There's a library called paretoset that calculates Pareto frontiers. It provides both numpy and numba implementations, and self-reports benchmark values that are comparable to the fastest implementations provided here:
paretoset