Resolve GPU framework library pins before choosing CUDA builds

Co-locating PyTorch and PaddlePaddle requires comparing their exact NVIDIA Python distribution requirements, beyond matching CUDA labels. Two CUDA builds can demand incompatible versions of the same distributions. Prefer a supported combination with non-conflicting distribution namespaces when available, rather than forcing installation over either framework's requirements. Then check dependency consistency and run both frameworks in the actual process and GPU environment. A clean resolver result proves package consistency, not shared-library compatibility, model parity or a reproducible deployment artifact. Record the resolved environment and verify the final deployment artifact before promotion.

Applies when: Multiple GPU frameworks share one Python inference environment. Keywords: PyTorch PaddlePaddle NVIDIA CUDA wheels dependencies pins compatibility