can you share the dataset?
#7
by
adeebDkheel
- opened
Hello,
I used the code at your github demo [https://github.com/01-ai/Yi-1.5/blob/ded4f4954bfbac0b538c447c36e78ca4f4100340/demo/web_demo.py]
and said 'hello' 'who are you' in arabic language, the model replied with "I'm part of openAI project"!
NOTE: I didn't use system message.
Actually if you use English, it will tell you that it is ChatGPT with some different representation of "hello" as well.
And even for the base model, it is the same case.
But I do not think this is a big deal.
Sorry, we are unable to share our data, and we have also noticed this issue. We have used some synthetic data, and you can construct your self-awareness data to correct it. This problem will not affect your use.
adeebDkheel
changed discussion status to
closed