为什么qwen3 1.7b eagle3 中文的猜中率很低，甚至猜不中。在线求?

我使用官方提供的代码，发现qwen3 1.7b eagle3 草稿模型对中文而言，猜中率很低，甚至猜不中， 请问真实情况是这样的吗？是不是训练草稿模型的时候，没有训练中文的情况？还是说我这边操作有误？有没有哪位大佬可以解答一下吗？非常感谢!

I used the official code provided and found that the qwen3 1.7b eagle3 draft model has a very low guessing rate for Chinese, and even cannot guess correctly. Is this the real situation? Is it true that when training the draft model, there was no training in Chinese? Or did I make a mistake in my operation? Is there any expert who can answer this? Thank you very much!

Start by reproducing the reported behavior with the official code and Chinese prompts using the Qwen3 1.7B Eagle3 draft model. No file, test, or exact operation details are provided, so compare the setup and observed guessing rate with the documented expected behavior. Done means determining whether this is expected model behavior or an operation or implementation problem, then documenting the evidence.