Closed pabl-o-ce closed 5 months ago
Hi @pabl-o-ce,
If you're talking about the ShareGPT_Vicuna_unfiltered dataset, it does not have a pairwise preference format necessary for ORPO. So in that case you can further develop pairs based on the dataset or maybe use it to perform SFT before applying ORPO.
Hi @nlee-208 yeah the main problem I was looking is to generate the negative response for complete a dataset that is on sharegpt
is there a way to modify a current dataset ShareGPT to ORPO?