Brandon Music developed SM120 support for NVFP4 Mixture-of-Experts kernels in the flashinfer-ai/flashinfer repository, expanding GPU compatibility to include RTX Blackwell hardware. He addressed architecture fragmentation by relaxing compute capability checks and enhancing JIT module generation, allowing seamless deployment on newer CUDA versions. Using CUDA and Python, Brandon improved architecture flag encoding and compute-capability detection, ensuring accurate nvcc flag generation and preserving user-defined suffixes. His work was validated through targeted benchmarks and practical tests, confirming both compatibility and performance gains. This update laid a robust foundation for future GPU families, reflecting a deep understanding of GPU programming and deep learning.