zerotosingularity/fastai2 repository overview

⁠Important

This container has moved to https://hub.docker.com/repository/docker/seemeai/fastai2⁠.

The existing tags will remain for a while, but we recommend switching to using https://hub.docker.com/repository/docker/seemeai/fastai2⁠.

There you will find:

• newer versions of the framework
• support for different PyTorch/torchvision versions

⁠Old info for reference

⁠DESCRIPTION

This Docker container allows you to easily get started with using fast.ai⁠'s version 2⁠.

This is not an official Fast.ai release, nor associated with Fast.ai.

The container is being maintained by proud members of the Fast.ai community, with a goal to be helpful to people wanting to use Fast.ai in a Docker container.

From what we can find, you will need a Linux machine and an NVidia GPU.

⁠LICENSE

MIT

⁠How to use this container

0. Install CUDA 10.2 on the host system https://developer.nvidia.com/cuda-toolkit⁠

1. The machine running the CUDA container only requires the NVIDIA driver, the CUDA toolkit doesn't have to be installed. from https://github.com/NVIDIA/nvidia-docker/wiki/CUDA⁠

2. Install Docker: (docker-ce or docker-ee) https://docs.docker.com/install/⁠

3. Install Nvidia-docker: https://github.com/NVIDIA/nvidia-docker⁠

4. Checkout code and run container

docker run --runtime=nvidia -it -p 8888:8888 -v $HOME/.fastai:/root/.fastai -v $HOME/.torch:/root/.torch -v $(pwd):/code --ipc=host zerotosingularity/fastai2:tagname

It uses three volumes:

• /root/.fastai: store the data for later use, so you don't have to redownload every time
• /root/.torch: store the data for later use, so you don't have to redownload every time
• /code: makes the current directory available in /code