Today, NVIDIA announced the general availability of the latest version of the TAO Toolkit. As a low-code version of the NVIDIA Train, Adapt and Optimize (TAO) framework, the toolkit simplifies and accelerates the creation of AI models for speech and vision AI applications.
With TAO, developers can use the power of transfer learning to create production-ready models customized and optimized for many use-cases. These include detecting defects, translating languages, or managing traffic—without the need for massive amounts of data.
This version boosts developer productivity with new pretrained vision and speech models. It also includes key new features such as ONNX model weights import, REST APIs, and TensorBoard integration.
Download TAO Toolkit 3.22.05 >>
Release highlights
Deploy TAO Toolkit as-a-Service with REST APIs: Build a new AI service or integrate into an existing one with REST APIs. You can manage and orchestrate the TAO Toolkit service on Kubernetes. With TAO Toolkit as-a-service IT managers can deliver scalable services using industry-standard APIs.
Bring your own model weights: Fine-tune and optimize your non-TAO models with TAO. Import pretrained weights from ONNX and take advantage of TAO features like pruning and quantization on your own model. This is supported for image classification and segmentation tasks.
Visualize with TensorBoard: Understand your model training performance by visualizing scalars such as training and validation loss, model weights, and predicted images in TensorBoard. Compare results between experiments by changing hyperparameters and choose the one that best fits your needs.
Pretrained models: Pretrained models speed up the customization process for you to fine-tune through the power of transfer learning, with less data.
Some of the new pretrained models in this latest version can:
- Apply data gathered from LIDAR sensors for robotics and automotive applications.
- Classify human actions based on human poses that can be used in public safety, retail, and worker safety use cases.
- Estimate keypoints on humans, animals, and objects to help portray actions or simply define the object shape.
- Create custom voices with just 30 minutes of recorded data to power smart devices, game characters, and quick service restaurants.
Enterprise support for TAO Toolkit is available with NVIDIA AI Enterprise, an end-to-end software suite for AI development and deployment. This new release of TAO Toolkit will be included in the next quarterly update to NVIDIA AI Enterprise.
Get started
- Visit the TAO Toolkit Getting Started page for instructional videos and quick start guides.
- Download the computer vision or conversational AI model from NGC.
- Watch demo videos of new features including Rest APIs, TensorBoard, and pretrained models.
Solutions using TAO Toolkit
- Develop and deploy AI-powered Robots with NVIDIA Isaac Sim and NVIDIA TAO.
- Read the partner blog from LexSet on how to use their platform to generate synthetic data and fine-tune object detection and segmentation models with the TAO Toolkit.
- Detecting animal health and phenotypes with the TAO Toolkit. Sign up for an upcoming webinar to learn more.
- RocketBoots improves inference performance in automated workforce management.
- Transform development efforts in a retail action recognition use case.