Right now the PreferenceDataset class randomly pick one of the provided prompt template. If I am not mistaken, this is not the cases for other dataset generator. We should also include this functionality for other datasets with an argument prompt_selection taking values 'all', 'random'. Then we also need to ensure traceability in the generated dataset. Probably via metadata.
Right now the PreferenceDataset class randomly pick one of the provided prompt template. If I am not mistaken, this is not the cases for other dataset generator. We should also include this functionality for other datasets with an argument
prompt_selectiontaking values 'all', 'random'. Then we also need to ensure traceability in the generated dataset. Probably via metadata.