Migrate from Nvidia TAO (Train, Adapt, Optimize)

The TAO Toolkit (short for Train, Adapt, Optimize) is a low-code AI model development framework that helps to build, fine-tune, and deploy deep learning models, without writing complex training code from scratch. It is part of NVIDIA’s DeepStream and TensorRT ecosystem and is designed mainly for computer vision and speech AI tasks.

Common model formats used by TAO at various stages of model development are:

Stage	Format	Description
Training	.tlt	TAO internal model checkpoint
Export	.etlt	Encrypted TensorRT-ready model
Deployment	.engine	TensorRT optimized inference engine
Optional	.onnx	Open format for interoperability (optional export)

To migrate from TAO, you need to export the model to ONNX, a format supported by Open Edge Platform components. You cannot use .tlt directly. You do it by simply including the –onnx flag in the tao export command:

tao export <task> --onnx_file output_model.onnx

Some models may not allow ONNX export. This regards all encrypted models (.etlt is a proprietary format meant for TensorRT deployment only), as well as the ones with NVIDIA’s proprietary backbones.