Extremely slow (3 t/s) and can't setup llama.cpp + a couple of errors

[sudo-saif]

Hey there, I have just tried Mux, and I face two problems that make the app non-usable at the moment.

1. Getting extremely low speeds with Qwen 3.6 35b a3b:

• Using OpenAI compatible URL for LM Studio, I get 3 t/s speed (note: my average LM Studio speed is 72 t/s)

2. Using Qwen 3.5 9b, it works fast but I get a couple of "Invalid type for 'input'." errors during chats:

• Speed: LM Studio ~ 86 t/s while Mux is ~80 t/s

3. When I set llama.cpp up via OpenAI compatible URL, then try to chat, it keeps showing "Cannot determine type of 'item'
   " error: