Once #85 is done, we should build an alternate scorer that uses MLM's like BERT. this code exposes a perplexity-like score for MLMs, it really seems not bad.
This wouldn't actually live in replacy but in its own repo, so see https://github.com/Qordobacode/replacy_article_agreer for inspiration
Once #85 is done, we should build an alternate scorer that uses MLM's like BERT. this code exposes a perplexity-like score for MLMs, it really seems not bad.
This wouldn't actually live in replacy but in its own repo, so see https://github.com/Qordobacode/replacy_article_agreer for inspiration