Enhance leaderboard display by adding a description and adjusting column widths for retrieval and reranking dataframes

app.py

@@ -4,8 +4,13 @@ from fuzzywuzzy import fuzz
 
 HEADER = """<div style="text-align: center; margin-bottom: 20px;">
     <h1>The Arabic RAG Leaderboard</h1>
-    <p style="font-size: 14px; color: #888;">The only leaderboard you will require for your RAG needs 🏆</p>
+    <p style="font-size: 16px; color: #888;">The only leaderboard you will require for your RAG needs 🏆</p>
 </div>
+
+This leaderboard presents the first comprehensive benchmark for Arabic RAG systems, evaluating both retrieval and re-ranking components. Our framework combines real-world queries with synthetic contexts in a dynamic evaluation cycle, ensuring fair and robust assessment of Arabic information retrieval systems. 
+<br>
+<br>
+For technical details, check our blog post <a href="https://huggingface.co/blog/Navid-AI/arabic-rag-leaderboard">here</a>.
 """
 
 RETRIEVAL_ABOUT_SECTION = """
@@ -107,14 +112,28 @@ def reranking_search_leaderboard(model_name, columns_to_show):
 
 def update_retrieval_columns_to_show(columns_to_show):
     global retrieval_df
-    dummy_df = retrieval_df.loc[:, columns_to_show]
-    columns_widths = [400] + [150] * (len(columns_to_show) - 1)
+    dummy_df = retrieval_df.loc[:, [col for col in retrieval_df.columns if col in columns_to_show]]
+    columns_widths = []
+    for col in dummy_df.columns:
+        if col == "Rank":
+            columns_widths.append(80)
+        elif col == "Model":
+            columns_widths.append(400)
+        else:
+            columns_widths.append(150)
     return gr.update(value=dummy_df, column_widths=columns_widths)
 
 def update_reranker_columns_to_show(columns_to_show):
     global reranking_df
-    dummy_df = reranking_df.loc[:, columns_to_show]
-    columns_widths = [400] + [150] * (len(columns_to_show) - 1)
+    dummy_df = reranking_df.loc[:, [col for col in reranking_df.columns if col in columns_to_show]]
+    columns_widths = []
+    for col in dummy_df.columns:
+        if col == "Rank":
+            columns_widths.append(80)
+        elif col == "Model":
+            columns_widths.append(400)
+        else:
+            columns_widths.append(150)
     return gr.update(value=dummy_df, column_widths=columns_widths)
 
 def main():
@@ -122,14 +141,16 @@ def main():
 
     # Prepare retrieval dataframe
     retrieval_df = load_retrieval_results(True, "Web Search Dataset (Overall Score)", ["Revision", "Precision", "Task"])
-    retrieval_columns_to_show = ["Model", "Web Search Dataset (Overall Score)", "Model Size (MB)", "Embedding Dimension", "Max Tokens", "Num Likes"]
-    
+    retrieval_columns_to_show = ["Rank", "Model", "Web Search Dataset (Overall Score)", "Model Size (MB)", "Embedding Dimension", "Max Tokens", "Num Likes"]
+    retrieval_columns_widths = [80, 400, 150, 150, 150, 150, 150]
+    retrieval_df.insert(0, "Rank", range(1, 1 + len(retrieval_df)))
     retrieval_cols = retrieval_df.columns.tolist()  # cache columns
 
     # Prepare reranking dataframe
     reranking_df = load_reranking_results(True, sort_col="Overall Score", drop_cols=["Revision", "Precision", "Task"])
-    reranking_columns_to_show = ["Model", "Overall Score", "Model Parameters (in Millions)", "Embedding Dimensions", "Downloads Last Month", "MRR", "nDCG", "MAP"]
-    reranking_columns_widths = [400, 150, 150, 150, 150, 150, 150]
+    reranking_columns_to_show = ["Rank", "Model", "Overall Score", "Model Parameters (in Millions)", "Embedding Dimensions", "Downloads Last Month", "MRR", "nDCG", "MAP"]
+    reranking_columns_widths = [80, 400, 150, 150, 150, 150, 150, 150, 150]
+    reranking_df.insert(0, "Rank", range(1, 1 + len(reranking_df)))
     reranking_cols = reranking_df.columns.tolist()  # cache columns
 
     with gr.Blocks() as demo:
@@ -153,12 +174,12 @@ def main():
                             )
 
                         retrieval_leaderboard = gr.Dataframe(
-                            value=retrieval_df[retrieval_columns_to_show],
+                            value=retrieval_df.loc[:, retrieval_columns_to_show],
                             datatype="markdown",
                             wrap=False,
                             show_fullscreen_button=True,
                             interactive=False,
-                            column_widths=reranking_columns_widths
+                            column_widths=retrieval_columns_widths
                         )
                         
                         # Submit the search box and the leaderboard