This project contains the source code to the TAO launcher interface. This launcher interface aims at providing a unified command line experience to the Transfer Learning Toolkit package. The DNN's in TAO may be implemented in TensorFlow, Keras or PyTorch. These frameworks are difficult to maintain in the same docker. In an attempt to abstract the final customers of TAO, to handle tao command abstracts these details away from the user.
This section covers a quick guide to start working with the launcher.
| Software | Version |
|---|---|
| python | >= 3.6.9 |
| docker-ce | >19.03.5 |
| docker-API | 1.40 |
| nvidia-container-toolkit | >1.3.0-1 |
| nvidia-container-runtime | 3.4.0-1 |
| nvidia-docker2 | 2.5.0-1 |
| nvidia-driver | >450 |
-
Install docker-ce by following the official instructions.
Once you have installed docker-ce, please follow the post-installation steps to make sure that the docker can be run without sudo.
-
Install nvidia-container-toolkit by following these installation instructions.
-
Login to the NGC staging area docker registry using the api key generated from
ngc.nvidia.com
docker login nvcr.io- To install the launcher, please run the following command from the root of the repo.
export PYTHONPATH=${PWD}:${PYTHONPATH}
python release/tao/setup.py developAdd the --user flag to the setup.py script, if you would like to install it locally to your user.
- Run the tao tasks using the changed tao cli structure.
tao <task_group> <task> <sub-task> <cli-args>This project is licensed under the Apache 2.0 License.