The question is not support. It is clear that LM Studio has nothing custom in term of actual inference code. The curiosity is that they are still stuck at LLaMa and Falcon, etc., giving Mistral’s performance.