Open-Assistant

NOTE: OpenAssistant is completed, and the project is now finished. Thank you to everyone who contributed! Check out our blog post for more information. The final published oasst2 dataset can be found on HuggingFace at OpenAssistant/oasst2.

What is Open Assistant?

Open Assistant is a project designed to provide access to a great chat-based large language model. Its goal is to foster innovation in language, similar to how stable-diffusion revolutionized art and image creation.

Useful Links

• Data Collection
• Chat
• Project Documentation

How To Try It Out

Chatting with the AI

The chat frontend is live at Open Assistant Chat. Log in and start chatting. Please provide feedback on the assistant's responses by reacting with a thumbs up or down.

Contributing to Data Collection

Access the data collection frontend at Open Assistant. Log in to start tasks that involve submitting, ranking, and labeling model prompts and responses, which directly help improve Open Assistant's capabilities.

The Vision

Open Assistant aims to build the assistant of the future, capable of meaningful work, API usage, dynamic research, and more. The focus is on making it open and accessible, allowing it to run efficiently on consumer hardware.

The Plan

Open Assistant follows a 3-step plan as outlined in the InstructGPT paper:

1. Collect High-Quality Samples: Crowdsourcing prompts and responses to create a high-quality dataset, avoiding spam and junk data.
2. Sample and Rank Completions: Users rank multiple completions of prompts, which is used to train a reward model.
3. RLHF Training: Reinforcement Learning from Human Feedback (RLHF) based on prompts and reward models.