▲ 2 pointsMXFP8 GEMM: Up to 99% of cuBLAS Performance Using CUDA and PTXdanielvegamyhre.github.ioby matt_d·4mo ago·0 comments·view on hn ↗