An idea

#1
by Cran-May - opened

Is it possible to build a multilingual model using the same approach (using the multilingual model developed by lightblue)? The model has shown the ability to outperform this 8B model in every way and supports more languages. I think this attempt is interesting.

Owner

Thanks!
I believe that the performance of this model is probably due to the performance of the original Llama-3-70B-instruct, and that the experiment did not yield very promising results.
However, I believe there is still room for improvement in this approach and will try a few more things.

Sign up or log in to comment