Update app.py

app.py

@@ -69,18 +69,18 @@ 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, max_tokens=1000, max_chunks=5):
+def generate_chunked_response(prompt, max_tokens=1000, max_chunks=5, temperature=0.7, repetition_penalty=1.1):
     API_URL = "https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.3"
     headers = {"Authorization": f"Bearer {huggingface_token}"}
     payload = {
         "inputs": prompt,
         "parameters": {
             "max_new_tokens": max_tokens,
-            "temperature": 0.7,
-            "top_p": 0.95,
+            "temperature": temperature,
+            "top_p": 0.4,
             "top_k": 40,
-            "repetition_penalty": 1.1,
-            "stop": ["</s>", "[/INST]"]  # Add stop tokens
+            "repetition_penalty": repetition_penalty,
+            "stop": ["</s>", "[/INST]"]
         }
     }
     
@@ -175,13 +175,13 @@ After writing the document, please provide a list of sources used in your respon
     
     return main_content, sources
 
-def chatbot_interface(message, history, use_web_search):
+def chatbot_interface(message, history, use_web_search, temperature, repetition_penalty):
     if use_web_search:
-        main_content, sources = get_response_with_search(message)
+        main_content, sources = get_response_with_search(message, temperature, repetition_penalty)
         formatted_response = f"{main_content}\n\nSources:\n{sources}"
     else:
-        response = get_response_from_pdf(message)
-        formatted_response = response  # No sources for PDF responses
+        response = get_response_from_pdf(message, temperature, repetition_penalty)
+        formatted_response = response
 
     history.append((message, formatted_response))
     return history
@@ -201,6 +201,11 @@ with gr.Blocks() as demo:
     chatbot = gr.Chatbot(label="Conversation")
     msg = gr.Textbox(label="Ask a question")
     use_web_search = gr.Checkbox(label="Use Web Search", value=False)
+    
+    with gr.Row():
+        temperature_slider = gr.Slider(minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Temperature")
+        repetition_penalty_slider = gr.Slider(minimum=1.0, maximum=2.0, value=1.1, step=0.1, label="Repetition Penalty")
+    
     submit = gr.Button("Submit")
 
     gr.Examples(
@@ -213,8 +218,12 @@ with gr.Blocks() as demo:
         inputs=msg,
     )
 
-    submit.click(chatbot_interface, inputs=[msg, chatbot, use_web_search], outputs=[chatbot])
-    msg.submit(chatbot_interface, inputs=[msg, chatbot, use_web_search], outputs=[chatbot])
+    submit.click(chatbot_interface, 
+                 inputs=[msg, chatbot, use_web_search, temperature_slider, repetition_penalty_slider], 
+                 outputs=[chatbot])
+    msg.submit(chatbot_interface, 
+               inputs=[msg, chatbot, use_web_search, temperature_slider, repetition_penalty_slider], 
+               outputs=[chatbot])
 
     gr.Markdown(
     """
@@ -223,7 +232,8 @@ with gr.Blocks() as demo:
     2. Select the PDF parser (pypdf or llamaparse) and click "Upload Document" to update the vector store.
     3. Ask questions in the textbox. 
     4. Toggle "Use Web Search" to switch between PDF chat and web search.
-    5. Click "Submit" or press Enter to get a response.
+    5. Adjust Temperature and Repetition Penalty sliders to fine-tune the response generation.
+    6. Click "Submit" or press Enter to get a response.
     """
     )