update readme

README.md

@@ -1 +1 @@
-bert-base-cased model trained on quora question pair dataset. The model achieves 89% evaluation accuracy
+bert-base-cased model trained on quora question pair dataset. The task requires to predict whether the two given sentences (or questions) are `not_duplicate` (label 0) or `duplicate` (label 1). The model achieves 89% evaluation accuracy