Experimenting with Zero-Knowledge Proofs of Training