cache

Aug 7, 2026

Commits on Aug 7, 2026

fix: store the KV cache at half precision, as llama.cpp does
bsbodden
committed
b1dc082
fix: offer a half-precision KV cache behind a flag, default off
bsbodden
committed
bd3bb4e