The “criteria-based evaluator sandbox for AI responses” that I have developed is now available within the CiberIA system and my AIsecTest, which I commercialize as part of AI security audits through model or system introspection.
This system can fully operate in English, Catalan -or in any other language- and is powered by the Apertus model.
I’m sharing a link to a simple open space that I’ve set up for anyone who wants to have a small taste of it. I normally use a full private environment to run demos for clients, but with this OPEN option you will be able to test it yourself. However, you’ll need to provide your own HF token and switch the machine (at no cost) to one with GPU support (recommended: Nvidia A10G Large with 12 vCPUs, 48 GB RAM, and 24 GB VRAM).
You can easily adjust the repository configuration to adapt it and run it. If you deploy it “as is,” it will throw a Runtime Error.
If you are one of my active clients, feel free to contact me and I will give you access to a private demo at no cost and with no additional setup required.
Link: https://huggingface.co/spaces/gcjordi/auditor-ciberia-apertus-space-OPEN
Top comments (0)