Zeroth-order (ZO) optimization enables memory-efficient fine-tuning of large language models through forward-only evaluations, but high-dimensional random pe…
机构:HKU
来源:arXiv 2611.16317 | AI4Papers 论文推荐平台