update readme

Author: oceanypt

README.md

@@ -11,12 +11,14 @@ Code for our paper [Preference Guided Reflective Sampling for Aligning Language
 
 ## Overview
 We propose a new sampling method named *Preference Guided Reflective Sampling for Aligning Language Models (PReS)*. *PReS* frames the data generation as the optimization process to a specified user preference described in natural language, such that "Can you give me a concise response without explanations?". 
+
 <img src="./figures/compare_random.png" height="200" alt="Description of Image">
 
 
 
 ## Method
 *PReS* employs a tree-based generation approach to optimize outputs aligned to user preference. 
+
 ![](./figures/method.png)