Also, their metric table is very interesting. It shows Falcon 7B and OpenLlama 7B much less favorably than other evaluations (including the HuggingFace leaderbo

Also, their metric table is very interesting. It shows Falcon 7B and OpenLlama 7B much less favorably than other evaluations (including the HuggingFace leaderboard, which I am kinda suspicious of), and instruct benchmarks like that aren't seen as much.