Showing cs.LGShow all
2 papers · 1 filter
cs.LG2026
Understanding Diversity Collapse in RLVR via the Lens of Overtraining
Suqin Yuan, Jinkun Chen, Jiyang Zheng +6
Reinforcement learning with verifiable rewards (RLVR) has become a key approach for enhancing the reasoning abilities of large language models. However, RLVR often suffers from \em…
cs.LG2026
Mitigating Mismatch within Reference-based Preference Optimization
Suqin Yuan, Xingrui Yu, Jiyang Zheng +4
Direct Preference Optimization (DPO) has become the de facto standard for offline preference alignment of large language models, but its reliance on a reference policy introduces a…