[BUG] Minicpm-o audio conversation issues

[janak2]

是否已有关于该错误的issue或讨论？ | Is there an existing issue / discussion for this?

• [x] 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions

该问题是否在FAQ中有解答？ | Is there an existing answer for this in FAQ?

• [x] 我已经搜索过FAQ | I have searched FAQ

当前行为 | Current Behavior

I am seeing some issues when conversing with MiniCPM:

1. MiniCPM-o is getting stuck in a loop. for example if i ask it “how are you?”, it might respond with “i’m sorry, but i cannot assist you with that”. It only breaks out of this loop after i move to a different topic like “what’s the capital of France”.

2. It sometimes randomly switches languages or generates noises/gibberish.

I understand that these are issues inherent to the model but are there ways to reduce the frequency of these errors?

期望行为 | Expected Behavior

No response

复现方法 | Steps To Reproduce

No response

运行环境 | Environment

- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA (`python -c 'import torch; print(torch.version.cuda)'`):

备注 | Anything else?

No response