Feature request

Support LLM-guided Self-Refinement MCTS inference method. It has the following features:

• LLM-as-Judge to provide review
• Proposer LLM generates rewriting of the answer, taking the review into consideration
• Perform self-refinement
• Utilized in various forms across research projects on LLM reasoning.

Motivation

more diverse exploration in tree search

Your contribution

Submitting a PR