[BUG/Help] <title>关闭do_Sample 后，只推理单条问题和推理该问题前又推理了其他问题输出不一致

[J-G-Y]

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

调用model.stream_chat（）函数做输出，将do_sample关闭，正常此刻推理同一条内容，答案输出应一致。但是在实际测试中，仅推理该条时和在该条前添加一条问题时的输出不一致，history也置为空了，但是没有效果。原因求解。

Expected Behavior

Steps To Reproduce

1

Environment

- OS:
- Python:
- Transformers:
- PyTorch:1.13.1
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) :12.0

Anything else?

No response