chromeNLP's picture
Update README.md
c18455a
metadata
tags:
  - Issue_fixed
  - textattack
  - textclassification
  - entailment
license: mit
datasets:
  - mnli
metrics:
  - accuracy

Fixed label mapping issue for textattack/bert-base-uncased-MNLI, if using the original model, the predicted label has systematic confusion with the huggingface MNLI dataset. See the Github issue: https://github.com/QData/TextAttack/issues/684. The fixed accuracy_mm is 84.44% and is 7% before the fix applied.