OpenChatKit - The first open-source ChatGPT

🤖⚡️ Discover OpenChatKit, the first open-source ChatGPT! 🌐🚀 Create specialized or general-purpose chatbots with a fine-tuned 20B model. Exciting features like live-updating sources, moderation, and seamless integration await you. Join the AI conversation today! #OpenChatKit #ChatGPT #AItools

OpenChatKit is an open-source project providing a powerful base for creating specialized and general-purpose chatbots, utilizing a 20 billion-parameter model fine-tuned for chat tasks.
The project collaborated with LAION and Ontocord to create the OIG-43M dataset used in training, focusing on various natural language tasks like dialogue, question answering, classification, extraction, and summarization.
It features an extensible retrieval system that augments responses with information from live-updating sources to provide up-to-date context.
OpenChatKit supports a moderation model to filter inappropriate or out-of-domain questions and offers sample code for integrating with APIs and web search.
Collaborators behind OpenChatKit are Together, LAION, and Ontocord, known for open-source foundation models, data annotation services, and machine learning solutions respectively.
The base model is GPT-NeoX-T-Chat-Base-20B, fine-tuned with the OIG-43M dataset, focusing on tasks like multi-turn dialogue, question answering, and more.
OpenChatKit excels in question answering, extraction, classification tasks, and performs well in few-shot prompts but requires improvements in knowledge-based closed question answering, coding tasks, and creative writing.
OpenChatKit's license is Apache License 2.0, allowing free usage, modification, and distribution, with access to source code, model weights, and datasets on GitHub and Hugging Face.
Feedback can be given through the OpenChatKit feedback app, and collaboration is encouraged on GitHub, Discord, Twitter, and Medium for sharing ideas and suggestions.