Developed MPS device support with precision-aware execution for the vllm-project/llm-compressor repository, enabling model compression workflows to run efficiently on Apple Silicon hardware. The implementation introduced device-aware precision selection throughout the compression, fusion, and transform stages, with a robust fallback to float32 for unsupported MPS operations. Python was used to update and expand unit tests, ensuring compatibility and coverage for the new precision logic. The work included end-to-end validation, confirming successful quantization and fast inference on MPS devices. Integration with compressed-tensors and improved device warnings enhanced reliability, unlocking production use of quantized models on Apple platforms.