Update app.py overview with ORPO paper link and revised description

app.py

@@ -113,13 +113,13 @@ collections = create_update_collections(datasets)
 languages = list(datasets.keys())
 
 overview = """
-This Space shows an overview of Direct Preference Optimization (DPO) datasets available on the Hugging Face Hub across different languages.
+This Space shows an overview of preference datasets, in particular DPO style datasets, available on the Hugging Face Hub across different languages
+Recently, [Odds Ratio Preference Optimization](https://huggingface.co/papers/2403.07691) ORPO has been demonstrated to be a powerful tool for training better performing language models directly from preference datasets. 
 
-Recently [Odds Ratio Preference Optimization](https://huggingface.co/papers/2403.07691) ORPO has been demonstrated to be a powerful tool for training better performing language models. 
-
-- ORPO training can be done using DPO style datasets 
+- ORPO can be done using DPO style datasets 
 - Is a key ingredient for training better models for every language having enough DPO datasets for different languages?
-- This Space aims to track the number DPO datasets are available for different languages and how many datasets are available for each language!"""
+- This Space aims to track the number DPO datasets available on the Hugging Face Hub for different languages. 
+"""
 
 dpo = """
 #### What is Direct Preference Optimization (DPO)?