Update app.py

app.py

@@ -402,13 +402,17 @@ Answer:
 """
 
 
+import re
+
 def clean_response(response_text):
     # Remove any metadata-like information and focus on the main content
-    cleaned_response = re.sub(r'\[Document\(.*?\),?|\{.*?\}', '', response_text, flags=re.DOTALL)
+    # Removes "Document(metadata=...)" and other similar patterns
+    cleaned_response = re.sub(r'Document\(metadata=.*?\),?\s*', '', response_text, flags=re.DOTALL)
+    cleaned_response = re.sub(r'page_content=".*?"\),?', '', cleaned_response, flags=re.DOTALL)
+    cleaned_response = re.sub(r'\[.*?\]', '', cleaned_response, flags=re.DOTALL)  # Remove content in brackets
     cleaned_response = re.sub(r'\s+', ' ', cleaned_response).strip()
     return cleaned_response
 
-
 def generate_answer(message, choice, retrieval_mode, selected_model):
     logging.debug(f"generate_answer called with choice: {choice} and retrieval_mode: {retrieval_mode}")
 
@@ -449,10 +453,10 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
                 logging.debug(f"Phi-3.5 Prompt: {prompt}")
 
                 response = selected_model(prompt, **{
-                    "max_new_tokens": 300,
+                    "max_new_tokens": 512,  # Increased to handle longer responses
                     "return_full_text": False,
-                    "temperature": 0.5,
-                    "do_sample": False,
+                    "temperature": 0.7,  # Adjusted to avoid cutting off
+                    "do_sample": True,   # Allow sampling to increase response diversity
                 })
 
                 if response: