Hacker News Reader

It's just a 27b model finetuned with LoRA?

that's a statement (as a rebuttal), not a question, right?

The question was implicit, as I'm not sure what is particularly expensive about that, since you presumably could do it with a single GPU