Update README.md

README.md

@@ -58,7 +58,10 @@ What's your evaluation based on the above unsafe content guidelines?
 
 ## Training and evaluation data
 
-More information needed
+The finetuning is comprised of three steps:
+1. Apply LLaMA-2-70B-chat to generate responses to harmless dataset from Anthropic
+2. Apply LLaMA-2-70B-chat and Chatgpt 3.5 to evaluate the (question, answer) pairs generated in Step 1 to make dataset for finetuning
+3. Apply the evaluation dataset from Step 2 to finetune LLaMA-2-7B-chat model using int8 quantization and Low-Rank Adaptation (LoRA)
 
 ## Training procedure