Update app.py

app.py

@@ -44,7 +44,7 @@ def initialize_model():
         return None, None
 
 # Generate response
-def generate_response(model, tokenizer, user_input, chat_history):
+def generate_response(model, tokenizer, user_input, chat_history, temperature=0.2, top_p=1.0, top_k=40):
     # Check if model and tokenizer are loaded
     if model is None or tokenizer is None:
         return "மாதிரி ஏற்றப்படவில்லை. 'மாதிரியை ஏற்று' பொத்தானைக் கிளிக் செய்யவும்."  # Model not loaded, please click 'Load Model' button
@@ -73,15 +73,16 @@ def generate_response(model, tokenizer, user_input, chat_history):
         # Tokenize input
         inputs = tokenizer(prompt_template, return_tensors="pt", padding=True)
         
-        # Generate response
+        # Generate response with user-specified parameters
         with torch.no_grad():
             output = model.generate(
                 inputs["input_ids"],
                 attention_mask=inputs["attention_mask"],
                 max_new_tokens=256,
                 do_sample=True,
-                temperature=0.3,
-                top_p=0.9,
+                temperature=temperature,
+                top_p=top_p,
+                top_k=top_k,
                 pad_token_id=tokenizer.eos_token_id,
                 eos_token_id=tokenizer.encode("<|im_end|>", add_special_tokens=False)[0] if "<|im_end|>" in tokenizer.get_vocab() else tokenizer.eos_token_id
             )
@@ -104,24 +105,66 @@ def generate_response(model, tokenizer, user_input, chat_history):
         print(f"Error generating response: {e}")
         return f"பிழை ஏற்பட்டது. மீண்டும் முயற்சிக்கவும்."  # Error occurred, please try again
 
+# Function to vote/like a response
+def vote(data, vote_type, model_name):
+    # This is a placeholder for the voting functionality
+    # You can expand this to save votes to a file or database
+    print(f"Received {vote_type} for response: {data}")
+    return data
+
 # Create the Gradio interface
 def create_chatbot_interface():
-    with gr.Blocks() as demo:
-        gr.Markdown("# தமிழ் உரையாடல் பொத்தான் (Tamil Chatbot)")
+    with gr.Blocks(css="css/index.css") as demo:
+        title = "# தமிழ் உரையாடல் பொத்தான் (Tamil Chatbot)"
+        description = "Tamil LLaMA 7B Instruct model with user-controlled generation parameters."
+        
+        gr.Markdown(title)
+        gr.Markdown(description)
         
         # Model loading indicator
         with gr.Row():
             model_status = gr.Markdown("⚠️ மாதிரி ஏற்றப்படவில்லை (Model not loaded)")
             load_model_btn = gr.Button("மாதிரியை ஏற்று (Load Model)")
         
+        # Define chatbot component to use with ChatInterface
         chatbot = gr.Chatbot(label="உரையாடல் (Conversation)", type="messages")
-        msg = gr.Textbox(label="உங்கள் செய்தி (Your Message)", placeholder="இங்கே தட்டச்சு செய்யவும்...")
-        clear = gr.Button("அழி (Clear)")
         
         # Model and tokenizer states
         model = gr.State(None)
         tokenizer = gr.State(None)
         
+        # Parameter sliders
+        with gr.Accordion("Generation Parameters", open=False):
+            temperature = gr.Slider(
+                label="temperature", 
+                value=0.2, 
+                minimum=0.0, 
+                maximum=2.0, 
+                step=0.05, 
+                interactive=True, 
+                info="Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic."
+            )
+            
+            top_p = gr.Slider(
+                label="top_p", 
+                value=1.0, 
+                minimum=0.0, 
+                maximum=1.0, 
+                step=0.01, 
+                interactive=True, 
+                info="0.1 means only the tokens comprising the top 10% probability mass are considered. Suggest set to 1 and use temperature. 1 means 100% and will disable it"
+            )
+            
+            top_k = gr.Slider(
+                label="top_k", 
+                value=40, 
+                minimum=0, 
+                maximum=1000, 
+                step=1, 
+                interactive=True, 
+                info="limits candidate tokens to a fixed number after sorting by probability. Setting it higher than the vocabulary size deactivates this limit."
+            )
+        
         # Function to load model on button click
         def load_model_fn():
             m, t = initialize_model()
@@ -129,41 +172,59 @@ def create_chatbot_interface():
                 return "✅ மாதிரி வெற்றிகரமாக ஏற்றப்பட்டது (Model loaded successfully)", m, t
             else:
                 return "❌ மாதிரி ஏற்றுவதில் பிழை (Error loading model)", None, None
-            
+        
         # Function to handle messages
-        def respond(message, chat_history, model_state, tokenizer_state):
+        def respond(message, chat_history, model_state, tokenizer_state, temp, tp, tk):
             # Check if model is loaded
             if model_state is None:
                 bot_message = "மாதிரி ஏற்றப்படவில்லை. முதலில் 'மாதிரியை ஏற்று' பொத்தானைக் கிளிக் செய்யவும்."
             else:
-                # Generate bot response
-                bot_message = generate_response(model_state, tokenizer_state, message, chat_history)
+                # Generate bot response with parameters
+                bot_message = generate_response(
+                    model_state, 
+                    tokenizer_state, 
+                    message, 
+                    chat_history, 
+                    temperature=temp, 
+                    top_p=tp, 
+                    top_k=tk
+                )
             
             # Append messages
             chat_history.append({"role": "user", "content": message})
             chat_history.append({"role": "assistant", "content": bot_message})
             
-            return "", chat_history, model_state, tokenizer_state
+            return "", chat_history
         
-        # Connect the interface components
+        # Chat interface
+        chat_interface = gr.ChatInterface(
+            fn=lambda message, history, m=model, t=tokenizer, temp=temperature, tp=top_p, tk=top_k: 
+                respond(message, history, m.value, t.value, temp, tp, tk)[1],
+            chatbot=chatbot,
+            examples=[
+                ["வணக்கம், நீங்கள் யார்?"],
+                ["நான் பெரிய பணக்காரன் இல்லை, லேட்டஸ்ட் iPhone-இல் நிறைய பணம் செலவழிக்க வேண்டுமா?"],
+                ["பட்டியலை வரிசைப்படுத்த பைதான் செயல்பாட்டை எழுதவும்."],
+                ["சிவப்பும் மஞ்சளும் கலந்தால் என்ன நிறமாக இருக்கும்?"],
+                ["விரைவாக தூங்குவது எப்படி?"]
+            ],
+            additional_inputs=[temperature, top_p, top_k]
+        )
+        
+        # Connect the model loading button
         load_model_btn.click(
             load_model_fn, 
             outputs=[model_status, model, tokenizer]
         )
         
-        msg.submit(
-            respond, 
-            [msg, chatbot, model, tokenizer], 
-            [msg, chatbot, model, tokenizer]
-        )
-        
-        clear.click(lambda: None, None, chatbot, queue=False)
+        # Add like button functionality
+        chatbot.like(vote, None, None)
         
-        return demo
+    return demo
 
 # Create and launch the demo
 demo = create_chatbot_interface()
 
 # Launch the demo
 if __name__ == "__main__":
-    demo.launch()
+    demo.queue(max_size=3).launch()