Update app.py

app.py

@@ -240,18 +240,14 @@ def summarize_content(content, model):
     if content is None:
         return "No content available to summarize."
 
-    # Approximate the token limit using character count
-    # Assuming an average of 4 characters per token
-    max_chars = 7000 * 4  # Leave some room for the prompt
-    if len(content) > max_chars:
-        content = content[:max_chars] + "..."
-    
     summary_prompt = f"""
-    Summarize the following content concisely:
-    {content}
+    Summarize the following news article in 10-15 lines. Focus on the key points, main events, and significant details. Ensure the summary is informative and relevant to current news:
+
+    {content[:3000]}  # Limit input to avoid token limits
+
     Summary:
     """
-    summary = generate_chunked_response(model, summary_prompt, max_tokens=200)
+    summary = generate_chunked_response(model, summary_prompt, max_tokens=300)  # Adjust max_tokens as needed
     return summary
 
 def rank_search_results(titles, summaries, model):
@@ -309,7 +305,6 @@ def ask_question(question, temperature, top_p, repetition_penalty, web_search):
                     summary = summarize_content(result["text"], model)
                     processed_results.append({
                         "title": result.get("title", f"Result {index}"),
-                        "content": result["text"],
                         "summary": summary,
                         "index": index
                     })
@@ -323,24 +318,16 @@ def ask_question(question, temperature, top_p, repetition_penalty, web_search):
 
         print(f"Number of processed results: {len(processed_results)}")
 
-        # Rank the results
-        titles = [r["title"] for r in processed_results]
-        summaries = [r["summary"] for r in processed_results]
-        try:
-            ranks = rank_search_results(titles, summaries, model)
-        except Exception as e:
-            print(f"Error in ranking results: {str(e)}. Using default ranking.")
-            ranks = list(range(1, len(processed_results) + 1))
-
-        print(f"Number of ranks: {len(ranks)}")
+        # For news requests, return the summaries directly
+        if "news" in question.lower():
+            news_response = "Here are the latest news summaries on this topic:\n\n"
+            for result in processed_results[:5]:  # Limit to top 5 results
+                news_response += f"Title: {result['title']}\n\nSummary: {result['summary']}\n\n---\n\n"
+            return news_response.strip()
 
-        # Update Vector DB
-        current_date = datetime.now().strftime("%Y-%m-%d")
-        update_vector_db_with_search_results(processed_results, ranks, current_date)
-        
-        # Prepare context for the question
-        context_str = "\n\n".join([f"Title: {r['title']}\nSummary: {r['summary']}\nRank: {ranks[i]}" 
-                                   for i, r in enumerate(processed_results)])
+        # For other questions, use the summaries as context
+        context_str = "\n\n".join([f"Title: {r['title']}\nSummary: {r['summary']}" 
+                                   for r in processed_results])
         
         prompt_template = """
         Answer the question based on the following web search results:
@@ -348,10 +335,12 @@ def ask_question(question, temperature, top_p, repetition_penalty, web_search):
         {context}
         Current Question: {question}
         If the web search results don't contain relevant information, state that the information is not available in the search results.
-        Provide a concise and direct answer to the question without mentioning the web search or these instructions:
+        Provide a concise and direct answer to the question:
         """
         prompt_val = ChatPromptTemplate.from_template(prompt_template)
         formatted_prompt = prompt_val.format(context=context_str, question=question)
+        
+        answer = generate_chunked_response(model, formatted_prompt)
     else:
         if database is None:
             return "No documents available. Please upload documents or enable web search to answer questions."
@@ -368,23 +357,7 @@ def ask_question(question, temperature, top_p, repetition_penalty, web_search):
         prompt_val = ChatPromptTemplate.from_template(prompt)
         formatted_prompt = prompt_val.format(history=history_str, context=context_str, question=question)
 
-    full_response = generate_chunked_response(model, formatted_prompt)
-    
-    # Extract only the part after the last occurrence of a prompt-like sentence
-    answer_patterns = [
-        r"Provide a concise and direct answer to the question without mentioning the web search or these instructions:",
-        r"Provide a concise and direct answer to the question:",
-        r"Answer:"
-    ]
-    
-    for pattern in answer_patterns:
-        match = re.split(pattern, full_response, flags=re.IGNORECASE)
-        if len(match) > 1:
-            answer = match[-1].strip()
-            break
-    else:
-        # If no pattern is found, return the full response
-        answer = full_response.strip()
+        answer = generate_chunked_response(model, formatted_prompt)
 
     if not web_search:
         memory_database[question] = answer
@@ -530,11 +503,19 @@ with gr.Blocks() as demo:
             top_p_slider = gr.Slider(label="Top P", minimum=0.0, maximum=1.0, value=0.9, step=0.1)
             repetition_penalty_slider = gr.Slider(label="Repetition Penalty", minimum=1.0, maximum=2.0, value=1.0, step=0.1)
             web_search_checkbox = gr.Checkbox(label="Enable Web Search", value=False)
-    
     def chat(question, history, temperature, top_p, repetition_penalty, web_search):
-        answer = ask_question(question, temperature, top_p, repetition_penalty, web_search)
+    answer = ask_question(question, temperature, top_p, repetition_penalty, web_search)
+    
+    if "news" in question.lower():
+        # Split the answer into individual news items
+        news_items = answer.split("---")
+        for item in news_items:
+            if item.strip():
+                history.append((question, item.strip()))
+    else:
         history.append((question, answer))
-        return "", history
+    
+    return "", history
     
     submit_button.click(chat, inputs=[question_input, chatbot, temperature_slider, top_p_slider, repetition_penalty_slider, web_search_checkbox], outputs=[question_input, chatbot])