This is likely due to Llama 3's requirement of a very specific prompt format, which is described in the Meta Llama 3 documentation at https://www.llama.com/docs/model-cards-and-prompt-formats/meta-llama-3