I probably need to benchmark iocaine locally to make sure, but I figure on my local machine, it'll likely hit similar numbers. Markov chains are pretty fast, and this is entirely CPU driven.
LLMs need entire GPUs just to output tens/hundreds of tokens per second.