API for faster key generation

Hello @Rish ,
How long does your keygen take ? Please be aware that the comment seems a bit outdated as we ran this key generation in around 2-3 minutes on our machines.

Besides, models like Convolutional Neural Networks (I believe that’s where you took this comment) have complex structures. This means that the underlying FHE circuit gets complex as well, and that the key generation will most probably take a longer time than “simpler” models like linear or tree-based models ! Currently, we do not provide any other key generation algorithm to make it faster. However, we are always working on improving such key features and you could expect some improvements in future releases.

As for your question on “low level operations”, could you maybe be a bit more specific ? I believe you might be more interested in our Concrete library, where you can define, compile and execute any operation or algorithm in FHE.

However, if you are talking about having the ability to call different steps for a Concrete-ML model, I recommend you to have a look at this documentation page. You will see how to easily call keygen, encrypt, decrypt and run in separate calls. Furthermore, if your goal is to deploy a model in a client-server setting, here’s how to use our dedicated API.

Hope this helps !