Which datasets were used for the instruct SFT?
#1
by anrilombard - opened
Hi, thanks for releasing the AfriqueQwen models, they've been really useful for my isiZulu work 🌍
Would you mind sharing which datasets went into the post-training for this instruct version? I couldn't find it on the card or in the AfriqueLLM paper, which seems to cover the base models. I'm mainly trying to get the licensing right for anything built on top of it, so knowing whether any of the SFT data was generated by other models (e.g. Llama or GPT outputs) would help a lot, and whether any benchmark train splits like AfroBench were included.
Thanks!