Comment by tom2026hn

Let me guess: the last crackdown on Hugging Face yielded better-than-expected results. They obtained the answers to the test benchmarks, and for some reason, an agent added those answers to the training set.