Update app.py

app.py

@@ -353,15 +353,32 @@ QA_CHAIN_PROMPT_2 = PromptTemplate(input_variables=["context", "question"], temp
 # Sure! Here's the information:
 # """
 
+# phi_custom_template = """
+# <|system|>
+# Sei un esperto della lingua italiana e un madrelingua italiano. Il tuo compito è fornire risposte concise, dirette e brevi basate sul documento fornito. Dovresti restituire le informazioni nel seguente formato:
+# - Nome del documento: (il nome del documento)
+# - Numero di pagina: (numero di pagina)
+# - Contenuto effettivo: (contenuto rilevante del documento)
+# Alla fine, fornisci una sezione separata per la risposta nel seguente formato:
+# - Risposta: (la risposta alla domanda)
+# Se non riesci a trovare la risposta nel documento, rispondi semplicemente con "Questa domanda è al di là delle mie conoscenze". Ecco i dettagli del documento da considerare:
+# <|end|>
+# <|user|>
+# {context}
+# Question: {question}<|end|>
+# <|assistant|>
+# Sure! The Responses are as follows:
+# """
+
 phi_custom_template = """
 <|system|>
-Sei un esperto della lingua italiana e un madrelingua italiano. Il tuo compito è fornire risposte concise, dirette e brevi basate sul documento fornito. Dovresti restituire le informazioni nel seguente formato:
-- Nome del documento: (il nome del documento)
-- Numero di pagina: (numero di pagina)
-- Contenuto effettivo: (contenuto rilevante del documento)
-Alla fine, fornisci una sezione separata per la risposta nel seguente formato:
-- Risposta: (la risposta alla domanda)
-Se non riesci a trovare la risposta nel documento, rispondi semplicemente con "Questa domanda è al di là delle mie conoscenze". Ecco i dettagli del documento da considerare:
+You are an expert in the Italian language and a native Italian speaker. Your task is to provide concise, direct and short answers based on the document provided. You should return the information in the following format:
+- Document name: (the name of the document)
+- Page number: (page number)
+- Top 5 Relevant Answer 
+-Actual Answer 
+
+If you can't find the answer in the document, simply reply with "This question is beyond my knowledge." Here are the document details to consider:
 <|end|>
 <|user|>
 {context}
@@ -604,55 +621,6 @@ import re
 
 #     return final_response
 
-# def clean_response(response_text):
-#     # Remove system and user tags
-#     response_text = re.sub(r'<\|system\|>.*?<\|end\|>', '', response_text, flags=re.DOTALL)
-#     response_text = re.sub(r'<\|user\|>.*?<\|end\|>', '', response_text, flags=re.DOTALL)
-#     response_text = re.sub(r'<\|assistant\|>', '', response_text, flags=re.DOTALL)
-
-#     # Extract the document name and page number
-#     document_match = re.search(r"Document\(metadata=\{'source': '(.+?)', 'page': (\d+)\}", response_text)
-    
-#     if document_match:
-#         document_name = document_match.group(1).split('/')[-1]  # Get the document name
-#         page_number = document_match.group(2)  # Get the page number
-#     else:
-#         document_name = "Unknown"
-#         page_number = "Unknown"
-
-#     # Remove the entire 'Document(metadata=...' and any mention of it from the response
-#     response_text = re.sub(r'Document\(metadata=\{.*?\},page_content=', '', response_text, flags=re.DOTALL)
-
-#     # Extract top 5 fetched results (assuming the top 5 results are the first 5 lines of content)
-#     top_results = response_text.split('\n')[:5]  # Adjust this as per your actual data structure
-
-#     # Remove any unwanted escape characters like \u and \u00
-#     response_text = re.sub(r'\\u[0-9A-Fa-f]{4}', '', response_text)
-
-#     # Ensure proper spacing between words and dates
-#     response_text = re.sub(r'([a-zA-Z])(\d)', r'\1 \2', response_text)
-#     response_text = re.sub(r'(\d)([a-zA-Z])', r'\1 \2', response_text)
-
-#     # Remove the phrase "Sure! The Responses are as follows:" from the actual content
-#     response_text = re.sub(r'Sure! The Responses are as follows:', '', response_text).strip()
-
-#     # Clean up the text by removing extra whitespace
-#     cleaned_response = re.sub(r'\s+', ' ', response_text).strip()
-
-#     # Format the final response with bullet points
-#     top_five_formatted = ''.join([f'{i+1}. {result.strip()}\n' for i, result in enumerate(top_results)])
-
-#     final_response = (
-#         f"Sure! Here is the response for your Query:\n"
-#         f"• Document name - {document_name}\n"
-#         f"• Page No - {page_number}\n"
-#         f"• Top 5 Fetched Results:\n{top_five_formatted}"
-#         f"• Actual Response - {cleaned_response}"
-#     )
-
-#     return final_response
-
-
 def clean_response(response_text):
     # Remove system and user tags
     response_text = re.sub(r'<\|system\|>.*?<\|end\|>', '', response_text, flags=re.DOTALL)
@@ -672,35 +640,31 @@ def clean_response(response_text):
     # Remove the entire 'Document(metadata=...' and any mention of it from the response
     response_text = re.sub(r'Document\(metadata=\{.*?\},page_content=', '', response_text, flags=re.DOTALL)
 
-    # Extract the actual content from the response
-    content_match = re.search(r'page_content="(.+?)"', response_text)
-    if content_match:
-        actual_content = content_match.group(1)
-    else:
-        actual_content = "No content available."
-
-    # Limit the actual content to a short, precise snippet
-    actual_content = actual_content[:200] + "..." if len(actual_content) > 200 else actual_content
-
-    # Extract top 5 fetched results (based on some identifier you have)
+    # Extract top 5 fetched results (assuming the top 5 results are the first 5 lines of content)
     top_results = response_text.split('\n')[:5]  # Adjust this as per your actual data structure
 
     # Remove any unwanted escape characters like \u and \u00
     response_text = re.sub(r'\\u[0-9A-Fa-f]{4}', '', response_text)
 
+    # Ensure proper spacing between words and dates
+    response_text = re.sub(r'([a-zA-Z])(\d)', r'\1 \2', response_text)
+    response_text = re.sub(r'(\d)([a-zA-Z])', r'\1 \2', response_text)
+
+    # Remove the phrase "Sure! The Responses are as follows:" from the actual content
+    response_text = re.sub(r'Sure! The Responses are as follows:', '', response_text).strip()
+
     # Clean up the text by removing extra whitespace
     cleaned_response = re.sub(r'\s+', ' ', response_text).strip()
 
     # Format the final response with bullet points
     top_five_formatted = ''.join([f'{i+1}. {result.strip()}\n' for i, result in enumerate(top_results)])
 
-    # Final formatted response
     final_response = (
         f"Sure! Here is the response for your Query:\n"
         f"• Document name - {document_name}\n"
         f"• Page No - {page_number}\n"
-        f"• Top 5 Fetched Results:\n{top_five_formatted if top_five_formatted.strip() else 'No results found.'}"
-        f"• Actual Response - {actual_content}"
+        f"• Top 5 Fetched Results:\n{top_five_formatted}"
+        f"• Actual Response - {cleaned_response}"
     )
 
     return final_response
@@ -709,6 +673,9 @@ def clean_response(response_text):
 
 
 
+
+
+
 # Define a new template specifically for GPT-4o-mini in VDB Details mode
 gpt4o_mini_template_details = f"""
 As a highly specialized assistant, I provide precise, detailed, and informative responses. On this bright day of {current_date}, I'm equipped to assist with all your queries about Birmingham, Alabama, offering detailed insights tailored to your needs.