feat/ Implemented CoT with Blocks statement

app.py

@@ -8,6 +8,7 @@ import torch
 from typing import Optional, Iterator, Dict, Any, List
 from threading import Thread
 from types import NoneType
+import traceback
 
 
 # Initialize logging and device information
@@ -1003,66 +1004,234 @@ knowledge_textbox = gr.Textbox(
     visible=False
 )
 
-chat_interface = gr.ChatInterface(
-        fn=generate,
-        type="messages",
-        multimodal=True,
-        additional_inputs=[
-            knowledge_textbox,
-            gr.Textbox(label="System prompt", lines=6),
-            gr.Slider(
-                label="Max new tokens",
-                minimum=1,
-                maximum=MAX_MAX_NEW_TOKENS,
-                step=1,
-                value=DEFAULT_MAX_NEW_TOKENS,
-            ),
-            gr.Slider(
-                label="Temperature",
-                minimum=0.1,
-                maximum=4.0,
-                step=0.1,
-                value=0.2,
-            ),
-            gr.Slider(
-                label="Top-p (nucleus sampling)",
-                minimum=0.05,
-                maximum=1.0,
-                step=0.05,
-                value=0.8,
-            ),
-            gr.Slider(
-                label="Top-k",
-                minimum=1,
-                maximum=1000,
-                step=1,
-                value=50,
-            ),
-            gr.Slider(
-                label="Repetition penalty",
-                minimum=1.0,
-                maximum=2.0,
-                step=0.05,
-                value=1.0,
-            ),
+with gr.Blocks() as demo:
+    gr.Markdown("# Ontology Generation with Chain-of-Thought")
+    
+    # State components for maintaining conversation
+    chat_state = gr.State([])
+    
+    chatbot = gr.Chatbot(type="messages")
+    message_input  = gr.Textbox(
+        label="message",
+        placeholder="Ask about the elicitation text...",
+        lines=2
+    )
+    
+    with gr.Row():
+        file_explorer = gr.FileExplorer(
+            glob="**/*.txt",
+            file_count="single",
+            label="Upload file",
+            show_label=True
+        )
+        knowledge_input = gr.Textbox(
+            label="Knowledge text",
+            lines=6,
+            visible=True
+        )
+    with gr.Accordion("Advanced Settings", open=False):
+        system_prompt_input = gr.Textbox(
+            label="System Prompt",
+            lines=4,
+            value=DEFAULT_SYSTEM_PROMPT
+        )
+        with gr.Row():
+            with gr.Column():
+                max_tokens_slider = gr.Slider(
+                    label="Max new tokens",
+                    minimum=1,
+                    maximum=14000,
+                    step=1,
+                    value=1024
+                )
+                temperature_slider = gr.Slider(
+                    label="Temperature",
+                    minimum=0.1,
+                    maximum=4.0,
+                    step=0.1,
+                    value=0.2
+                )
+            
+            with gr.Column():
+                top_p_slider = gr.Slider(
+                    label="Top-p (nucleus sampling)",
+                    minimum=0.05,
+                    maximum=1.0,
+                    step=0.05,
+                    value=0.8
+                )
+                top_k_slider = gr.Slider(
+                    label="Top-k",
+                    minimum=1,
+                    maximum=1000,
+                    step=1,
+                    value=50
+                )
+                repetition_penalty_slider = gr.Slider(
+                    label="Repetition penalty",
+                    minimum=1.0,
+                    maximum=2.0,
+                    step=0.05,
+                    value=1.0
+                )
+
+    # Example prompts
+    examples = gr.Examples(
+        examples=[
+            ["List the main classes in the ontology"],
+            ["Summarize the object properties in the ontology"],
+            ["Create a new property to link Person and Environment"]
         ],
-        stop_btn=True,
-        cache_examples=False,
-        show_progress="full",
-        examples=["Donne moi les outils et matériaux utilisés dans ce texte et génère des classes ontologiques sur cette base en format Turtle."]
-        run_examples_on_click=False
+        inputs=message_input
     )
+    
+    def user_message(message:str, history:List[Dict[str, str]]):
+        """Add user message to chat history.
 
