Not sure how this will look exactly. Would like to be able to support AWS Sagemaker models directly in AWS lambda layer. This will allow for serverless sagemaker model serving.
Notes
- Sagemaker python 3 is python 3.5
- Will this require custom miniconda runtime? Personally not sure
- At least some of sagemaker models save as .protobuf - these need a .proto to make sense but I have yet to find them.
- Prefer to support this without dependencies on boto or sagemaker
- Probably need to support core ML libraries : numpy pandas scipy, but may be able to get away with only one at a time of: pytorch, tensorflow, sklearn.
Not sure how this will look exactly. Would like to be able to support AWS Sagemaker models directly in AWS lambda layer. This will allow for serverless sagemaker model serving.
Notes