F-score on CORD dataset

[taosong2019]

Thanks for the excellent work! I am trying to reproduce the result on CORD dataset. However, I find the f-score results in your paper are somewhat different from that in LayoutLMv2 paper. Specifically, LayoutLMv2*-base achieves 96.05 and LayoutLMv2*-large achieves 97.24 in your paper. While in LayoutLMv2 paper, LayoutLMv2-base achieves 94.95 and LayoutLMv2-large achieves 96.01. Could you give an example of BROS fine-tuning on CORD dataset? Thanks!