Nvidia has come together with ServiceNow and Hugging Face to introduce a new family of open-access large language models (LLMs) for code generation.
The StarCoder2 platform was developed by the BigCode community with performance, transparency, and cost-effectiveness considered.
StarCoder2’s extensive reach is founded on the training of 619 programming languages. The AI code generator comes in three flavors: 3 billion, 7 billion and 15 billion parameters.
StarCoder2 brings code generation to all
According to the announcement, the smaller variants were created to provide strong performance while managing compute costs. The smallest model was built together with ServiceNow and promises to match the performance of the original StarCoder’s 15 billion-parameter option, while the mid-spec option gets the support of Hugging Face.
StarCoder2’s 15 billion-parameter option was trained on Nvidia accelerated infrastructure.
The considerable improvements mean that, while greater performance can be unlocked by the Nvidia-accelerated option, even the most basic variant is a considerable advancement on previous generations, and requires less sophisticated infrastructure.
Jonathan Cohen, VP for Applied Research at Nvidia. “Nvidia’s collaboration with ServiceNow and Hugging Face introduces secure, responsibly developed models, and supports broader access to accountable generative AI that we hope will benefit the global community.”
Moreover, StarCoder2 uses a new code dataset called The Stack v2, which incorporates new training techniques to understand low-resource programming languages, mathematics, and program source code discussions.
Besides performance and efficiency improvements, the organizations insist that StarCoder2 adheres to ethical AI practices, such as using responsibly sourced data under license from the digital commons of Software Heritage. Developers can also opt out of having their data used for training.