Update README.md

README.md

@@ -25,6 +25,6 @@ configs:
     path: data/train-*
 ---
 
-This is a cleaned version from the Quora dataset that solves a single problem with the original dataset - sentences can have multiple ids. 
+This is a cleaned version from the Quora dataset that solves a single problem with the original dataset - sentences can have multiple ids. We also removed 2 rows from the original dataset that contained an empty string - this causes problem when trying to run an embedding because a empty string cannot... be embedded by OpenAI and has no valid representation?
 
 Created to minimise data leakage from your train, test and validation sets by allowing you to segregate and split by ID.