Skip to content

Latest commit

 

History

History
65 lines (37 loc) · 1.72 KB

README.md

File metadata and controls

65 lines (37 loc) · 1.72 KB

Thinking-Flow-MATH

Introduction

We are a team of Netmind.AI, our website is also Netmind.AI

This repository store the netmindai competition project in the AAAI2024 COMPETITION ON MATH PROBLEM SOLVING - TRACK2 leaderboard.

Using guidance:

Step 1: Prepare the enviroment:

git clone https://github.com/protagolabs/Thinking-Flow-MATH
cd Thinking-Flow-MATH 
pip install -r requirements.txt

Step 2: Set your own OpenAI-Key

You should follow the OpenAI website to set the OpenAI-Key, before you using any code from this repository.

If you just want to get all the answer from this project. Before you run the solving_competition.sh, please complete the openai-key in scripts/solve_competition.py line: 12-13

Step 3: Run the script to ght all the answer from this competition:

cd Thinking-Flow-Math 
bash solving_competition.sh

Some Using example:

You want to using this repo in jupyter:

See the notebook: Thinking-Flow-MATH/scripts/solve_math_question.ipynb

For the no-coding using:

(TODO) We also build a webapp demo by streamlit: Thinking-Flow-MATH/streamlit_webapp/thinking_flow_netmindai.py

Explain for this project

Multi-Agents System for MATH-SOP (Thinking-FLow)

(We also have a paper(under reviewing now) for this Thinking-FLow. But the prompts in this paper is old version. If you want to know more detail about this project, please read this paper.) We construct a multi-agent system using a large language model-based agent. The agent system includes:

  1. Plan—Agent
  2. Solving-Agent
  3. MathCoding-Agent
  4. Summary-Agent
  5. Evaluate-Agent

And for 1-4, we also have another correction-version for each agents.

External tool, Retrieval-Module: The description: