So this is apparently an "open" (you have to apply) model which outperforms GPT-3 and can be run on 4x3090's. Seems like a pretty big step forward. I'm especial

So this is apparently an "open" (you have to apply) model which outperforms GPT-3 and can be run on 4x3090's. Seems like a pretty big step forward. I'm especially interested that they got it down to INT4 for inference.