Nvidia revamps its AI software platforms to advance speech, recommender systems and inference
Nvidia Corp. today announced a major upgrade to its Nvidia AI and Nvidia AI Enterprise platforms, with new features designed to advance workloads such as speech, recommender systems, hyperscale inference and more again.
The improvements were announced at Nvidia CG 2022 alongside the launch of the new Nvidia AI Accelerated Programwhich aims to guarantee the performance and reliability of artificial intelligence applications built from the platform.
the Nvidia AI Platform is a suite of tools comprising SDKs and AI frameworks that developers can use to design, deploy, manage and scale AI across multiple nodes to power complex training workloads, inference and machine learning.
One of its key components is Nvidia Triton, an open-source hyperscale model inference solution that now includes a model browser to help accelerate the deployment of optimized AI models, a management service for efficient scale in Kubernetes and a forest inference library to enable tree-based model inference.
Another key component, Nvidia Riva 2.0, has been updated. Riva is a speech AI SDK that includes various pre-trained models with high recognition rates that enable twice the accuracy of generic speech recognition services, Nvidia said. The latest version offers voice recognition in seven languages, deep learning-based text-to-speech for both male and female voices, and custom tuning with Nvidia TAO Toolkit.
The platform also comes with the latest version of Nvidia NeMo Megatron 0.9, which is a framework for training large language models, and Nvidia Merlin 1.0, a brand new component that Nvidia describes as an “end recommendation framework”. end-to-end” for building high-performance recommender systems. Finally, Nvidia AI now comes with Nvidia Maxine, an audio and video quality enhancement software development kit that enables real-time communication with AI.
Nvidia said a number of high-end customers are using Nvidia AI, which is available starting today. They include Snap Inc., the creator of the Snapchat messaging platform.
“On Snapchat, our community plays with Lenses more than 6 billion times a day,” said Alan Bekker, Head of Conversational AI at Snap. “Snap is using Nvidia Riva to power our AI-powered voice capabilities and bring them to Lens Studio creators to create a next generation of compelling AR experiences.”
As for Nvidia AI Enterprise, the professional version of the Nvidia AI platform, it is now optimized, certified and supported on all major data center and cloud infrastructure platforms, the company said. The last version, Nvidia Enterprise 2.0is now supported on Red Hat OpenShift and VMware vSphere with Tanzu.
Additionally, Nvidia AI Enterprise 2.0 supports more AI software containers for improved training and inference. For example, support for the Nvidia TAO toolkit means developers will be able to refine pre-trained AI models and customize them more easily, even if they lack expertise in AI or training data.
Nvidia said the latest version of Nvidia AI Enterprise has already been put to good use by NTT Communications, a subsidiary of Nippon Telegraph and Telephone Corp., to accelerate its research and development of natural language processing applications and intelligent video analysis.
“Many of our application developers are now using accelerated computing and need an internal infrastructure that provides an easy-to-use and cost-effective GPU-enabled environment,” said Shoichiro Henmi, Technology Division Manager, Innovation Center , at NTT Communications. “We are confident that Nvidia AI Enterprise will provide an ideal solution as an AI-enabled platform to support large scale development in our VMware vSphere, Kubernetes and cloud infrastructure.”