ricdomolm/mini-coder-4b
The ricdomolm/mini-coder-4b is a 4 billion parameter model distilled from Qwen 3 Coder 30B A3B, specifically optimized for software engineering tasks. It demonstrates strong performance on the SWE-bench Verified Bash only benchmark, outperforming larger models like gpt-oss-120b. This model is designed for efficient code generation and problem-solving within agentic frameworks, making it suitable for automated software development workflows.
Loading preview...
Overview
ricdomolm/mini-coder-4b is a compact yet powerful 4 billion parameter language model, distilled from the larger Qwen 3 Coder 30B A3B. It is specifically engineered for software engineering (SWE) tasks, demonstrating performance that "punches above its weight" by outperforming models like gpt-oss-120b on the SWE-bench Verified Bash only benchmark.
Key Capabilities
- High Performance on SWE Tasks: Achieves a pass@1 score of 26.8 and pass@100 of 60.2 on SWE-bench Verified Bash only, surpassing several larger models.
- Efficient Training: Trained on 400k trajectories using the lightweight mini-swe-agent scaffolding and the SWE-smith dataset.
- Resource-Friendly Fine-tuning: Unlike many agentic SWE models,
mini-coder-4bcan be post-trained on a single 80GB GPU or smaller, making it accessible for more developers. - Seamless Agentic Integration: Designed to work effectively with
mini-swe-agent, a lightweight and scalable agentic framework, facilitating RL fine-tuning.
Good For
- Automated Software Development: Ideal for use in agentic systems that require robust code generation and problem-solving capabilities.
- Resource-Constrained Environments: Its smaller size and efficient training make it suitable for developers with limited GPU resources.
- Research and Development: Provides a strong base for further fine-tuning and experimentation in software engineering AI.