Closed nathanielrindlaub closed 1 year ago
This will be a very good thing to do now that we're taking on the cost of hosting Megadetector ourselves (#55).
I also would like to revisit instance types to make sure we're not over provisioning.
Closing this out as it will no longer be relevant with MIRAv2.
Consider consolidating mira-small and mira-large into a multi model endpoint? Would that be cheaper to host?
The sagemaker Tensorflow Serving container looks like it might be able to support multiple models on it's own. We should run that to ground.