ColossalChat is a free and open-source alternative to ChatGPT introduced by Colossal-AI. Its GitHub project has already garnered over 26K stars.

Colossal AI is a leading open-source large AI model solution today, being the first company to release a complete **RLHF (Reinforcement Learning with Human Feedback)** mechanism based on LLaMA pre-trained models. This mechanism includes supervised data collection, supervised fine-tuning, reward model training, and RL fine-tuning. They have shared **ColossalChat**—an open-source project very similar to the original ChatGPT technology solution.

The open-source project ColossalChat includes the following components:

• Demo Demonstration:An interactive demo that allows users to try out the model online without requiring registration or joining a waitlist.
• Training Code:Open-source complete RLHF (Reinforcement Learning with Human Feedback) training code, including models with 70 billion and 130 billion parameters.
• Dataset:An open-source bilingual Chinese-English dataset with 104K entries, providing a diverse set of training examples.
• Inference:Allows inference for 70 billion parameter models with 4-bit quantization, requiring only 4GB of GPU memory.
• Model Weights:Easily replicable on a single server with minimal computational resources, allowing for quick deployment and usage.