Any paper that's trained more than one model size on the same data affirms the same thing. Llama 13b was better than 7b and Llama 66b was better than 33b. If

Any paper that's trained more than one model size on the same data affirms the same thing. Llama 13b was better than 7b and Llama 66b was better than 33b. If you're bothered with how general a statement I'm making then Ok, point is that all training so far has pointed towards that.