MAD-OPD: Breaking the Ceiling in On-Policy Distillation via Multi-Agent Debate
MAD-OPD: 突破基于策略的蒸馏的天花板:多智能体辩论
机构 * School of Artificial Intelligence and Automation, Huazhong University of Science and Technology(华中科技大学人工智能与自动化学院) ; Alibaba Group(阿里巴巴集团)
AI总结 MAD-OPD通过多智能体辩论机制突破单教师限制,提出多教师辩论框架和基于任务的分歧原理,提升代理任务和代码生成性能。
Comments Preprint. 9-page main paper + appendix. 8 figures, 7 tables. Code: https://github.com/chiefovoavicii/MAD-OPD