Stability AI has launched Stable Code 3B, an upgraded three billion parameter AI system designed for automatic code generation and completion. The system offers enhancements like larger context size and improved completion quality to push the boundaries of AI-assisted software development. Despite its smaller size, Stable Code 3B can efficiently run on readily available hardware like laptops, matching or exceeding the code completion quality of larger models.
Stable Code 3B builds upon Stability AI’s Stable LM natural language model with additional training on software engineering data like code repositories and programmer forums. It supports 18 programming languages, including Python, JavaScript, Java, C++, and Go. The model’s training process incorporates Rotary Position Embeddings (RoPE) to expand the context size for improved performance, similar to Meta’s Llama 2 Long. This technique allows for context lengths up to 100k tokens.
In addition to suggesting new lines of code, Stable Code 3B can also fill in large missing sections in existing code, an ability known as Fill in the Middle (FIM). This advanced feature allows it to automatically write entire functions or components.
Stable Code 3B’s performance outshines comparable models like StarCoder and positions Stability AI as a leader in the fast-moving field of AI-generated code. With impressive benchmarks and increased accessibility, Stable Code 3B aims to bring enhanced AI code completion to a broader audience, further accelerating the integration of generative AI into software development workflows across industries.
By automating rote coding tasks, Stable Code 3B enables developers to become more productive, creative, and focus their efforts on more complex challenges. As AI-generated code systems continue to advance, the potential for these technologies to revolutionize software development across industries is undeniable.
