wls04/ultrafeedback-length-bias

--- dataset_info: features: - name: chosen list: - name: content dtype: string - name: role dtype: string - name: rejected list: - name: content dtype: string - name: role dtype: string - name: prompt_text dtype: string - name: length_diff dtype: int64 - name: tokenizer dtype: string splits: - name: train num_bytes: 146580787 num_examples: 30567 download_size: 78799632 dataset_size: 146580787 configs: - config_name: default data_files: - split: train path: data/train-* ---