Update app.py

app.py

@@ -77,7 +77,7 @@ def update_vectors(files, parser):
     
     return f"Vector store updated successfully. Processed {total_chunks} chunks from {len(files)} files using {parser}."
 
-def generate_chunked_response(prompt, model, max_tokens=1000, num_calls=3, temperature=0.2, should_stop=False):
+def generate_chunked_response(prompt, model, max_tokens=1000, num_calls=5, temperature=0.2, should_stop=False):
     print(f"Starting generate_chunked_response with {num_calls} calls")
     client = InferenceClient(model, token=huggingface_token)
     full_response = ""
@@ -172,7 +172,7 @@ def respond(message, history, model, temperature, num_calls, use_web_search):
         for partial_response, _ in get_response_from_pdf(message, model, num_calls=num_calls, temperature=temperature):
             yield partial_response
 
-def get_response_with_search(query, model, num_calls=3, temperature=0.2):
+def get_response_with_search(query, model, num_calls=5, temperature=0.2):
     search_results = duckduckgo_search(query)
     context = "\n".join(f"{result['title']}\n{result['body']}\nSource: {result['href']}\n" 
                         for result in search_results if 'body' in result)
@@ -197,7 +197,7 @@ After writing the document, please provide a list of sources used in your respon
                 main_content += chunk
                 yield main_content, ""  # Yield partial main content without sources
 
-def get_response_from_pdf(query, model, num_calls=3, temperature=0.2):
+def get_response_from_pdf(query, model, num_calls=5, temperature=0.2):
     embed = get_embeddings()
     if os.path.exists("faiss_database"):
         database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
@@ -211,7 +211,9 @@ def get_response_from_pdf(query, model, num_calls=3, temperature=0.2):
 
     prompt = f"""Using the following context from the PDF documents:
 {context_str}
-Write a detailed and complete response that answers the following user question: '{query}'"""
+Write a detailed and complete response that fully answers the following user question. 
+Ensure your response covers all relevant information and is not cut off: '{query}'
+If the response is long, please continue until you have provided a comprehensive answer."""
 
     client = InferenceClient(model, token=huggingface_token)
     
@@ -219,7 +221,7 @@ Write a detailed and complete response that answers the following user question:
     for i in range(num_calls):
         for message in client.chat_completion(
             messages=[{"role": "user", "content": prompt}],
-            max_tokens=1000,
+            max_tokens=2000,
             temperature=temperature,
             stream=True,
         ):