Skip to content

ggml-cuda: GEMM weight row padding + one-time K-padded f16 dequant for prefill#57

Draft
roberteg16 wants to merge 3 commits into
gfx11from
rogarcia.cuda-gemm-pad-prefill-dequant
Draft

ggml-cuda: GEMM weight row padding + one-time K-padded f16 dequant for prefill#57
roberteg16 wants to merge 3 commits into
gfx11from
rogarcia.cuda-gemm-pad-prefill-dequant