Nvidia Launches New Face Hug Inference Service, AI Industry Solutions

Unveiled at SIGGRAPH 2024, the AI ​​computing giant’s work powers Hugging Face’s new Inference-as-a-Service and helps developers bring generative AI capabilities to industrial sectors like manufacturing and robotics via new microservices.

Nvidia continued its efforts to simplify AI application development with a new inference-as-a-service offering via the Hugging Face model repository and the introduction of new microservices for industrial generative AI applications.

At the SIGGRAPH 2024 conference in Denver on Monday, the AI ​​computing giant announced that the new Hugging Face service will run on Nvidia’s DGX Cloud and build on its inference microservices to help developers “rapidly deploy” popular large language models like Meta’s Llama 3 family and Mistral’s AI model suite.

(Related: 12 Big Announcements From Nvidia, Intel, and AMD at Computex 2024)

Officially known as Nvidia NIM, these microservices consist of AI models packaged in optimized containers that developers can integrate into their applications. The company first introduced them in early June, offering support for more than 40 models developed by Nvidia and others.

NIM is available to businesses through the Nvidia AI Enterprise software suite, which costs $4,500 per GPU per year. It is free for members of the Nvidia Developer Program.

According to the Santa Clara, California-based company, Hugging Face’s new inference service enables developers to “rapidly prototype using open-source AI models hosted on the Hugging Face Hub and deploy them to production.”

The new Hugging Face service complements the Train on DGX Cloud model repository service that was announced with Nvidia a year ago at SIGGRAPH 2023.

New NIM microservices enable industrial GenAI use cases

Nvidia also announced new NIM microservices that will help developers bring generative AI capabilities to industry sectors like manufacturing and robotics.

These include what it calls the “world’s first generative AI models for development OpenUSD,” referring to the open 3D framework developed by Pixar that Nvidia uses to connect its Omniverse platform to other 3D applications.

Available now or in preview soon, the first batch of OpenUSD NIM libraries enable developers to search OpenUSD, 3D, and image data libraries using text or images; generate realistic materials for 3D objects; assemble OpenUSD-based scenes using text prompts; and increase the resolution of physical simulations using AI-based scaling, among other things.

Nvidia has also released a data connector between the Unified Robotics Description Format and OpenUSD, which is intended to help developers move robotics data between design, simulation, and reinforcement learning applications, among others. The company also gives developers the ability to develop their own OpenUSD data connectors through the new OpenUSD Exchange developer toolkit.

“Until recently, digital worlds were primarily used by creative industries; now, with the advancements and accessibility that NVIDIA NIM microservices bring to the OpenUSD platform, industries of all kinds can build physically-based virtual worlds and digital twins to drive innovation while preparing for the next wave of AI: robotics,” Rev Lebaredian, vice president of Omniverse and simulation technologies at NVIDIA, said in a statement.