Update README.md

README.md

@@ -67,7 +67,8 @@ This benchmark evaluates model performance on multiple-choice question answering
 
 We provide two versions of the prompt template, depending on the language setting.
 
-** English Prompt **
+**English Prompt**
+
 ```
 Question: {question}
 Choices:
@@ -79,7 +80,8 @@ Based on the image and the question, choose the most appropriate answer.
 **Only output a single letter (A, B, C, or D)**. Do NOT output any other text or explanation.
 ```
 
-** Chinese Prompt **
+**Chinese Prompt**
+
 ```
 问题: {question}
 选项:
@@ -92,7 +94,8 @@ D. {choice_D}
 只输出单个字母 (A, B, C 或 D)，不要输出选项内容，也不要输出任何解释。
 ```
 
-** Evaluation Protocol **
+**Evaluation Protocol**
+
 If the model’s output is not one of A, B, C, or D, we use GPT-4o to map the output to A–D based on the option content. 
 The final evaluation reports the absolute accuracy of the benchmark in both English and Chinese versions.