SWE-Dev-32B: An Open-Source Software Engineering Agent
zai-org/SWE-Dev-32B is a 32.8 billion parameter model built as an open-source agent for complex software engineering tasks. Derived from the Qwen-2.5-Coder-32B-Instruct architecture, this model is part of the SWE-Dev family, which also includes 7B and 9B variants. It leverages a comprehensive pipeline for creating developer-oriented datasets from GitHub repositories, focusing on practical applications like issue tracking, code localization, and test case generation.
Key Capabilities & Performance
- Software Engineering Automation: Designed to act as an agent for automating various software development workflows.
- High SWE-bench Performance: Achieved a 36.6% solve rate on the challenging SWE-bench-Verified benchmark, approaching the performance of proprietary models like GPT-4o.
- Data and Inference Scaling: Research indicates that both scaling training data quality (especially with reinforcement fine-tuning) and increasing inference rounds significantly boost performance on SWE-bench. For instance, the solve rate improved from 34.0% at 30 inference rounds to 36.6% at 75 rounds.
Good For
- Automated code generation and bug fixing.
- Generating test cases for software projects.
- Assisting with code localization and issue tracking within GitHub repositories.
- Researchers and developers exploring advanced software engineering agents.