Do you support multi-turn conversations?

[vackosar]

In the description here: https://github.com/thunlp/ChatEval#prepare-dataset

I see that you support question-answer. But do you support multi-turn also, or how to make it work?

[chanchimin]

Actually, we did not test the performance on evaluating multi-turn conversations. You might experience performance degradation due to the long context window. However, you are free to directly substitute what you want to evaluate into the "response" slot. Please refer to the data preparation section, as you've pointed out. The text will then be inserted into the prompt template before we query the LLM.