Introducing Stable Code Instruct 3B
Key Takeaways
Stable Code Instruct 3B is an instruction-tuned Code LM based on Stable Code 3B. With natural language prompting, this model can handle a variety of tasks such as code generation, math and other software development related queries.
This model provides state of the art performance at the 3B scale and outperforms larger size models such as CodeLlama 7B Instruct and performs comparably with StarChat 15B in software engineering related tasks.
The model is now available with a Stability AI Membership for commercial use. The weights and code for Stable Code Instruct 3B are now available on Hugging Face.
For more details, view our technical report here.
Stable Code Instruct 3B is our latest instruction-tuned large language model, built on top of Stable Code 3B. This model enhances code completion and supports natural language interactions, aiming to improve the efficiency and intuitiveness of programming and software development related tasks. Our analysis suggests that Stable Code Instruct 3B outperforms comparable models such as Codellama 7B Instruct, and DeepSeek-Coder Instruct 1.3B in various coding-related tasks. Details of these findings are available in our technical report. The model is now available with a Stability AI Membership for commercial use. The weights and code for Stable Code Instruct 3B are now available on Hugging Face.
Capabilities
Based on insights from the Stack Overflow 2023 Developer Survey, Stable Code Instruct 3B focuses on languages such as Python, Javascript, Java, C, C++, and Go, which are among the most popular and impactful for a diverse range of developers. While these languages were selected as the primary focus for training, the model also includes training on additional broadly adopted languages such as SQL, PHP, and Rust as detailed in our technical report.
Stable Code Instruct 3B delivers strong test performance even in languages that were not initially included in the training set, such as Lua. This proficiency may stem from its understanding of the underlying coding principles and its ability to adapt these concepts across diverse programming environments, leveraging the inherent predictability of coding tasks.
Stable Code Instruct 3B is not only proficient in code generation but also in FIM (Fill in the Middle) tasks, database queries, code translation, explanation, and creation. Its instruction tuning enables it to comprehend and act on nuanced instructions, facilitating a broad spectrum of coding tasks beyond simple code completion including significant abilities in mathematical understanding, logical reasoning, and processing of complex technical narratives surrounding software development.
Performance
In a direct comparison with leading models like Codellama 7B Instruct and DeepSeek-Coder Instruct 1.3B, Stable Code Instruct 3B offers superior performance in a range of coding tasks. Our testing, as outlined in the technical report, shows that Stable Code Instruct 3B not only matches but often surpasses these models in code completion accuracy, understanding of natural language instructions, and versatility across different programming languages.
In the above graphs, we compare all three models using the Multi-PL benchmark to evaluate the strength of outputs generated in various languages. As a result, we have found Stable Code Instruct 3B outperformed CodeLlama notably in all languages despite its lower parameter count.
Technical Details
Key technical aspects of Stable Code Instruct 3B include:
Language Support: While primary attention is given to major programming languages such as Python, C++, and JavaScript, the model's training dataset encompasses a broad spectrum of languages. To see the full list of languages the model was trained on, see our technical report.
Instruction Following: The instruction tuning of this model allows it to interpret and execute complex and nuanced requests. This capability allows for Stable Code Instruct 3B to deliver more relevant and context-aware code suggestions, enhancing productivity for developers and enabling them to iterate and improve on outputs
Hardware Accessibility: The compact size and low-hardware-requirement of Stable Code Instruct 3B makes it accessible to a broader audience. It supports code completion, FIM tasks, and understands nuanced programming queries, enabling developers to work more efficiently and effectively with less effort.
Stable Code Instruct 3B can be used now for commercial purposes with a Stability AI Membership. For non-commercial use, you can download the model weights on Hugging Face and view our technical report here.
To stay updated on our progress follow us on Twitter, Instagram, LinkedIn, and join our Discord Community.
