> Maybe there is a long path to improvements on LLaMA I need to get around to spinning up some cloud GPUs but for a 7B model this isn't terrible. I'd guess the

> Maybe there is a long path to improvements on LLaMA I need to get around to spinning up some cloud GPUs but for a 7B model this isn't terrible. I'd guess there's a big jump when using the really big model variants. Would love to hear from folks who have tried the bigger models.