fix(qwen35moe): size KV reservation by n_full + cache-type, not n_layer×f16#454

davide221 merged 1 commit into

Luce-Org:mainLuce-Org/lucebox-hub:mainfrom

dusterbloom:ship/qwen35moe-kv-reservationdusterbloom/lucebox-hub:ship/qwen35moe-kv-reservationCopy head branch name to clipboard