What Happened
Independent interviewer Kimmonismus announced they will conduct an in-person interview with Kari Briski, NVIDIA's Vice President of Generative AI Software for Enterprise, this Thursday. According to the announcement, NVIDIA will record the interview and publish it afterward.
The stated topic of discussion will be NVIDIA's Nemotron family of models, with a specific focus on the recent development of Nemotron 3 Super.
Context
Nemotron is NVIDIA's family of open-source large language models designed for commercial use. The models are positioned as alternatives to offerings from Meta (Llama) and Mistral AI, with permissive licensing for enterprise integration and customization.
In May 2024, NVIDIA released the Nemotron 3 Super series, which includes three model sizes (1.5B, 4B, and 8B parameters). The "Super" designation indicates these models are instruction-tuned variants, optimized for following user prompts and performing specific tasks, as opposed to the base Nemotron 3 models which are primarily for continued pre-training or research.
Key technical features of the Nemotron 3 Super models, as per their initial release, include:
- Support for context windows up to 128,000 tokens.
- Training on a 9 trillion token dataset.
- Availability on the NVIDIA NGC catalog and Hugging Face.
Kari Briski leads the team responsible for the software stack and developer tools that enable enterprises to deploy and customize these generative AI models, including the Nemotron family, on NVIDIA's hardware platforms.
The article is based solely on the announcement of an upcoming interview. Specific technical details, benchmarks, or new features regarding Nemotron 3 Super discussed in the interview will be available after NVIDIA publishes the recording.
