3 ms·And Llama 2's training data was less aggressively deduplicated.by minimaxir 3y agoAnd Llama 2's training data was less aggressively deduplicated.