UVA Research Computing will soon debut a new generative AI (GenAI) platform designed specifically to serve the UVA research community. This central tool will give researchers a way to leverage GenAI capabilities in their research with improved data security, increased efficiency, clear ownership, and reduced costs.
The platform will feature two large language models (LLMs): a mid-powered model (120 billion parameters) and a high-powered model (1 trillion parameters). The LLMs will run on a shared graphics processing unit (GPU) infrastructure, enabling simultaneous inquires of up to 1,000 per hour at no charge.
These models will support a wide range of qualitative and quantitative analyses, including work already being done with other Research Computing services such as analyzing social media sentiment, exploring the context of laws around the commonwealth, and more.
Within Research Computing’s Afton and Rivanna systems, researchers can access the LLMs via a web portal or application programming interface (API). The web portal interface will resemble familiar AI chatbots, allowing users to submit questions or requests in natural language and receive responses generated from the models’ trained data. API access will support more advanced use cases, including complex analyses and large-scale or automated workflows.
This GenAI platform is part of Research Computing’s Early Access program, designed to give full access to new services sooner. During the Early Access phase, the team will complete final optimizations and testing, and tech support will be limited until the full launch.
This project underscores Research Computing’s commitment to research excellence, strong data stewardship, and responsible AI innovation and was guided by best practices demonstrated at peer R1 institutions.
Research Computing is part of UVA Information Technology Services, the central service provider for information technology services across the University with over 100 services.
More information about the new GenAI platform will be available soon.