Skip to content
r/LocalLLaMA · Communities

New set of FP4 attention kernels for B300, achieving up to 1.69x speedup over FA4

submitted by /u/tuananh_org [link] [comments]

Read original