OpenChatKit logo

OpenChatKit

Free

9.0k

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
Company
Together

About OpenChatKit

OpenChatKit is an open-source toolkit for building chat models. It includes instruction-tuned language models (Pythia-Chat-Base-7B, GPT-NeoXT-Chat-Base-20B), a moderation model, and an extensible retrieval system for up-to-date responses. The kit provides training and fine-tuning code, an inference test harness, and was trained on the OIG-43M dataset in collaboration with Together, LAION, and Ontocord.ai. All code is available under the Apache 2.0 license.

Key Features

Instruction-tuned language models (Pythia-Chat-Base-7B, GPT-NeoXT-Chat-Base-20B)
Moderation model for safer outputs
Extensible retrieval system for up-to-date responses
Training and fine-tuning code for custom models
Fine-tuning support for Llama-2-7B-32K-beta
Command-line inference test harness
Built on the OIG-43M training dataset
Apache 2.0 license

Pros & Cons

Pros
  • Completely free and open-source under Apache 2.0
  • Includes a moderation model for safer outputs
  • Extensible retrieval system enables up-to-date responses
  • Provides pre-trained chat models (Pythia-Chat-Base-7B, GPT-NeoXT-Chat-Base-20B)
  • Training and fine-tuning code available for customization
  • Trained on the large OIG-43M dataset
Cons
  • Requires technical expertise to set up and use
  • Large model sizes require substantial computational resources
  • Documentation is limited to GitHub README
  • Community support only, no official customer support

Best For

Building custom chatbots for customer supportCreating domain-specific conversational AIResearch and experimentation in open-source language modelsDeveloping retrieval-augmented generation systemsEducational projects for learning about chat model training

FAQ

What models are included in OpenChatKit?
OpenChatKit includes Pythia-Chat-Base-7B, GPT-NeoXT-Chat-Base-20B, and fine-tuning code for Llama-2-7B-32K-beta.
What license is OpenChatKit released under?
OpenChatKit is released under the Apache 2.0 license.
How can I get started with OpenChatKit?
You can get started by installing dependencies (Miniconda, Git LFS), creating an environment from the environment.yml file, and running the inference bot with 'python inference/bot.py --model togethercomputer/Pythia-Chat-Base-7B'.
What data was OpenChatKit trained on?
OpenChatKit was trained on the OIG-43M training dataset, a collaboration between Together, LAION, and Ontocord.ai.
Does OpenChatKit include a moderation model?
Yes, OpenChatKit includes a moderation model for filtering outputs.