Test data leakage & minor questions

[Hanzhang-lang]

Thanks for your work. Your research is immensely valuable and has been of significant assistance to me. However, I have a few questions regarding your article that I hope you could clarify.

1. In the process of reasoning, you mentioned the use of Large Language Models (LLMs). I am curious about the potential impact of test data leakage in the LLM training data. I observed that some common-sense questions might be answerable directly through the language model without the need for a knowledge graph model. If so, do you consider this issue in your work?
2. I understand that the reasoning and planning processes can be decoupled. Have you explored the possibility of employing a zero-shot learning approach without fine-tuning? I am interested to know if such an approach has been tested and what the outcomes were, if any.

Your insights on these matters would be greatly appreciated.

[lijiabao2]

You can find the answer you want in this article: "simple is effective: the roles of..."

[YRVGFO9588]

Greetings, I hope you are all doing well.

The link of pre-trained weights: https://huggingface.co/rmanluo/RoG can not be found. I hope you can update it.