Exactly the same as #235, I opened another one because the previous one was closed by the author.
The browser itself freezes for a good time, sometimes a few seconds, when doing chat.generate. Tried using web workers, but as expected that doesn't work because Chrome itself freezes.
Seems like a major performance problem

The problem seems to become worse with a larger context window.
The freezes are longest on the first time inference is performed on a page refresh.
Slightly noticeable with small models like Qwen2.5-0.5b, very noticeable with larger ones like Llama-8b.
Windows 11 Chrome 136