iamPi/mini-coder-1.7b
iamPi/mini-coder-1.7b is a 1.7 billion parameter model distilled from Qwen 3 Coder 30B A3B, designed for code generation and software engineering tasks. It demonstrates performance exceeding larger models like SWE-agent-LM 7B on SWE-bench Verified Bash only. This model is optimized for agentic workflows, particularly with the mini-swe-agent framework, and supports a 32768 token context length.
Loading preview...
Overview
mini-coder-1.7b is a compact yet powerful 1.7 billion parameter language model, distilled from the larger Qwen 3 Coder 30B A3B. It is specifically engineered for software engineering (SWE) tasks, showcasing impressive capabilities for its size.
Key Capabilities & Performance
This model excels in code-related problem-solving, particularly within agentic frameworks. It has been trained on 400,000 trajectories using the lightweight mini-swe-agent scaffolding and the SWE-smith dataset of GitHub issues. Notably, mini-coder-1.7b outperforms the 7B parameter SWE-agent-LM on the SWE-bench Verified Bash only benchmark, achieving a pass@1 score of 18.6 compared to 15.2.
Unique Differentiators
Unlike many existing agentic SWE models, mini-coder-1.7b is a dense model, benefiting from a more mature fine-tuning ecosystem. Its smaller size allows for post-training on a single 80GB GPU or smaller, making it highly accessible for developers. It integrates seamlessly with mini-swe-agent, providing a scalable and developer-friendly framework for tasks like generating SWE-bench trajectories locally using inference engines such as vLLM.