[namlem]

Does this issue happen with llama models too? If you ask them to repeat a word they'll eventually leak their training data?