Serverless PyTorch model serving on AWS with lambda

This project illustrates how to serve a PyTorch model in production as a REST API using a simple CIFAR model for illustration from the PyTorch CIFAR tutorial.

Blog post on paepper.com/blog

For full details about the approach and how to run it yourself, see my blog post covering the details

Project Overview

This is the project tree overview:

Folder model/

Inside this folder are all the files which are relevant to our model. This is:

• The trained PyTorch model: simplecifar_jit.pth
• The target classes of the images: classes
• Both packed together as a bundle for AWS S3: model.tar.gz

Folder aws/

This folder is for the serverless part. It contains:

• The lambda function: pytorch/prediction.py
• The AWS SAM template: template.yaml
• The packaged bundle to upload to AWS: packaged.yaml
• An environment file to test the lambda function locally: env.json