CodeMonkeys: Scaling Test-Time Compute for Software Engineering

TL;DR

CodeMonkeys is a system that enhances large language models' ability to solve real-world software issues by iteratively editing codebases through combined serial and parallel test-time compute, improving success rates on GitHub issues.

Contribution

We introduce CodeMonkeys, a novel approach that scales test-time compute for software engineering tasks by combining iterative editing with multi-trajectory sampling, achieving significant problem-solving improvements.

Findings

Resolved 57.4% of issues from SWE-bench.

Achieved 66.2% success with ensemble candidate selection.

Demonstrated effective parallel and serial scaling of test-time compute.

Abstract

Scaling test-time compute is a promising axis for improving LLM capabilities. However, test-time compute can be scaled in a variety of ways, and effectively combining different approaches remains an active area of research. Here, we explore this problem in the context of solving real-world GitHub issues from the SWE-bench dataset. Our system, named CodeMonkeys, allows models to iteratively edit a codebase by jointly generating and running a testing script alongside their draft edit. We sample many of these multi-turn trajectories for every issue to generate a collection of candidate edits. This approach lets us scale "serial" test-time compute by increasing the number of iterations per trajectory and "parallel" test-time compute by increasing the number of trajectories per problem. With parallel scaling, we can amortize up-front costs across multiple downstream samples, allowing us to identify relevant codebase context using the simple method of letting an LLM read every file. In order to select between candidate edits, we combine voting using model-generated tests with a final multi-turn trajectory dedicated to selection. Overall, CodeMonkeys resolves 57.4% of issues from SWE-bench Verified using a budget of approximately 2300 USD. Our selection method can also be used to combine candidates from different sources. Selecting over an ensemble of edits from existing top SWE-bench Verified submissions obtains a score of 66.2% and outperforms the best member of the ensemble on its own. We fully release our code and data at https://scalingintelligence.stanford.edu/pubs/codemonkeys.