Worked on expanding the NVIDIA/onnxruntime-genai repository by integrating support for both vanilla and quantized ChatGLM3 models within the Model Builder. Focused on ensuring consistent behavior and reliability across model configurations through the implementation of comprehensive parity checks. Utilized Python and applied deep learning and model optimization techniques, particularly quantization, to enhance deployment flexibility. Validated end-to-end model-builder flows to improve production readiness and reduce integration friction for customers. The work emphasized robust feature validation and cross-team collaboration, resulting in expanded model compatibility and faster time-to-value for users deploying ChatGLM3 models in diverse environments without introducing major bugs.