[Question]将qwen3.5-27B转换为torch dist后，再转换回hf格式，使用sglang部署，在tp=4/8的情况下模型会输出且只输出大量‘！’

Your Question

将qwen3.5-27B转换为torch dist后，再转换回hf格式，使用sglang部署，在tp=4/8的情况下模型会输出且只输出大量‘！’，只有在tp=1/2的情况才能正常输出，不知道这个情况应该怎么解决。

What I've Tried

首先使用最原始的hf模型权重在tp=8的情况下是可以正常输出的，并且在我进行grpo训练过程中模型都是可以正常roll out的，我已经尝试过将原始hf模型，训练后的torch dist模型转换回hf格式部署，都会出现这种错误，另外我使用了很长的tools做输入，如果将tools长度减少，模型可以恢复正常。

Environment (if relevant)

• slime version:
• Python version:
• PyTorch version:
• CUDA/ROCm version:
• GPU type and count:
• OS:

Additional Context

No response

Pre-submission Checklist

• I have read the CONTRIBUTING.md and understand the collaboration scope.
• I have read the documentation and FAQ and my question is not answered there.
• I have searched for existing issues and my question has not been asked before.

No repository file, test, or entry point is named. Start by reproducing the HF-to-torch dist-to-HF conversion and sglang deployment at TP=1/2/4/8, comparing original and converted weights while reducing the long tools input; done means identifying a reproducible cause or confirmed workaround with complete environment details.