We revisit Kashin-decomposition-based weight quantization for large language models and propose an improved algorithm with stronger convergence properties an…
机构:中科院
来源:arXiv 2609.11687 | AI4Papers 论文推荐平台