[Bug]: Poor support for Hebrew

[shohamyamin]

Model Series

Qwen2.5

What are the models used?

Qwen2.5-Turbo-1M-Demo

What is the scenario where the problem happened?

when using hebrew in the input

Is this a known issue?

• I have followed the GitHub README.
• I have checked the Qwen documentation and cannot find an answer there.
• I have checked the documentation of the related framework and cannot find useful information.
• I have searched the issues and there is not a similar one.

Information about environment

I used the Qwen/Qwen2.5-Turbo-1M-Demo in hugging face

Log output

input: 
"היי אתה יכול לרשום לי פסקה בעברית על כנסת ישראל?"

output:
"แน่นอน! 以下是关于以色列议会的段落，但我将用希伯来语书写：

כנסת ישראל היא הגוף המחוקק העיקרי במדינת ישראל. היא כוללת 120 נציגים שנבחרים בבחירות כלליות אחת לשנתיים. הכנסתי的第一个职能是制定法律，监管国家的行政机构，并批准预算。ckeet以色列议会的成员被称为国会议员（MK），他们代表不同的政党或独立候选人。议会大厦位于耶路撒冷。议会的历史可以追溯到1949年，当时第一届议会选举举行。自那时以来，议会经历了许多变化和发展，反映了以色列社会的多样性和复杂性。"

Description

Steps to reproduce

This happens to Qwen/Qwen2.5-Turbo-1M-Demo and every qwen model
The problem can be reproduced with the following steps:

1. asking a qwen model a question in Hebrew like: "היי אתה יכול לרשום לי פסקה בעברית על כנסת ישראל?" meaning "write me a paragraph in Hebrew about Israeli Knesset?"

Expected results

The results are expected to be:
The all answer should be in Hebrew without any chines letters or irrelevant English words

Attempts to fix

I have tried several ways to fix this, including:

1. trying different system prompts
2. trying different questions

Anything else helpful for investigation

I find that this problem also happens to all qwen models. it seems that the support for Hebrew is not good at all

[leo-smi]

FREE PALESTINE