Fast.ai hotdog classifier

Serverless classifier trained with AWS Sagemaker and deployed on AWS Lambda

See blog post: https://medium.com/@pvhee/building-a-serverless-not-hotdog-classifier-with-aws-sagemaker-and-lambda-dd473cf6506a

Model training

Check out the Jupyter notebook for model training using fast.ai and make sure to save your model to AWS S3.

Inference in production

We use the serverless framework to get our model inference on AWS Lambda.

serverless deploy
...
endpoints:
  POST - https://ax45wn5lv2.execute-api.us-east-1.amazonaws.com/dev/invoke

Example inference on this hotdog.

curl -d "{\"url\":\"https://leitesculinaria.com/wp-content/uploads/fly-images/96169/best-hot-dog-recipe-fi-400x300-c.jpg\"}" -H "Content-Type: application/json" -X POST https://ax45wn5lv2.execute-api.us-east-1.amazonaws.com/dev/invoke

Load testing inference

We use artillery.io to load test our model inference.

cd loadtest/
artillery run artillery.yml