A arxiv.org Matrix-vector multiplication implemented in off-the-shelf DRAM for Low-Bit LLMs 230 points cpldcpu 1 year ago 53 comments