Update app.py

app.py

@@ -23,15 +23,22 @@ embeddings_path = 'corpus_embeddings.pt'
 corpus_embeddings = torch.load(embeddings_path, map_location=torch.device('cpu'))
 
 def semantic_search(query):
+    print("Data columns:", data.columns)
+    print("First few rows:", data.head())
     query_embedding = model.encode(query, convert_to_tensor=True)
-    search_hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=5)
+    search_hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=3)
     
     results = []
     for hit in search_hits[0]:
-        text = " ".join(data.iloc[hit['corpus_id']]['tokens'])
-        results.append(f"Score: {hit['score']:.4f} - Text: {text}")
+        # 使用更安全的方法來訪問數據
+        row = data.iloc[hit['corpus_id']]
+        if 'tokens' in row:
+            text = " ".join(row['tokens'])
+        else:
+            text = str(row)  # 如果沒有 'tokens'，就轉換整行為字符串
+        results.append((hit['score'], text))
     
-    return "\n".join(results)
+    return results
 
 iface = gr.Interface(
     fn=semantic_search,