Can this handle smaller models like Qwen 1.5B, or does it need some real intelligence to get the tool calling to work?