Worked on the pytorch-labs/tritonbench repository to deliver an environment-driven Warp Specialization configuration, known as AutoWS, for Triton 3.3.x compatibility. This feature replaced legacy attribute-based checks with a new approach that leverages the ENABLE_AUTO_WS environment variable, ensuring that warp specialization is determined accurately based on deployment context. The implementation focused on environment configuration, library integration, and performance optimization using Python. By reducing configuration drift and improving reliability, the update enhanced inference performance predictability across different environments. The work demonstrated a methodical approach to maintaining compatibility and deployment reliability in evolving machine learning infrastructure scenarios.