Skip to content

Dataset · allenai

Human-ChatGPT Conversations Without Toxic Content

allenai/WildChat-4.8M

3.2M human-ChatGPT dialogues filtered for non-toxic user inputs and responses.

What's inside

A collection of 3,199,860 conversations between humans and ChatGPT. In version 4.8M, only non-toxic user inputs and ChatGPT responses filtered by OpenAI Moderations are retained. The data is text-based and intended for generation and question-answering tasks.

What you can do with it

  • SFT of chat models on real dialogues
  • Evaluating toxicity filtering
  • Training question-answering systems
Tasks
Text generation, Question answering
Size
1M–10M rows
Downloads this month
24.2K
Likes
219

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("allenai/WildChat-4.8M")