training data

#1
by aari1995 - opened

hi guys,

thank you again for this awesome model!

I was just wondering what data you used for the training, especially from when, as it would be interesting to know, for example the model "already" knows that Olaf Scholz is german chancellor. Or was it fineweb de?

I would be happy over some feedback about the source / or at least the year.

THank you!!

VAGO solutions org

Hi Aari,

thanks for your feedback. We did not focus on training "new knowledge" into the model. The focus was rather on German grammar and wording. I could imagine that such knowledge comes from the original Qwen2 model. We did not use fine web de, we used our own curated data set for CPT.

Best regards

Sign up or log in to comment