qwen reranker的模型训练数据里是下面这样的,但是这样没办法添加think,qwen的reranker模型本身是带有think的,所以想问下有带think的训练方式么 `{"query": "query", "positive": ["relevant_doc1", "relevant_doc2", ...], "negative": ["irrelevant_doc1", "irrelevant_doc2", ...]}`