dinobby/MAgICoRE

23

stars

3

commits

Python

primary language

Sep 19, 2024

updated

README

MAgICoRe: A Multi-Agent Coarse-to-Fine Refinement Framework for Reasoning

Justin Chih-Yao Chen, Archiki Prasad, Swarnadeep Saha, Elias Stengel-Eskin, and Mohit Bansal

image

Installation

This repository is tested on Python 3.10.12. All dependencies can be installed as follows:

pip install -r requirements.txt

Run Experiment

Step 1: use the following script to generate $k$ reasoning chains for each problem.

CUDA_VISIBLE_DEVICES=0,1 python generate.py --k 40

Step 2: use the following script to get RM scores for the first iteration.

CUDA_VISIBLE_DEVICES=0,1 python annotate.py --k 40

Step 3: use the following script to refine hard instances.

CUDA_VISIBLE_DEVICES=0,1 python refine.py --k 40

k corresponds to the number of generations for each problem.

Citation

@article{chen2024magicore,
  title={MAgICoRe: A Multi-Agent Coarse-to-Fine Refinement Framework for Reasoning},
  author={Chen, Justin Chih-Yao and Prasad, Archiki and Saha, Swarnadeep and Stengel-Eskin, Elias and Bansal, Mohit},
  journal={arXiv preprint arXiv:2409.12147},
  year={2024}
}

Contributors

dinobby

3 commits

dinobby/MAgICoRE

23

stars

3

commits

Python

primary language

Sep 19, 2024

updated

README

MAgICoRe: A Multi-Agent Coarse-to-Fine Refinement Framework for Reasoning

Justin Chih-Yao Chen, Archiki Prasad, Swarnadeep Saha, Elias Stengel-Eskin, and Mohit Bansal

image

Installation

This repository is tested on Python 3.10.12. All dependencies can be installed as follows:

pip install -r requirements.txt

Run Experiment

Step 1: use the following script to generate $k$ reasoning chains for each problem.

CUDA_VISIBLE_DEVICES=0,1 python generate.py --k 40

Step 2: use the following script to get RM scores for the first iteration.

CUDA_VISIBLE_DEVICES=0,1 python annotate.py --k 40

Step 3: use the following script to refine hard instances.

CUDA_VISIBLE_DEVICES=0,1 python refine.py --k 40

k corresponds to the number of generations for each problem.

Citation

@article{chen2024magicore,
  title={MAgICoRe: A Multi-Agent Coarse-to-Fine Refinement Framework for Reasoning},
  author={Chen, Justin Chih-Yao and Prasad, Archiki and Saha, Swarnadeep and Stengel-Eskin, Elias and Bansal, Mohit},
  journal={arXiv preprint arXiv:2409.12147},
  year={2024}
}

Contributors

dinobby

3 commits

Languages

Python

100.0%