Optimizing an NVFP4 Blockscaled GEMM on RTX PRO 6000 Blackwell GPU (SM120)

FlashAttention-4: Algorithm and Kernel Pipelining

Categorical Foundations for Cute Layouts