Skip to content

Dataset · OpenAssistant

Open Assistant conversation trees

OpenAssistant/oasst2

Multilingual prompter-assistant message trees for training and evaluating assistants.

What's inside

OASST2 contains message trees: the root is the initial prompt, which has branch responses that can have their own replies. Roles alternate strictly between prompter and assistant. Languages include English, Spanish, Russian, German, Polish, and Thai; size ranges from 100k to 1M examples; format is tabular and text data.

What you can do with it

  • Training an SFT assistant on dialogues
  • Building a reward model for RLHF
  • Evaluating a Russian-language assistant
Languages
English, Spanish, Russian, German, Polish, Thai, Vietnamese, Swedish and 4 more
Size
100K–1M rows
Downloads this month
23.5K
Likes
315

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("OpenAssistant/oasst2")