Thanks for the quick feedback. However I wasn't quite sure what do you mean by "require the provision of options themselves for the test" as well as the missing category in ttqa

As for #1, we will do some ablation study first and check whether newline affects LLMs readings first

Thanks for the quick feedback. However I wasn't quite sure what do you mean by "require the provision of options themselves for the test" as well as the missing category in ttqa

As for #1, we will do some ablation study first and check whether newline affects LLMs readings first

Thank you for the response.
what does "require the provision of options themselves for the test" mean, some questions have options, while others do not.
It might be helpful to provide an example to clarify the prompt structure required for evaluation.

If no documentation is provided, it may mislead the user.

for each q in subjects:
  return """{q.question}
A: {q.A}
B: {q.B}
C: {q.C}
D: {q.D}"""

@ZoneTwelve we do have an evaluation library as well as a preprint paper coming soon, which would include the prompting method we use.

This is one of the prompts we used for one of the evaluated model:

System : 你是一位專業的中文AI主力，以下是關於閩南語考試單選題，請選出正確的答案。<｜end▁of▁sentence｜> USER: 台語：「三碗飯，兩碗菜湯。」的意思是指？
A. 翁仔某互相幫贊扶持
B. 營養著愛平衡愛顧予夠
C. 飼囝無論飯，飼爸母算頓
D. 食飯皇帝大
答案：<｜end▁of▁sentence｜> ASSISTANT:

We follow the prompting strategy by C-eval, except text is written in traditional chinese

@ZoneTwelve The evaluation code is now public, feel free to try it out : https://github.com/ikala-corp/ievals

@ikala-ray Thank you for notifying me. I will take a look at your amazing work.

I hope you guys can keep this issue open until #1 is fully resolved.
That's very important to us to track the progress and ensure that all related problems are addressed before closing this matter.

Here are the rest of the issues that need to be addressed

• did newline affect LLMs readings?
• TTQA v2's categories are missing.

did newline affect LLMs readings?

depends on your LLM, some matters some doesn't. It would be nice if you could do the study as it's not covered in our ablation study.

TTQA v2's categories are missing.

we decided not to include categories as its not really part of our core contributions. If you want just refer to the original dataset

No problem, I'll be sure to inform everyone immediately if I come across any relevant studies.

I still appreciate your guy's amazing work.

Closed by as no plans.