Khánh Duy implemented MPS device support with precision-aware execution in the vllm-project/llm-compressor repository, enabling Apple Silicon compatibility for model compression workflows. Using Python and leveraging machine learning and quantization techniques, Khánh Duy introduced device-aware precision selection across compression, fusion, and transform stages, with a fallback to float32 for unsupported MPS operations. The work included updating and expanding unit tests to validate the new precision path and ensure compatibility, as well as coordinating with dependencies to improve device warnings and parallelism handling. This engineering effort reduced runtime errors and enabled production use of compressed models on MPS devices.