Update README.md

README.md

@@ -60,7 +60,7 @@ We used 99% of the 10M examples as a train set, and the remaining ~ 100K example
 
 ## Training procedure 👨🏻‍💻
 ### Preprocessing
-The texts are lowercased and tokenized using WordPiece and a shared vocabulary size of approximately 110,000. The beginning of a new document is marked with `[CLS]` and the end of one by `[CLS]`
+The texts are lowercased and tokenized using WordPiece and a shared vocabulary size of approximately 110,000. The beginning of a new document is marked with `[CLS]` and the end of one by `[SEP]`
 The details of the masking procedure for each sentence are the following:
 - 15% of the tokens are masked.
 - In 80% of the cases, the masked tokens are replaced by `[MASK]`.