This leaderboard is not the best for comparing model architectures, the dataset and finetuning have too much influence. I think perplexity on a particular datas

This leaderboard is not the best for comparing model architectures, the dataset and finetuning have too much influence. I think perplexity on a particular dataset would be a better way to compare