It's just as fast, but not nearly as clever. I can push the context size to 120k locally, but quality of the work it delivers starts to falter above say 40k. Ge

It's just as fast, but not nearly as clever. I can push the context size to 120k locally, but quality of the work it delivers starts to falter above say 40k. Generally you have to feed it more bite-sized pieces, and keep one chat to one topic. It's definitely a step down from SOTA.