Zero-shot Conversational Summarization Evaluations with small Large Language Models