This project allows you to build a Docker container to build, train and deploy fast.ai based Deep Learning models with Amazon SageMaker.
Since the fast.ai library is based on PyTorch, this project builds upon the SageMaker PyTorch Container meaning that you can bring your own fast.ai scripts for training and deploying your models using the PyTorch Estimator, Model and Predictor objects in the SageMaker SDK.
Amazon SageMaker utilizes Docker containers to run all training jobs & inference endpoints. Make sure you have installed Docker on your development machine in order to build the necessary Docker images.
The Docker images are built from the following Dockerfile.
We have a utility script that builds 2 Docker images, a GPU based image and a CPU based image, on your machine locally and pushes them to your ECR repository.
To build the images run the following script:
./build_and_push.sh
This sample code is made available under a modified MIT license. See the LICENSE file.