Nvidia announces raft of 'NIMs' to speed up Gen AI apps
Nvidia has significantly expanded its NIM ecosystem, growing to over a hundred pre-packaged microservices that streamline the deployment of optimized AI models. By containerizing both commercial and open-source models, Nvidia enables developers to easily integrate capabilities like retrieval-augmented generation and speech recognition into applications. This expansion supports diverse industries, from robotics to digital biology, and integrates with industry standards like Open USD to facilitate better 3D simulation and collaboration tools across teams. A key innovation is the partnership with Hugging Face to offer "Inference-as-a-service," which runs on Nvidia’s cloud infrastructure. This service dramatically improves performance, allowing models like Meta’s Llama 3.1 to operate up to five times faster than on standard hardware. While the service currently focuses on Nvidia’s curated NIMs, it demonstrates how optimized software layers can maximize the efficiency of underlying open-source and partner models, reducing the friction typically associated with running complex AI workloads. This development is highly relevant to the open data and open source communities because NIMs provide a standardized, production-ready framework for deploying open-source AI models. By packaging models with appropriate commercial licenses and optimizing them for specific hardware, Nvidia lowers the barrier to entry for organizations wanting to utilize open-source AI without managing complex infrastructure. It highlights a trend where open models gain enterprise viability through commercial optimization, making them more accessible and efficient for widespread industrial application while maintaining their open origins.
Source: zdnet.comPublished on 2024-07-31
Related news
- Nvidia Powers New Hugging Face Inference Service, AI Industrial Solutions
- Zuck got so excited talking open platforms and AI with Jensen Huang that he dropped a big old F-bomb
- A new White House report embraces open-source AI
- Apple skips Nvidia's GPUs for its AI models, uses thousands of Google TPUs instead
- Beware of AI 'model collapse': How training on synthetic data pollutes the next generation