hello? 虽然是一个推理模型，但有的方面也太离谱了吧
我可以接受一个模型专门用于math和coding，但一个最简单的hello都输出了一串数学解答，是不是太离谱了呢？希望作者可以解释一下这个现象
明显过拟合了
Our model is an experimental research prototype dedicated to mathematical reasoning, released specifically to validate the claims in our paper. It relies on a math-only base model with further post-training focused on math and code. Consequently, it has not been aligned for general conversational capabilities. We do not recommend using this model for general chat, as it is biased towards responding from a problem-solving perspective. Additionally, please note that running inference with quantized versions may lead to increased hallucinations when testing general conversation scenarios.
明显过拟合了
我们的训练过程经过严格去污，可以泛化到数学、竞赛类编程内的其他未见过的题目。我们不推荐将该模型用于日常对话等领域进行测试，因为该模型由Qwen2.5数学base模型进行数学、code、stem领域后训练得到，并未针对性做RLHF等用于日常问答的优化。该问题不属于过拟合问题。
