This is pretty standard in the machine learning world. You'll open-source the code and weights trained on a public data set (these are often licensed specifical

This is pretty standard in the machine learning world. You'll open-source the code and weights trained on a public data set (these are often licensed specifically for non-commercial use). But in production, you'll be using different weights trained on a proprietary data set.