Post-training quantization (PTQ) to FP4 has emerged as a key technique for reducing large language model inference cost, especially on Blackwell GPUs, which …
机构：MIT
来源：arXiv 2610.32243 | AI4Papers 论文推荐平台