README.md

metadata

pipeline_tag: text-generation
library_name: transformers
license: mit

FETCH: Efficient Tree Search for LLM Reasoning

This repository contains model checkpoints for the FETCH framework, as described in the paper Don't Get Lost in the Trees: Streamlining LLM Reasoning by Overcoming Tree Search Exploration Pitfalls.

The FETCH framework improves the efficiency and accuracy of LLM-based reasoning by addressing over-exploration and under-exploration issues in tree search algorithms.

Pre-trained models mentioned in the paper's README are available on Hugging Face Hub:

• Policy: xmu-nlp/Llama-3-8b-gsm8k
• Verifier: xmu-nlp/Llama-3-8b-gsm8k-value-A and xmu-nlp/Llama-3-8b-gsm8k-value-B
• Embedding: xmu-nlp/simcse-large-gsm8k

For more details and usage instructions, please refer to the GitHub repository.