This was an attempt to use Zeta v098, a GPU BestFirst-search with thousands of parallel threads, with ANNs, but one single thread is too slow for ANN inference.