This project generates a docker image enables runs a fastAi software stack that can run on a modern AMD graphics card. It runs on a rocm implementation of pytorch 1.7.0. This docker image runs a Jupyter Lab that can be used to run the fastAi library.
Notes: as of 2020-11-21, FastAi requires pytorch 1.7.0, if that changes to something else, then this git repo will need to be updated accordingly.
This docker image contains:
- Jupyter Lab
- Pytorch
- Fast.ai
- Rocm Utilities
- Python 3.6
- nbdev
- fastbook (FastAi course material libraries)
Initial Setup (This downloads the files, builds the docker container etc)
sudo docker build --tag rocm .Usage
sudo docker run -it -v Host/Data/Location:/data -p 1337:1337 --privileged \
--rm --device=/dev/kfd --device=/dev/dri --group-add video --shm-size 16gb rocm The Host/Data/Location folder is exposed to the Docker container and is where the Jupyter lab starts from. Anything not in the data folder is lost when you rebuild the container.