aws-lambda-pytorch-image-classification-example

Example of implementing a pytorch image classifier service using AWS lambda. Cold start time can be 10-20 seconds. Subsequent inferences happen in approx 300-500 ms.

docker build -t lambda-pytorch-example .

docker run -p 9000:8080 \
-e AWS_ACCESS_KEY_ID=$AWS_ACCESS_KEY_ID \
-e AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY \
aws-lambda-pytorch-image-classification-example:latest

To connect to the shell of the running container, check the container name with docker container ls then docker exec -it <container_name> /bin/bash. Run the request.py file to make a request.

CICD

A github action is used to push the latest release to ECR, enter required credentials in repo secrets

References

• https://github.com/gokavak/lambda-docker-image-pytorch-xgboost
• https://github.com/aws-samples/aws-lambda-docker-serverless-inference/tree/main/pytorch-inference-docker-lambda