Great work by the team, I like that they re-used ArrayFire instead of Yet Another Tensor Library feels refreshing to use native code again to build ML Pipelines for these kind of tasks.