Hysen LabsOpen-Assistant
OpenAssistant is a chat-based assistant that understands tasks, can interact with third-party systems, and retrieve information dynamically to do so.
Open Assistant, the finished open data project
A project that aimed to give everyone access to a chat based large language model, now completed, with its final oasst2 dataset published on Hugging Face.
Completed, and saying so
The README leads with a note in block letters: OpenAssistant is completed, and the project is now finished, with thanks to everyone who contributed. A blog post is linked for more information. The final published oasst2 dataset sits on Hugging Face at the OpenAssistant/oasst2 id. That status turns the repository into an archive of the project and its data rather than an actively developed codebase. Python is the primary language, the license is Apache-2.0, and the homepage is open-assistant.io.
What it was trying to do
The mission was straightforward: give everyone access to a great chat based large language model. The README draws a parallel to stable diffusion, which helped the world make art and images in new ways, and hopes Open Assistant can improve the world by improving language itself. It was not going to stop at replicating ChatGPT; the vision included writing emails and cover letters, doing meaningful work, using APIs, and dynamically researching information.
The data collection machinery
Two front ends carried the project. A chat frontend let people log in and chat, with a request to react with thumbs up or down on responses. A data collection frontend let users take on tasks, submitting, ranking, and labelling model prompts and responses to improve the assistant's capabilities. The original plan, documented in the README, was to follow the three steps of the InstructGPT paper, starting with high quality human generated instruction and fulfillment samples. For developers, the README describes running the full stack locally with Docker, with notes for MacOS M1 chips.
Community notes