There’s a similar tch-rs project wrapping libtorch, and in general, deeclaring neural networks is particularly intuitive in functional languages.