How to reproduce results on name disambigution

[GentleZhu]

Hi,

Thanks for the great work. I have some troubles on reproducing the name disambiguation results on whoiswho v3 . I am using validation set and the connected component on output embedding cosine similarity matrix, where most of the paper will be in single cluster. Also can you release the evaluation metric of macro pairwise F1, I am getting 0.02 on whoiswho validation using "v2-sim" of OAGBERT