v0.30.0: [Build] Fix DeepGEMM CUDA 12.9 release builds (#57554)
This update resolves build failures specifically affecting DeepGEMM integration within the vLLM framework when using CUDA 12.9. It ensures compatibility for users deploying vLLM on the latest NVIDIA CUDA toolkit version.
Verified State Diff
Impact & Verification Analysis
Developers and infrastructure engineers deploying vLLM on systems configured with NVIDIA CUDA 12.9.
Ensures the vLLM project remains compatible with the latest NVIDIA software stack, preventing technical debt and enabling users to leverage the newest CUDA features for inference performance.
Full Fact Overview
The patch addresses a build-time dependency or configuration issue involving DeepGEMM, a high-performance library for General Matrix Multiply operations often used in LLM inference acceleration. By cherry-picking commit eb87980, the vLLM project maintains alignment with the CUDA 12.9 ecosystem, preventing compilation errors that would otherwise block deployment on systems utilizing this specific toolkit version.