Together, a team of former OpenAI fellows has launched the first open source ChatGPT alternative. They have released a 20B Chat-GPT model, fine-tuned for chat using EleutherAI’s GPT-NeoX-20B, which includes over 43 million instructions. The model has been released under the Apache-2.0 license.
The company provides an intuitive platform that combines data, models, and computing to assist AI researchers, developers, and businesses in better leveraging and advancing AI.
OpenChatKit 0.15 was recently released under the Apache-2.0 license, making the code, model weights, and training datasets available to the public for free. OpenChatKit’s strong, open-source foundation allows for the development of both domain-specific and general-purpose chatbots. Users can submit feedback, and community members can use the OpenChatKit tools to add new datasets, all contributing to the growing corpus of open training data, resulting in better LLMs.
Together collaborated with LAION and Ontocord to construct the dataset used for training. OpenChatKit’s chat model, which has 20 billion parameters and was trained on 43 million instructions, supports reasoning, multi-turn discussions, knowledge, and answer generation.
For a chatbot to be useful, it must be capable of regulating responses, following instructions in normal language, and maintaining context throughout the conversation. The OpenChatKit framework consists of a generic chatbot and the necessary components to develop specialized bots.
Main Points To Highlight:
- EleutherAI’s GPT-NeoX-20B, a large language model optimized for chats with over 43 million instructions on 100% carbon-negative compute, serves as the foundation for the framework.
- Documentation containing a series of customization recipes, which can be used to optimize the model’s performance to achieve high accuracy on a user’s tasks, is available as an open-source resource on GitHub under the Apache-2.0 license.
- An extensible retrieval system that allows you to add information from a document repository, API, or other live-updating information sources at inference time.
- A moderation model derived from GPT-JT-6B is accessible on HuggingFace under the Apache-2.0 license. The moderation model selects which queries the bot should answer.
Similar to other chatbots based on language models, GPT-NeoXT-Chat-Base-20B is limited in some ways. For example, the model may not provide accurate or relevant responses when confronted with novel or unclear questions that fall outside its training data. The team encourages involvement from a diverse range of groups and individuals to build a more resilient and inclusive chatbot.
Checkout Demo and Model of Open Source Alternative Of ChatGPT ‘OpenChatKit’.
Related Stories: