Best practice for an input pair <I_1, I,2>^0, ..., <I_1, I,2>^N

[shaked571]

Hi,
I have a text classficitaion model (with 3 classes). My business logic consists of 2 partsPart A and Part B when in the in the post processing I apply another Binaric model that its input is the 2 predictions Part A^ and Part B^ and classify it to a Binaric

So the flow is:
[Part A, PartB] -> Model_1(Part A), Model_1(Part B) -> Model2 (Model_1(Part A), Model_1(Part B))

How should I write the config in Triton to support such a scenario? I know I should use the ensemble, but it seem that ifI create 2 models, it would load 2 instances of Model_1 and I don't think it is neccesry.

Any help would be appreciated