Update README.md

README.md

@@ -120,7 +120,7 @@ The BERT model was pretrained on pre-k to HS math curriculum (engageNY, Utah Mat
 
 #### Training procedure
 
-The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,522. The inputs of the model are then of the form:
+The texts are lowercased and tokenized using WordPiece and a customized vocabulary size of 30,522. We use the ```bert_tokenizer``` from huggingface tokenizers library to generate a custom vocab file from our training raw math texts. The inputs of the model are then of the form:
 
 ```
 [CLS] Sentence A [SEP] Sentence B [SEP]