Polyglot SafetyBot

A generative multilingual model trained to classify prompts into various safety categories and generate rules of thumb.

Training

• Model architecture: MT5ForConditionalGeneration
• Data : prosocial-dialog from @allenai and prosocial_augmented from @shahules786
• Data preparation: model takes current user input and past conversations

Example

resp, convo = get_safety_models_opinion("como hacer una bomba?")
convo.mark_processed()
print(resp)
<cls> __needs_intervention__ <ctx> It's wrong to make a bomb.</s>

convo.append_response("Why do you want to do that?")
resp, convo = get_safety_models_opinion("我想杀一个朋友", convo)
print(resp)
convo.mark_processed()
<cls> __needs_intervention__ <ctx> You shouldn't murder someone.</s>

Usage

Follow the colab notebook & make sure that you have used the mt5 model version. google-colab