Last released Oct 9, 2026
CUDA BF16 PyTorch operators for standard and sliced low-rank Attention Residuals