What’s Out There
← Back to swiping
Language & textFreeCommunity project

OpenAssistant Conversations (OASST1)

OpenAssistant (LAION-affiliated volunteers)

161,443 chat messages written and ranked by volunteers in 35 languages, with no model in the loop.

A crowd-sourced assistant-style conversation corpus. One row is one message, linked into conversation trees with human quality ratings and labels. It includes more than 10,000 fully annotated trees, making it one of few instruction sets written entirely by people.

Access

Cost
FreeNo charge.
License
Open licenseApache 2.0
Sign-up
No sign-up
How you get it
API and Bulk downloadParquet, JSON Lines

What's in it

One row is
One row per message in a conversation tree
Covers
Global volunteers, 35 languages (English and Spanish dominate)
Years
2022-2023
Updated
One-time release
Size
161,443 messages and 461,292 quality ratings in 10,000+ complete trees
message textrole (prompter or assistant)parent messagelanguagequality ratingsrank among siblings

Quality

4 of 5

Rare fully human-written chat data with rankings. Volunteer quality varies and topics skew toward what volunteers found interesting.

  • Some replies are low effort, wrong, or copied from other sources
  • Language mix is uneven
  • Volunteers were not vetted experts

Connects to

  • Databricks Dolly 15k: Another human-written instruction set, from employees rather than volunteers
  • No Robots: Professionally written instruction set for comparison
  • UltraChat: Synthetic alternative at far larger scale

Questions it could answer

  • How do ratings differ by language?
  • Which conversation branches do raters prefer, and what do they share?

Similar data sets

  • LMSYS-Chat-1M

    One million real conversations with 25 LLMs, collected from Vicuna demo and Chatbot Arena visitors.

    Language & textLMSYS (UC Berkeley and collaborators)Free

  • Chatbot Arena Conversations

    33,000 side-by-side model chats where real users voted for the better answer.

    Language & textLMSYS (UC Berkeley and collaborators)Free

  • YODAS v3

    Over 1.1 million hours of CC-BY web audio in more than 100 languages.

    Audio & musicESPnet (Carnegie Mellon University and collaborators)Free

  • MADLAD-400

    A Common Crawl corpus in 419 languages whose lower-resource portions were manually audited.

    Language & textGoogle Research and Allen Institute for AIFree

  • OPUS Parallel Corpora

    Over 100 billion aligned sentence pairs across more than 1,000 languages.

    Language & textUniversity of Helsinki (Language Technology Research Group)Free

  • HH-RLHF (Helpful and Harmless)

    Human preference comparisons that helped define how chat assistants are trained to be helpful and harmless.

    Language & textAnthropicFree