Closed chncwang closed 4 years ago
hi @chncwang , sorry for a late reply, neurips deadline : )
Yes I imagine this would be possible but I am not sure how popular this demand is. Feel free to fork this repo and implement it!
Let me know if you have any question about the code.
hi @chncwang , sorry for a late reply, neurips deadline : )
Yes I imagine this would be possible but I am not sure how popular this demand is. Feel free to fork this repo and implement it!
Let me know if you have any question about the code.
Thank you all the same!
When evaluating conversation models using bert-score, a natural idea is that is is beneficial to add the source sentence before both the hypothesis and the reference to form better contextual embeddings. But when calculating cosine similarity, the source sentence should not be involved.