What benchmarks did they use? It seems like on larger tasks, having the LLM be familiar with the language through a large volume of training data will compactne

What benchmarks did they use? It seems like on larger tasks, having the LLM be familiar with the language through a large volume of training data will compactness and tenseness. See https://danluu.com/pl-tokens/ https://danluu.com/pl-tokens/