While reading the post, it didn’t strike me as odd. I worked with small services wrapping ML models and 200 lines looks very slick and I don’t recall anyone considering weights as part of the code, and even the model training code is treated differently that the cliente code. Seems legit
