ChatGLM-6B is a 6.2 billion parameter bilingual language model developed by THUDM, based on the General Language Model (GLM) framework. It is optimized for natural and fluent dialogue in both Chinese and English, supporting applications in conversational AI, question answering, and assistance. Trained on approximately 1 trillion tokens, the model benefits from supervised fine-tuning, feedback self-training, and reinforcement learning with human feedback to align its outputs with human preferences. It is particularly tailored to Chinese dialogue scenarios while maintaining strong English capabilities. The model can be deployed locally on consumer-grade GPUs, requiring as little as 6GB of VRAM using INT4 quantization. ChatGLM-6B is open-source and free for academic and commercial use upon registration. It features easy-to-use APIs, sample chatbot interactions, and is backed by the GLM research family, with further upgrades available in newer versions like ChatGLM2.
Features
- 6.2B parameter bilingual model trained on Chinese and English
- Built on the GLM architecture optimized for dialogue
- Supports quantization down to INT4 for local deployment on 6GB GPUs
- Includes supervised fine-tuning and RLHF for aligned responses
- Open-source and free for commercial use after registration
- Specially optimized for Chinese Q&A and conversation fluency
- Offers easy integration via Transformers API
- Part of the larger ChatGLM series including more powerful successors
Project Samples
