Update app.py

app.py

@@ -8,7 +8,7 @@ tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForSequenceClassification.from_pretrained(model_name)
 model.eval()
 
-# Function to compute relevance score (in logits) and dynamically adjust threshold
+# Function to compute relevance scores (in logits) and dynamically adjust threshold
 def get_relevance_score_and_excerpt(query, paragraph1, paragraph2, paragraph3, threshold_weight):
     # Handle empty input for paragraphs
     paragraphs = [p for p in [paragraph1, paragraph2, paragraph3] if p.strip()]
@@ -73,13 +73,19 @@ def get_relevance_score_and_excerpt(query, paragraph1, paragraph2, paragraph3, t
     # Sort paragraphs by logit (descending)
     ranked_paragraphs.sort(key=lambda x: x["logit"], reverse=True)
 
-    # Prepare output
-    relevance_scores = [round(p["logit"], 4) for p in ranked_paragraphs]
-    highlighted_texts = [p["highlighted_text"] for p in ranked_paragraphs]
+    # Prepare output: Combine scores and highlighted text in a readable format
+    output_html = "<table border='1' style='width:100%; border-collapse: collapse;'>"
+    output_html += "<tr><th style='padding: 8px;'>Relevance Score (Logits)</th><th style='padding: 8px;'>Highlighted Paragraph</th></tr>"
 
-    return "\n".join([f"Relevance Score (Logits): {score}" for score in relevance_scores]), "\n\n".join(highlighted_texts)
+    for item in ranked_paragraphs:
+        output_html += f"<tr><td style='padding: 8px; text-align: center;'>{round(item['logit'], 4)}</td>"
+        output_html += f"<td style='padding: 8px;'>{item['highlighted_text']}</td></tr>"
 
-# Define Gradio interface with a slider for threshold adjustment and ability to add multiple paragraphs
+    output_html += "</table>"
+
+    return output_html
+
+# Define Gradio interface with a slider for threshold adjustment and multiple paragraphs input
 interface = gr.Interface(
     fn=get_relevance_score_and_excerpt,
     inputs=[
@@ -90,8 +96,7 @@ interface = gr.Interface(
         gr.Slider(minimum=0.02, maximum=0.5, value=0.1, step=0.01, label="Attention Threshold")
     ],
     outputs=[
-        gr.Textbox(label="Relevance Scores (Logits)"),
-        gr.HTML(label="Highlighted Document Paragraphs")
+        gr.HTML(label="Ranked Paragraphs")
     ],
     title="Cross-Encoder Attention Highlighting with Reranking",
     description="Adjust the attention threshold to control token highlighting sensitivity. Multiple paragraphs can be added and reranked based on their logits.",