Low-bit weight--activation quantization reduces the memory and computation needed to deploy large language models, but activation rounding can still degrade …
机构:中科院
来源:arXiv 2610.35955 | AI4Papers 论文推荐平台