fix(cuda): turbo KV fused-MMA — use_sparse off-by-one and missing SMEM swizzle - #12

fix(cuda): turbo KV fused-MMA — use_sparse off-by-one and missing SMEM swizzle#12

wilky2005 wants to merge 2 commits into

thecodacus:perfthecodacus/llama.cpp:perffrom

wilky2005:fix/turboquant-fused-mmawilky2005/llama.cpp:fix/turboquant-fused-mmaCopy head branch name to clipboard