At GTC 2022, NVIDIA announced Riva 2.0, Merlin 1.0, new features to NVIDIA Triton, and more.
Overview
At GTC 2022, NVIDIA unveiled significant updates to its AI software suite, focusing on advancements in speech AI, recommenders, and inference optimization. The updates include the launch of Riva 2.0, Merlin 1.0, Triton enhancements, and new features in the NeMo framework and Maxine SDK.
What You'll Learn
1
How to build real-time speech AI applications using NVIDIA Riva 2.0
2
Why NVIDIA Merlin 1.0 accelerates recommender workflows
3
How to optimize inference deployment with NVIDIA Triton
4
When to use the NVIDIA NeMo framework for training large language models
5
How to enhance audio and video communications with NVIDIA Maxine
Key Questions Answered
What are the key features of NVIDIA Riva 2.0?
NVIDIA Riva 2.0 provides world-class automatic speech recognition in seven languages, neural-based text-to-speech capabilities, domain-specific customization with the TAO Toolkit and NeMo, and support for deployment in cloud, on-prem, and embedded platforms.
How does NVIDIA Merlin 1.0 improve recommender systems?
NVIDIA Merlin 1.0 accelerates recommender workflows by providing an end-to-end framework for data preprocessing, feature transforms, training, optimization, and deployment, allowing data scientists to scale faster with less code.
What new features were introduced in NVIDIA Triton?
New features in NVIDIA Triton include the Triton FIL backend for model explainability, Triton Management Service for fleet management on Kubernetes, and support for inference on AWS Inferentia, enhancing deployment capabilities.
What updates were made to the NVIDIA NeMo framework?
The latest version of the NVIDIA NeMo framework includes a hyperparameter tuning tool, reference recipes for T5 and mT5 models, cloud support for Azure, and distributed data preprocessing scripts to shorten training time.
How does NVIDIA Maxine enhance audio quality?
NVIDIA Maxine includes features like audio super resolution, acoustic echo cancellation, noise removal, and room echo cancellation, all designed to improve the clarity and quality of real-time audio communications.
Technologies & Tools
AI SDK
Nvidia Riva
Used for real-time speech recognition and text-to-speech applications.
ML Framework
Nvidia Merlin
Accelerates recommender workflows and simplifies model deployment.
Inference Server
Nvidia Triton
Facilitates fast and scalable AI inference for applications.
Framework
Nvidia Nemo
Used for training large language models with advanced techniques.
SDK
Nvidia Maxine
Enhances audio and video communication through AI.
Key Actionable Insights
1Leverage NVIDIA Riva 2.0 to create real-time speech applications that can operate in multiple languages, enhancing accessibility and user experience.This is particularly useful for developers targeting global markets or diverse user bases, as it allows for more inclusive applications.
2Utilize NVIDIA Merlin 1.0 to streamline the deployment of recommender systems, significantly reducing the amount of code needed for implementation.This can lead to faster iterations and more efficient workflows for data scientists and machine learning engineers.
3Adopt the new features of NVIDIA Triton to simplify the management of AI inference deployments across various environments, including Kubernetes.This is essential for organizations looking to scale their AI applications efficiently while maintaining performance.