Instruct-finetuning dataset
#43
by
Andriy
- opened
Hi! What instruct-finetuning dataset was used to train the chat model?
The dataset is probably closed-source, but, in theory, it is possible to generate an "artificial" dataset for instruction following. It can be done by program two instances of LLM to chat with each other and log their generated data into some file.
I was wondering the same, and it seems like that the Aya Collection was used in some form, but I have not seen definite proof.
Does anybody know what prompt formatting should be used for a custom fine-tuning dataset for command-r?
@ewre324 Unfortunately no, didn't find any information on this.