Concrete ML is a machine learning framework built on top of Concrete Numpy (CN). Concrete ML provides FHE compatible models, that are trained with sklearn or torch, and implements quantization which makes the models compatible with FHE constraints. Finally, Concrete ML implements model inference code that is written using Concrete Numpy, which compiles to FHE.
If you can’t find the ML model you’re looking for in Concrete ML, it is possible to use Concrete Numpy directly to write its inference implementation in FHE. To do so you could take example from Concrete ML tools or use them in your model implementation. See the developer guide for information about building custom models.