How the backbone of coherence score is trained？

[Yisen-Feng]

I truly appreciate the fantastic work you've done, and I'm grateful for the shared code. However, as I attempt to apply the provided method to a different dataset, I've encountered a challenge: How can I obtain the coherence score?

I'm interested in understanding more about the training process for the coherence score's backbone. Specifically, could you provide additional details on how it is trained?Will the parameters of the backbone be released?

[Loewen-Hob]

I also want to know how to train, looking forward to it!