example_cupy module
Module for timing comparison of qmatmul’s CUDA-based backends: direct vs. proposed-algorithm, each via numba-cuda and via CuPy (cuBLAS-backed). Runs a single fixed-size quaternion matrix-matrix product (configurable via the module-level constants below; defaults to the dense-layer dimensions from the QCNN example) with a warm-up call per approach followed by repeated, timed runs, reporting the median.