How to Deploy ML Models Using Gravity AI and Meadowrun

1

Amazon ECR

When you download a container from Gravity AI, it comes as a .tar.gz file that needs to get uploaded to a container registry in order to work. There are slightly longer instructions in the Gravity AI documentation, but here’s a short version of how to create an ECR (Elastic Container Registry) repository then upload a container from Gravity AI to it:.

#Container Registry #Cloud Hosting #Object Storage 39 social mentions

2

Apache Arrow

As a bit of an aside, you could imagine a way to get the best of both worlds with an extension to Docker that would allow you to publish a container that exposes a Python API, so that someone could call sentiment = call_container_api(image="huggingface/transformers", "my input text") directly from their python code. This would effectively be a remote procedure call into a container that is not running as a service but instead spun up just for the purpose of executing a function on-demand. This feels like a really heavyweight approach to solving dependency hell, but if your libraries are using a cross-platform memory format (hello Apache Arrow!) under the covers, you could imagine doing some fun tricks like giving the container a read-only view into the caller’s memory space to reduce the overhead. It’s a bit implausible, but sometimes it’s helpful to sketch out these ideas to clarify the tradeoffs we’re making with the more practical bits of technology we have available.

#Databases #NoSQL Databases #Relational Databases 33 social mentions