seemeai/fastai repository overview

⁠Description

This Docker container lets you easily get started with using fast.ai⁠ (https://github.com/fastai/fastai⁠).

This is not an official Fast.ai release, nor associated with Fast.ai. If you are looking for the official fast.ai container, they are located here⁠

The container is being maintained by seeme.ai⁠, with a goal to be helpful to people wanting to use fast.ai in a Docker container.

To get the best experience, you will need a linux machine with a NVidia GPU.

⁠License

MIT⁠

⁠How to use this container

1. Install Docker: https://docs.docker.com/install/⁠

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

3. Run container

docker run -it --gpus all --shm-size=32g -p 8888:8888 -v $HOME/.fastai:/root/.fastai -v $HOME/.cache:/root/.cache seemeai/fastai:tagname

It uses two volumes to persist the data:

• /root/.fastai
• /root/.cache

--shm-size: set the RAM you want the container to use

⁠Some more details on the tags

⁠Naming

Tags are named using the PyTorch⁠ version and the fastai version together with a build version.

pt-%PYTORCH_VERSION-fa-%FASTAI_VERSION%-%BUILD_VERSION%

For example, the tag:

pt-1.11.0-fa-2.7.2-1.1.0

consists of:

• PyTorch v1.11.0
• Fastai v2.7.2
• Build v1.1.0

⁠A quick note on torchvision

Every PyTorch version comes with its own torchvision, which is installed but not included in the tag naming.

More info on torchvision⁠

⁠Base container

The current base container is: pytorch/pytorch:$torch_version-cuda11.3-cudnn8-runtime

where torch_version is 1.10.0 or 1.11.0

⁠Contact

[email protected]⁠