MutaGReP: Execution-Free Repository-Grounded Plan Search for Code-Use
Zaid Khan, Ali Farhadi, Ranjay Krishna, Luca Weihs, Mohit Bansal, Tanmay Gupta
2025-02-25
Summary
This paper talks about MutaGReP, a new way to help AI language models use large code repositories more efficiently when solving coding tasks
What's the problem?
When AI needs to write code using a big code repository, it's hard to give it all the information it needs without overwhelming it. Putting the whole repository into the AI's memory is inefficient and can actually make it perform worse
What's the solution?
The researchers created MutaGReP, which works like a smart search engine for code. It breaks down the coding task into smaller steps and finds the most relevant parts of the code repository for each step. It does this by using AI to explore different plans and match them with the right code symbols, without actually running any code
Why it matters?
This matters because it makes AI coding assistants much more efficient and powerful. MutaGReP allows AI models to perform as well as much larger models while using only a fraction of the memory. It even helps solve really tough coding problems that were previously too hard for AI. This could make AI coding tools more useful and accessible for developers working on complex projects
Abstract
When a human requests an LLM to complete a coding task using functionality from a large code repository, how do we provide context from the repo to the LLM? One approach is to add the entire repo to the LLM's context window. However, most tasks involve only fraction of symbols from a repo, longer contexts are detrimental to the LLM's reasoning abilities, and context windows are not unlimited. Alternatively, we could emulate the human ability to navigate a large repo, pick out the right functionality, and form a plan to solve the task. We propose MutaGReP (Mutation-guided Grounded Repository Plan Search), an approach to search for plans that decompose a user request into natural language steps grounded in the codebase. MutaGReP performs neural tree search in plan space, exploring by mutating plans and using a symbol retriever for grounding. On the challenging LongCodeArena benchmark, our plans use less than 5% of the 128K context window for GPT-4o but rival the coding performance of GPT-4o with a context window filled with the repo. Plans produced by MutaGReP allow Qwen 2.5 Coder 32B and 72B to match the performance of GPT-4o with full repo context and enable progress on the hardest LongCodeArena tasks. Project page: zaidkhan.me/MutaGReP