Update README.md

README.md

@@ -17,6 +17,7 @@ Additionally, we expanded the independent evaluation axis to include harmlessnes
 ```
 prefix: Advice-seeking Question
 suffix: **List** of Answer Advice for each Question (prefix also is a form of the list but duplicated for efficient coding)
+sft_index: The response used for SFT post-training in the list.
 reward: Upvotes score of each advice in the list(=answer=response)
 label: 'safe' means those QAs are safe. 'unsafe' means those QAs are not safe. (These labels are automated labelled from LPT/ULPT forums.)
 ```