Worked on enhancing GPU-enabled build reliability for the bentoml/BentoML repository by updating CUDA base image compatibility for Debian distributions. Addressed the challenge of supporting newer CUDA versions, specifically 12.6.x and 12.8.x, by implementing dynamic image selection logic and updating the default to CUDA 12.8.1. Developed robust resolution mechanisms to ensure consistent GPU support across various Debian builds, resolving previously non-functional configurations. Added comprehensive unit tests for CUDA image resolution and version mapping, improving maintainability and guarding against regressions. Utilized Python development, containerization, and unit testing to enable smoother deployment of GPU inference workloads on modern NVIDIA hardware.