Low-bit quantization reduces the memory footprint and inference cost of large language models (LLMs), yet accurate FP4 quantization remains challenging. Prac…
机构:华为
来源:arXiv 2611.16426 | AI4Papers 论文推荐平台