ServiceNow, Hugging Face, and NVIDIA Collaborate to Launch Open-Access LLMs for Code Generation

ServiceNow, Hugging Face, and NVIDIA have joined forces to introduce StarCoder2, a family of open-access large language models (LLMs) designed for code generation. This collaborative effort aims to set new standards in performance, transparency, and cost-effectiveness within the realm of generative AI.

Empowering Developers with Enhanced Code Generation Capabilities

StarCoder2, developed by the BigCode community, a collaboration stewarded by ServiceNow and Hugging Face, offers developers a suite of advanced tools for code generation. Trained on a vast array of programming languages, StarCoder2 models can be further customized and integrated into enterprise applications to perform specialized tasks such as source code generation, workflow creation, text summarization, and more. These capabilities are poised to accelerate innovation and improve productivity across various industries.

Model Variants and Performance

StarCoder2 is available in three model sizes, each offering powerful performance tailored to different compute requirements. These variants include a 3 billion-parameter model developed by ServiceNow, a 7 billion-parameter model by Hugging Face, and a 15 billion-parameter model by NVIDIA, leveraging NVIDIA NeMo and trained on NVIDIA accelerated infrastructure. Despite their varying sizes, these models deliver impressive performance, with the 3 billion-parameter model matching the capabilities of the original 15 billion-parameter model.

Advancing Responsible AI Practices

The collaborative effort underscores a commitment to responsible AI practices, ensuring transparency and accountability in model development. Harm de Vries, lead of ServiceNow's StarCoder2 development team, highlighted the significance of open scientific collaboration and ethical data practices in advancing generative AI capabilities.