OpenChatKit

lastmod 2023-04-19created 2023-03-142 edits

An open-source project for implementing ChatGPT

Features

• Source code, model weights, and the full training dataset released under Apache-2.0
• Four main components
  • Instruction-Tuned LLM: GPT-NEOX-20B fine-tuned on a set of 43 million instructions
  • Customization Recipe: lets you fine-tune the model for your own tasks
  • Extensible Retreival System: augments responses with information pulled from document repos, APIs, and other sources that can be updated in real time at inference time
  • Moderation Model: GPT-JT-6B fine-tuned and designed to filter which questions the bot will answer
• Collaboration between LAION and Ontocord: the OIG dataset containing 43 million instructions (released separately)

See also

• ChatGPT
• RedPajama - An open-source project that recreates the LLaMA dataset

Favorite site

• Github - togethercomputer/OpenChatKit
• Announcing OpenChatKit — TOGETHER