Summary of Wpo: Enhancing Rlhf with Weighted Preference Optimization, by Wenxuan Zhou et al.
WPO: Enhancing RLHF with Weighted Preference Optimization
by Wenxuan Zhou, Ravi Agrawal, Shujian Zhang, Sathish Reddy Indurthi, Sanqiang Zhao, Kaiqiang Song, Silei Xu, Chenguang Zhu
First submitted to arxiv on: 17 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel strategy to align large language models (LLMs) with human values through reinforcement learning from human feedback (RLHF). The approach, called Weighted Preference Optimization (WPO), addresses the distributional gap problem in off-policy preference optimization by reweighting preference pairs to resemble on-policy data. WPO enhances the optimization process without additional costs and outperforms Direct Preference Optimization (DPO) on instruction following benchmarks like Alpaca Eval 2 and MT-bench. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps make large language models more useful for people by using feedback from humans. It’s a way to align these models with human values, making them better at doing what we want them to do. The method is called Weighted Preference Optimization (WPO) and it helps fix a problem that happens when we use preference data collected from other models. WPO makes the optimization process work better without needing more resources. |
Keywords
» Artificial intelligence » Optimization » Reinforcement learning from human feedback » Rlhf
