Weight-only post-training quantization (PTQ) relies heavily on reconstruction loss minimization to preserve model quality at low precision. We show that the …
机构:中科大
来源:arXiv 2610.11226 | AI4Papers 论文推荐平台