Summary of Token-supervised Value Models For Enhancing Mathematical Problem-solving Capabilities Of Large Language Models, by Jung Hyun Lee et al.
Token-Supervised Value Models for Enhancing Mathematical Problem-Solving Capabilities of Large Language Models
by Jung Hyun Lee, June Yong Yang, Byeongho Heo, Dongyoon Han, Kyungsu Kim, Eunho Yang, Kang Min Yoo
First submitted to arxiv on: 12 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 research paper presents a new class of verifiers called token-supervised value models (TVMs) to improve the performance of large language models (LLMs) in mathematical problem-solving tasks. The authors highlight the limitations of existing inference strategies, which rely on verifiers designed for Best-of-N search and are sub-optimal for tree search techniques at test time. TVMs assign a probability to each token reflecting its likelihood of reaching the correct final answer, enabling direct evaluation of partial solutions during tree search. Experimental results show that combining tree-search-based inference strategies with TVMs outperforms existing verifiers, achieving improved accuracy in mathematical problem-solving tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps improve how computers solve math problems by creating better “verifiers” for large language models (LLMs). Currently, these LLMs use methods from a field called “Best-of-N search”, which isn’t ideal for solving math problems. The researchers propose a new way to evaluate partial solutions during the problem-solving process, giving promising intermediate steps a chance to be considered. This approach leads to better results when testing LLMs’ ability to solve math problems. |
Keywords
» Artificial intelligence » Inference » Likelihood » Probability » Supervised » Token
