Start-up Hugging Face and ServiceNow Research have launched a free alternative to code-generating AI systems, called “StarCoder”.
Unlike DeepMind’s AlphaCode, Amazon’s CodeWhisperer and OpenAI’s Codex, which powers GitHub’s Copilot, StarCoder is licensed for royalty-free use by anyone, including corporations. The system was trained on more than 80 programming languages and text from GitHub repositories. StarCoder can integrate with Microsoft’s Visual Studio Code code editor and can answer questions about code, following basic instructions. Machine learning engineer Leandro von Werra claims that StarCoder matches or outperforms the AI model from OpenAI used to power initial versions of Copilot.
StarCoder is part of Hugging Face’s and ServiceNow’s BigCode project, which aims to develop “state-of-the-art” AI systems for code in an “open and responsible” way. The project has more than 600 personnel and focuses on subtopics including collecting datasets, implementing training methods for code models, developing an evaluation suite and discussing ethical best practices. BigCode seeks to address controversies around AI-powered code generation, such as the use of public source code by GitHub and OpenAI to train and monetise Codex.
The BigCode project provides a way for developers to “opt out” of The Stack, an initiative similar to efforts elsewhere to let artists remove their work from text-to-image AI training datasets. In machine learning, parameters define the skill of the system on a problem, such as generating code. Releasing a powerful code generation model allows anyone to fine-tune and adapt it to their own use cases and will enable countless downstream applications.
