Support for ONNX Models

[m4salem]

Thanks for this great work ! It works with pyfunc models however this faces the problem when having no pypi installable packages. Instead, can this support ONNX models so the inference endpoint works out of the box ?

[acere]

Thanks @m4salem , glad you find this sample useful.
Can you help me understand better the "problem when having no pypi installable packages"? if you have a reproducible example, happy to look into it more in detail

Similarly, if you have an example of an ONNX model it would be helpful

[m4salem]

Thanks @acere for the quick response.
1- Scenario 1: Having a model in Databricks Mlflow that relies on in-house package that can't be served with the provided container in the sample.
2- Scenario 2: Logging an ONNX model to MLFlow does not trigger the model transfer functionality in this sample (S3 bucket and model are not created on AWS), the reason being is it supports only python function, tf, xgboost, and sklearn as far as I explored in the code.