Update src/about.py

src/about.py

@@ -41,26 +41,29 @@ Addressing the gaps in existing LLM evaluation frameworks, this benchmark is spe
     2. Synthetically generated data (newly created for Persian LLMs)
     3. Naturally collected data (reflecting indigenous cultural nuances)
     
-### Key Datasets in the Benchmark
+## Key Datasets in the Benchmark
 > The benchmark integrates the following datasets to ensure a robust evaluation of Persian LLMs:
+>
 > **Translated Datasets**
 >    • Anthropic-fa
 >    • AdvBench-fa
->   • HarmBench-fa
+>    • HarmBench-fa
 >    • DecodingTrust-fa
+>
 > **Newly Developed Persian Datasets**
 >    • ProhibiBench-fa: Evaluates harmful and prohibited content in Persian culture.
 >    • SafeBench-fa: Assesses safety in generated outputs.
 >    • FairBench-fa: Measures bias mitigation in Persian LLMs.
 >    • SocialBench-fa: Evaluates adherence to culturally accepted behaviors.
+>
 > **Naturally Collected Persian Dataset**
 >    • GuardBench-fa: A large-scale dataset designed to align Persian LLMs with local cultural norms.
 
 ### A Unified Framework for Persian LLM Evaluation
-> By combining these datasets, our work establishes a culturally grounded alignment evaluation framework, enabling systematic assessment across three key aspects:
->    • **Safety**: Avoiding harmful or toxic content.
->    • **Fairness**: Mitigating biases in model outputs.
->    • **Social Norms**: Ensuring culturally appropriate behavior.
+ By combining these datasets, our work establishes a culturally grounded alignment evaluation framework, enabling systematic assessment across three key aspects:
+    • **Safety**: Avoiding harmful or toxic content.
+    • **Fairness**: Mitigating biases in model outputs.
+    • **Social Norms**: Ensuring culturally appropriate behavior.
 
     
 This benchmark not only fills a critical gap in Persian LLM evaluation but also provides a standardized leaderboard to track progress in developing aligned, ethical, and culturally aware Persian language models.