Update README.md

README.md

@@ -16,6 +16,16 @@ tags:
 
 The purpose of the experiment is to understand the quality of the pre-trained Phi-2 model. The good news is that `phi-2-sft` can follow open-ended user instructions well.
 
+## Decoding
+
+Format your prompt as 
+```
+"""### Human: {instruction}
+
+### Assistant:"""
+```
+where `instruction` is your query.
+
 ## Training
 
 The model was fine-tuned on a sanitized version of the single-turn OASST dataset linked above.