Which report column should I use

[SinuoWang]

Hi,

Thank you for releasing the code of this impressive work!

However, I am not sure which report column should I use for pre-training. I saw "LLM Report Impression" and "xinhuo" in the csv file provided. Cuz they both look like the report generated by LLM, I am not sure which column should I choose. And could you also let me know what is the differences between these two? Thanks in advance

[laihaoran]

Hi, thanks for your interest in our work!

To clarify:

LLM Report Impression: This column was generated using Vicuna-7B, and serves as one version of automatically generated impressions.

xinhuo: This column was generated using the iFlytek Spark Cognitive Large Model.

In our paper, we used the xinhuo reports (iFlytek Spark), as they showed better quality and performance compared to the Vicuna-7B results. So if you want to reproduce the pre-training setup in our work, we recommend using the xinhuo column.

Hope this helps!