Prompt generation training for large language models