Codes to process existing feedback dataset to pairwise training format.
See process_and_upload_dataset.py for details of processing.
to recreate the dataset, please check the process_and_upload_datasets.sh. After running this script, data is uploaded to 🤗Unified_feedback.
python to_mixinstruct_format.py --unified_feedback_dataset_name "all" --split "train" --to_save_data_dir "data/unified_feedback"
python to_mixinstruct_format.py --unified_feedback_dataset_name "all" --split "val" --to_save_data_dir "data/unified_feedback"