In the latest version, we introduced several exciting new features, including source code availability with the Enterprise license.
Release highlights:
- Automatically fine-tune hyperparameters with AutoML
- Experience turnkey deployment of TAO into various cloud services – Azure ML, Google Vertex AI, Azure Kubernetes Services (AKS), Amazon EKS, and more
- Integrate TAO with third-party MLOPs services - W&B and ClearML
- Experience TAO on Google Colab
- Complete access of TAO source code and model weights for the pretrained models without encryption using NVIDIA AI Enterprise software suite
- New transformer-based vision models (CitySemSegformer, Peoplenet Transformer) and pretrained models for retail and smart city tasks (Retail Object Detection, Retail Object Recognition, and Person ReIdNet)
- Support for newer GPUs - H100, L40 and L4
The latest version can be downloaded here
Read dev news, blogs, and other resources
- Dev news : Boost AI Development with the new NVIDIA TAO Toolkit
- Dev Blog: Train like a pro using AutoML in TAO
- Dev Blog: Creating custom AI models using TAO Toolkit with Azure ML
- Jupyter notebooks and startup scripts
- Experience TAO on Google Colab
- Running TAO models in DeepStream application (GitHub)
- Serving TAO models in Triton service (GitHub)