Congrats. I'm late to the party but this looks promising! By any chance, can you provide more info on the performance achieved? I see a pickled model. Do you guys have a sample hello world code you could add to the repo for inference to optimally exploit it? I see various use cases! Thanks a lot!