Build Applications Faster Using Pre-Trained AI Models

June 13, 2022

The NVIDIA TAO (train, adapt, optimize) Toolkit is getting more components and models.

Check out more Technology Talks videos.

NVIDIA's TAO (train, adapt, optimize) Toolkit is designed to streamline development and integration of artificial-intelligence (AI) and machine-learning (ML) models from a range of frameworks like PyTorch and TensorFlow. The latest release of this low-code platform includes new and updated pre-trained vision and speech models.

The toolkit lets developers import ONNX model weights to take advantage of its ability to prune and apply quantization to optimize models (Fig. 1). This feature is supported for image classification and segmentation models.

1. TAO now allows developers to bring in their own model weights (1) and use the toolkit to train, adapt and optimize you model (2). A REST interface can be used to integrate the framework with applications in the cloud (3).

Support for TensorFlow's TensorBoard visualization tool has been added to TAO. This visualization can apply to metrics to highlight validation loss and accuracy as well present the model graph.

It's now possible to deploy the TAO Toolkit-as-a-Service (TaaS) using REST APIs. Developers are able to build AI services and integrate TAO into existing applications and services. It also can orchestrate TAO Toolkit services on Kubernetes. The enterprise support for TAO Toolkit is part of NVIDIA AI Enterprise, an end-to-end software suite for AI development and deployment.