ChatGLM Introduction

ChatGLM models are based on the General Language Model (GLM) architecture and utilize advanced training techniques such as supervised fine-tuning, feedback bootstrapping, and reinforcement learning with human feedback. The latest ChatGLM3-6B incorporates a more diverse training dataset, extended training steps, and improved training strategies. It supports multi-turn dialogues and introduces new features like tool invocation (Function Call), code execution (Code Interpreter), and complex Agent tasks. The models can be deployed on consumer-grade hardware thanks to quantization techniques, requiring as little as 6GB of GPU memory for the INT4 quantization level. ChatGLM also offers different versions optimized for specific tasks, such as long-text dialogue (ChatGLM3-6B-32K) and a base model (ChatGLM3-6B-Base) for further fine-tuning.