Browse Datasets

OASST1

OpenAssistant Conversations for RLHF

rlhf dialogue preference-data 161k messages Apache 2.0
Access Dataset

Description

Human-generated, human-annotated assistant-style conversation corpus with quality ratings. Useful for training reward models.

Quick Start

Python 
from datasets import load_dataset

dataset = load_dataset("OpenAssistant/oasst1", split="train")

Project Ideas

Swahili RLHF

Translate OASST1 prompts to Swahili for RLHF dataset

Advanced

Cross-lingual Reward Model

Train reward model on English, test transfer to African languages

Advanced

Ethical Considerations

  • User-generated content