Looks like it errors out in LM Studio using the Unsloth quants, apparently the Unsloth team has already posted patches for llama.cpp to support this.