NVIDIA releases the latest version of Tao toolkit to further simplify and accelerate AI model creation

today ,NVIDIA Announce the official release of the latest version of TAO tool kit . As NVIDIA Training 、 Adapt and optimize (TAO) Low code version of the framework , The toolkit simplifies and accelerates voice and vision AI Application's AI Model creation . 

With the help of TAO, Developers can use the power of migration learning to create production ready models that are customized and optimized for many use cases . This includes detecting defects 、 Translate languages or manage traffic , Without a lot of data . 

This version improves the efficiency of developers through new pre training visual and speech models . It also includes key new features , for example ONNX Model weight import 、REST API and TensorBoard Integrate . 

Release highlights

- Use REST API Deploy TAO Toolkit as a service ：  Building new AI Service or use REST API Integration into existing services . You can go to Kubernetes Manage and arrange TAO Toolkit service . With the help of TAO Toolkit Namely service ,IT Managers can use industry standards API Provide scalable services .

- Built in model weights ： Use TAO Fine tuning and optimization are not TAO Model . from ONNX Import pre training weights , And use it on your own model TAO function , For example, pruning and quantification . This supports image classification and segmentation tasks .

- Use TensorBoard Visualizing ： By means of TensorBoard Visualizing scalars in （ Such as training and verification losses 、 Model weights and prediction images ） To learn about your model training performance . Compare the results between experiments by changing the super parameters , And choose the one that best suits your needs . 

- Pre training model ： The pre training model speeds up the customization process , You can fine tune by transferring the power of learning , Use less data at the same time . 

Some new pre training models in this latest version can ： 

• Apply data collected from lidar sensors to robotics and automotive applications .
• It can be used for public safety 、 Human body postures for retail and worker safety use cases classify human body movements .
• Estimated human 、 Key points of animals and objects , To help depict actions or simply define the shape of objects .  
• just 30 Minutes of recorded data to create a custom sound , For smart devices 、 Game characters and fast food restaurants provide motivation .

About NVIDIA TAO Toolkit learning , You can refer to this official link ：

https://developer.nvidia.com/tao-toolkit-get-started?ncid=so-yout-102946&sfdcid=EM20#nvidia-tao

In this video ,NVIDIA Show how to use TAO Toolkit stay TensorBoard Visual model training progress in .