Nvidia unveiled its latest series of open large language models (LLMs), named the Llama Nemotron family, on Monday. The tech company emphasized the necessity for more advanced AI models to support the increasing complexity of artificial intelligence agents. By highlighting the need for enhanced power and efficiency, Nvidia asserted that the Nemotron models are equipped to create and deploy AI agents suited for various applications. Enterprises will have access to these models through the Nvidia NIM microservice.
Nvidia Introduces Nemotron Family of AI Models
In a blog post, Nvidia detailed its new open-source LLM series, known as Nemotron. This lineup also includes the Cosmos Nemotron vision language models (VLMs), designed for building AI agents that can analyze and respond to visual content such as images and videos. The company stated that these vision-centric agents could be utilized in various fields, including autonomous machines, healthcare facilities, retail environments, warehouses, sports events, film production, and news reporting.
The Llama Nemotron models are constructed upon Meta’s Llama foundation models and are reported to be optimized for developing AI agents. While specific architectural details were not disclosed, Nvidia noted that these models are trained using advanced techniques alongside high-quality datasets. Capabilities such as instruction following, chat interactions, function calling, coding, mathematics, and more can be developed using these models. Additionally, Nemotron is said to optimize the size of AI agents, facilitating easier deployment.
Nvidia indicated that companies like SAP, ServiceNow, and other providers of AI agent platforms would be among the first to implement the new Llama Nemotron models.
The Nemotron and Cosmos Nemotron models will be offered in three parameter sizes: Nano, Super, and Ultra. The Nano model is designed for cost-effectiveness with low latency priorities. In contrast, the Super model emphasizes high accuracy and can operate on a single GPU. The Ultra model represents the highest-accuracy option and is aimed at data center-scale applications.
According to Nvidia, enterprises will be able to Access the Nemotron model family through downloadable options and the NIM microservice. These models will also be made available via application programming interfaces (APIs). Although the models are classified as open-source, their use will be limited to academic and research purposes.
