How does it support multi-turn conversations?
jujeongho0 opened this issue · comments
jujeongho0 commented
Hi. Thanks for the great work.
I was wondering about the meaning of "WizardLM-2 adopts the prompt format from Vicuna and supports multi-turn conversation.", so I'm leaving an issue.
I know that the datasets evolved through the EVOL process are only single-turn (maybe trained with the prompt format from Vicuna), so I'm wondering how it supports multi-turn conversation.
Or, did you create a multi-turn dataset with the EVOL process and train on it in the form of a multi-turn conversation?
Thanks.