With Webgpu users should be able to run other models as well. I have built chatllm.pro cheome extension that can run all sort of small models incliding the phi2 natively using a wasm library. You will be able to chat with the page as well