[Request ]Fine-tuned BERT version used for discrimination

[zainsarwar865]

I am conducting research on deep fake text and currently trying to replicate the results on the other models used as discriminators in the paper. The paper mentions a fine-tuned version of BERT in which you extended the maximum sequence length to 1024 by initializing new encodings.

Can you please upload those models? Also, could you tell what top-p threshold was used to generate text fed to BERT for discrimination?

Thanks!