Create README.md

README.md

@@ -0,0 +1,7 @@
+## BERT-large finetuned on MNLI.
+
+The [reference finetuned model](https://github.com/google-research/bert) has an accuracy of 86.05, we get 86.7:
+
+```
+{'eval_loss': 0.3984006643295288, 'eval_accuracy': 0.8667345899133979} 
+```