-with gr.Blocks() as demo:
-    fe = gr.FileExplorer(
-        glob="**/*.txt",
-        file_count="single",
-        label="Upload an elicitation file",
-        show_label=True
-    )
-    fe.change(append_text_knowledge, fe, knowledge_textbox)
-    chat_interface.render()
+        Args:
+            message (str): The User Message to send
+            history (List[Dict[str,str]]): The previous chat conversation history.
+        """
+        if message.strip() == "":
+            return history, message
+    
+        history = history + [{"role":"user", "content": message}]
+        return history, ""
 
+    def bot_response(history, knowledge, system_prompt, max_tokens, temp, top_p, top_k, rep_penalty):
+        """Generate assistant response with visible thinking.
+
+        Args:
+            history (List[Dict[str, str]]): The previous chat conversation history
+            knowledge (Any): Documents to pass as knowledge to the multimodal model
+            system_prompt (str): System prompt that the model follows
+            max_tokens (int): Max number of allowed output tokens
+            temp (float): Model's Temperature
+            top_p (int): Model's Top p value
+            top_k (int): Model's Top k value
+            rep_penalty (float): Model's repetition penalty
+            
+        Returns:
+            history (List[Dict[str, str]]): The history of the conversation updated
+        """
+        try:
+            if not history or history[-1]["role"] != "user":
+                return history
+            
+            user_message = history[-1]["content"]
+            # thinking message with pending status
+            history.append({
+                "role": "assistant",
+                "content": "Je réfléchis étape par étape...",
+                "metadata": {
+                    "title": "Réflexion",
+                    "status": "pending"
+                }
+            })
+            yield history
+            
+            thinking_conversation = []
+            if system_prompt:
+                thinking_conversation.append({"role": "system", "content": system_prompt})
+            if knowledge:
+                thinking_conversation.append({
+                    "role": "assistant",
+                    "content": f"Voici le document que je dois comprendre: {knowledge}\n\nJe vais l'analyser étape par étape."
+                })
+                
+            for msg in history[:-2]: # All msg except user message and thinking part
+                thinking_conversation.append(msg)
+            
+            thinking_prompt = user_message + "\n\nRéfléchis étape par étape.Identifie d'abord les entités, puis les relations, puis organise hiérarchiquement avant de formaliser."
+            thinking_conversation.append({"role": "user", "content": thinking_prompt})
+            
+            # GENERATE THINKING
+            thinking_result = generate_llm_response(
+                thinking_conversation,
+                max_new_tokens=max_tokens * 2,
+                temperature=temp,
+                top_p=top_p,
+                top_k=top_k,
+                repetition_penalty=rep_penalty
+            )
+            
+            # update the thinking message
+            history[-1] = {
+                "role": "assistant",
+                "content": thinking_result,
+                "metadata": {
+                    "title": "Réflexion",
+                    "status": "done"
+                }
+            }
+            yield history
+        
+            final_conversation = []
+            if system_prompt:
+                    final_conversation.append({"role": "system", "content": system_prompt})
+                
+            if knowledge:
+                final_conversation.append({
+                    "role": "assistant",
+                    "content": f"J'ai analysé ce document: {knowledge}"
+                })
+            
+            for msg in history[:-1]: # exclude thinking
+                if "metadata" not in msg or "title" not in msg.get("metadata", {}):
+                    final_conversation.append(msg)
+
+            final_conversation.append({
+                "role": "assistant",
+                "content": f"Voici mon analyse étape par étape:\n{thinking_result}\n\nMaintenant je vais formaliser le résulat final."
+            })
+            final_answer = generate_llm_response(
+                final_conversation,
+                max_new_tokens=max_tokens,
+                temperature=temp * 0.8,  # Lower temperature for final answer
+                top_p=top_p,
+                top_k=top_k,
+                repetition_penalty=rep_penalty
+            )
+            history.append({
+                "role": "assistant",
+                "content": final_answer
+            })
+            yield history
+        
+        except Exception as e:
+            error_traceback = traceback.format_exc()
+            print(f"Error traceback:\n{error_traceback}")
+            
+            history.append({
+                "role": "assistant",
+                "content": f"An error occurred: {str(e)}\n\nTraceback details:\n{error_traceback}"
+            })
+            yield history
+    
+    file_explorer.change(
+        append_text_knowledge,
+        file_explorer,
+        knowledge_input
+    )
+    
+    message_input.submit(
+        user_message,
+        inputs=[message_input, chatbot],
+        outputs=[chatbot, message_input]
+    ).then(
+        bot_response,
+        inputs=[
+            chatbot,
+            knowledge_input,
+            system_prompt_input,
+            max_tokens_slider,
+            temperature_slider,
+            top_p_slider,
+            top_k_slider,
+            repetition_penalty_slider
+        ],
+        outputs=chatbot
+    )
 if __name__ == "__main__":
     auth = HuggingFaceLogin()
     if auth.login():