language:
- en
- de
- ru
- fr
- es
license: mit
size_categories: - 1K<n<10K
task_categories: - text-classification
dataset_info:
features:- name: answer_1
dtype: string - name: refusal_1
dtype: string - name: answer_2
dtype: string - name: refusal_2
dtype: string
splits: - name: spanish
num_bytes: 2752983
num_examples: 1402 - name: russian
num_bytes: 3898653
num_examples: 1381 - name: french
num_bytes: 3257319
num_examples: 1728 - name: german
num_bytes: 3162383
num_examples: 1695 - name: english
num_bytes: 3330130
num_examples: 1799
download_size: 9078297
dataset_size: 16401468
configs:
- name: answer_1
- config_name: default
data_files:- split: spanish
path: data/spanish-* - split: russian
path: data/russian-* - split: french
path: data/french-* - split: german
path: data/german-* - split: english
path: data/english-*
- split: spanish
Data description
This dataset is designed to train and evaluate models for the task of refusal detection in generated responses. The dataset consists of input prompts sourced from the lmsys/lmsys-chat-1m collection, encompassing a variety of languages including English, German, French, Russian, and Spanish. To increase refusal diversity, the responses and refusals were generated using two models, Gemini Flash 1.5 and LLaMA-3.3-70b.
The dataset is primarily intended to train classifiers to distinguish between useful responses and refusals, which are typically statements where the model declines to answer or acknowledges its limitations in providing a response. These refusals were manually labeled and paired with their respective input prompts.
Language splits:
Spanish: 1402
Russian: 1381
French: 1728
German: 1695
English: 1799