NVIDIA TAO Toolkit
Last updated
Last updated
The NVIDIA TAO Toolkit built on TensorFlow and PyTorch, uses the power of transfer learning while simultaneously simplifying the model training process and optimizing the model for inference throughput on the target platform. The result is an ultra-streamlined workflow. Take your own models or pre-trained models, adapt them to your own real or synthetic data, then optimize for inference throughput. All without needing AI expertise or large training datasets.
Only available for enterprise customers
As this integration uses GPU hours for training, this integration is only available for enterprise customers. View our pricing for more information.
Check out NVIDIA's documentation for information on getting started as a first-time user with the TAO Toolkit.
Now, clone one of the following GitHub repositories in order to bring your TAO model into your Edge Impulse enterprise organization and projects:
Then, follow the instructions in the README.md of the respective repo to integrate and run the pipeline locally or pushed to your Edge Impulse organization.
The block is now available under any of your projects via Create impulse > Add new learning block.
To use a different TAO model you can modify one of the example repositories in the previous step.
If the model is available in the 'Image Classification (PyT)' or 'Image Classification (TF1)' applications, you just need to change the specs file.
If your model is available in another application, then:
Modify the Dockerfile to pull from the right container.
Modify dataset-conversion/ei_to_tao_image_classification.py to do the dataset conversion, and write out a valid specs file.
Modify run.sh to call the correct TAO runtime commands.
With everything setup you can now build your machine learning model with these tutorials:
