During October 2025, Tianyue Zhao developed CogVLM model support for the ggml-org/llama.cpp repository, focusing on expanding visual-language capabilities. Zhao implemented GGUF and tensor mappings for the visual encoder, integrated these into the model conversion script, and added new graphs for CogVLM and CogVLM CLIP. Working primarily in C++ and Python, Zhao addressed compile-time and runtime issues by adjusting graph context and encoder configurations, ensuring the model compiled and ran as intended. This work enhanced deployment readiness and maintainability, enabling end-to-end CogVLM usage and broadening the applicability of llama.cpp for deep learning and machine learning workflows.