Currently, large language models (LLMs) have gained very strong reasoning capabilities, with a key factor being test-time ...