Nvidia launches Llama Nemotron AI models for developers

Nvidia has launched a new series of AI models called Llama Nemotron at its GPU Technology Conference on Tuesday. These models focus on reasoning and are designed for advanced artificial intelligence workflows. The company aims to help developers and businesses create AI agents that can work both independently and as part of a team to tackle complex tasks. The Llama Nemotron models build on Meta's Llama 3 series, with Nvidia enhancing them through post-training improvements. As a result, these models show better performance in areas like mathematical reasoning, coding, and decision-making. Nvidia claims the new models are up to 20 percent more accurate and can perform five times faster than similar open-source models. There are three versions of the Llama Nemotron models: Nano, Super, and Ultra. The Nano model is ideal for edge-based tasks, while the Super model offers a balance of accuracy and processing power. The Ultra model is designed for high-demand environments, such as multi-GPU servers. Nvidia used the DGX Cloud and specially generated synthetic data for the post-training of these models. The company has also shared its tools and methods with the open-source community. In partnership with enterprises like Microsoft and SAP, Nvidia is making these models available for integration into various applications. Businesses can access the Nano and Super models through Nvidia's platform and on Hugging Face. The models come under the Nvidia Open Model License Agreement, allowing both research and commercial use.