Agent-Based Code Repair

Context

Large language models (LLMs) have became popular over the last few years, one of the reason being the quality of the outputs these models generate. LLM Agents are a step further, they allow LLMs to use memory, tools and sequential thinking.

Motivation

• Can we utilize LLM Agents for reasoning about code, to repair it for instance?
• Does additional feedback from tools that agents can use improve the quality of repairs?

Goal

The student will follow 3 steps:

1. Review about Code Agents
2. Practical direction: Code Repair, Vulnerability Repair, Code Generation
3. Experiments and analysis of results where the second step is to be discussed further

Requirements

• Knowledge of Machine Learning, PyTorch or TensorFlow
• Passion for learning about state-of-the-art methods and models

Pointers

•  https://arxiv.org/abs/2401.07339
•  https://huggingface.co/blog/smolagents
•  https://github.com/All-Hands-AI/OpenHands