Dealing with models/logic that do not fit into triton available backends.

[okyspace]

What are the options to deal with models/logic that does not work with python, tf, torchscript, onnx, etc ?

I understand we can have a conda env to run but I cannot find a way to make this work in the ensemble. Is there some ref or less customised way?