When evaluating the Qwen3-30B-A3B-Thinking-2507 model, which LMStyle should be specified? DeepSeekCodeInstruct or CodeQwenInstruct ? Thanks!