Hacker News Reader

Out of curiosity, how much more time will it cost to run this on general CPUs of users computer?

It takes a couple seconds per reply. GPUs aren't as powerful as everyone would have it, at least not for ML; you'd be lucky to get a 10x speedup.