馃殌 Feature Request
Bert Score (https://github.com/Tiiiger/bert_score) new version (0.3.1) supports oop implementation. Current vizseq uses the functional implementation which could be updated to oop implementation.
Motivation
Currently, using Bert Score in a validation loop causes re-loading the model again and again. This can be avoided with oop implementation.
Pitch
I see two solutions: (i) create a separate scorer named bert_score_oop (ii) in the current implementation of bert_score add argument whether to use oop implementation or not.
Are you willing to open a pull request?
Yes, I can send a pull request
馃殌 Feature Request
Bert Score (https://github.com/Tiiiger/bert_score) new version (
0.3.1) supports oop implementation. Currentvizsequses the functional implementation which could be updated to oop implementation.Motivation
Currently, using Bert Score in a validation loop causes re-loading the model again and again. This can be avoided with oop implementation.
Pitch
I see two solutions: (i) create a separate scorer named
bert_score_oop(ii) in the current implementation ofbert_scoreadd argument whether to use oop implementation or not.Are you willing to open a pull request?
Yes, I can send a pull request