I often see awesome papers but don’t see any way to replicate it short of coding the thing from scratch (if it actually worked). I think many of these are conce

I often see awesome papers but don’t see any way to replicate it short of coding the thing from scratch (if it actually worked). I think many of these are concepts that work in very controlled circumstances where the inference setup is probably well calibrated for the setting. I’d venture getting a more general purpose library or tool would be far more work.