Reduce PyTorch project bundle size to be able to deploy to Lambda

elcronos / COVID-19

COVID-19 Detector from x-rays using Computer Vision and Deep Learning

314 stars 68 forks source link

Reduce PyTorch project bundle size to be able to deploy to Lambda #30

Open franciscohanna92 opened 4 years ago

franciscohanna92 commented 4 years ago

The project is too heavy to run on AWS Lambda. It throws the following error when trying to deploy:

STDERR: ERROR: Could not install packages due to an EnvironmentError: [Errno 28] No space left on device

Torch is the heaviest dependency. Almost 800MB.

nietzscheson commented 4 years ago

AWS have a service: https://aws.amazon.com/es/pytorch/ locally we could use the PyTorch Docker image. In production the AWS service...

elcronos commented 4 years ago

Hmm we could have sage maker for that but it's a bit expensive. We will use it just if it's the last option.

ajhenaor commented 4 years ago

What about dockerizing the model and deploying it using AWS ECS?

aenriquerg commented 4 years ago

Lamdbas will last until we reach a high traffic (it will become expensive). We should consider basing the architecture on AWS ECS, so it can help to handle cost.