This repository provides a pure C++ implementation of ChatGLM series models and GLM-4 based on the ggml library, similar in design to llama.cpp. It enables efficient CPU and GPU inference with memory optimization through int4 and int8 quantization, supports P-Tuning v2 and LoRA fine-tuned variants, and offers streaming generation. The project includes Python bindings, web demos, and API servers for deployment.