ggml-cuda: pad GEMM weight rows to avoid cache-set aliasing#34
Draft
roberteg16 wants to merge 1 commit into
Draft
ggml-cuda: pad GEMM weight rows to avoid cache-set aliasing#34roberteg16 wants to merge 1 commit into
roberteg16 wants to merge 1 commit into
