Skip to content

Dataset · allenai

Prosocial response dialogues

allenai/prosocial-dialog

100K+ multi-turn English dialogues for training safe conversational agents.

What's inside

ProsocialDialog is a large multi-turn English conversational dataset. It teaches conversational agents to respond to problematic content according to social norms. Responses rely on commonsense social rules (rules-of-thumb, RoTs); the data was created collaboratively by humans and AI. Includes diverse unethical, problematic, biased, and toxic situations.

What you can do with it

  • Fine-tune a conversational model for safe responses
  • Evaluate toxicity and prosociality classifiers
  • Create a LoRA for responses based on social rules
Tasks
Text classification
Languages
English
Size
100K–1M rows
Downloads this month
1.3K
Likes
119

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("allenai/prosocial-dialog")