Worked on the bytedance-iaas/vllm repository to deliver improvements in QuarkW8A8Fp8 quantization handling, focusing on enhancing compatibility and error resilience across quantization schemes. Addressed a targeted issue in Quark ptpc by refining weight and input configuration management, which reduced runtime quantization errors and expanded support for quantized inference models. Leveraged Python and PyTorch to implement these changes, emphasizing robust error handling and deployment reliability. The work demonstrated a strong grasp of machine learning quantization techniques and effective integration within a complex codebase, ultimately enabling broader adoption and smoother deployment of quantized models in production environments.