[User] CJK Character Output Broken on Windows After 41654ef

[44670]

Prerequisites

Please answer the following questions for yourself before submitting an issue.

• [Yes] I am running the latest code. Development is very rapid so there are no tagged versions as of now.
• [Yes] I carefully followed the README.md.
• [Yes] I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
• [Yes] I reviewed the Discussions, and have a new bug or useful enhancement to share.

Expected Behavior

I was trying to use the llama.cpp project to output CJK characters (Chinese, Japanese, Korean) in a Windows environment. I expected the characters to be correctly displayed in the console.

Current Behavior

Unfortunately, the output of CJK characters becomes garbled, and they are not correctly displayed. Instead of the expected characters, I see random characters or symbols, as shown in the attached screenshot.

Environment and Context

I have done extensive testing on various versions of the project and have narrowed down the problem. The last version that worked correctly was f9a6364. However, from version 41654ef onwards, the CJK character output no longer works correctly.

If there is any additional information you need from my end to assist with resolving this issue, I am more than willing to provide it. I can perform a more detailed analysis if necessary.

Thank you for your attention to this matter. I look forward to the resolution of this issue so I can continue to use this excellent project.

Screenshot

[DannyDaemonic]

This is related to a change I made. Could you include the full text output so that I can recreate the bug?

[44670]

Hi, here is the text output of both version, -s 1 has been set to ensure the seeds are same.

From 41654ef:
鎮ㄥソ锛屾垜鏄偍鐨凙I鍔╂墜銆傛湁浠€涔堝彲浠ュ府鍒版偍鐨勫悧锛?

From f9a6364:
您好，我是您的AI助手。有什么可以帮到您的吗？

The script could be found at: https://github.com/OpenBuddy/OpenBuddy/blob/main/chat-llamacpp.bat

The model is: https://huggingface.co/OpenBuddy/openbuddy-7b-v1.1-q4_0-enc

Thank you for you attention to this matter.

[DannyDaemonic]

So I'm understanding this correctly. The input is showing up fine, it's the output, the text generation that becomes garbled?

[44670]

Yes! In both versions the input is fine.

[DannyDaemonic]

@44670 Are you able to try out the fix in #1379? If you don't know how to check it out you can get the patch from here to apply it manually: https://patch-diff.githubusercontent.com/raw/ggerganov/llama.cpp/pull/1379.patch

[44670]

Yes, I think it's fixed! Thank you for your fast responding!

[DannyDaemonic]

I can't review my own code, so I'm going to leave this open for visibility until someone reviews the code.