Update app.py

app.py

@@ -2,7 +2,7 @@ import gradio as gr
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
 
-# 1. Download the Model
+# 1. Download Your Mistral Model (Corrected ID)
 print("⬇️ Downloading model...")
 model_path = hf_hub_download(
     repo_id="XY26/dual-frame-mistral-7",
@@ -13,8 +13,8 @@ model_path = hf_hub_download(
 print("⚙️ Loading engine...")
 llm = Llama(
     model_path=model_path,
-    n_ctx=4096,        # Context window
-    n_threads=2,       # CPU threads
+    n_ctx=4096,
+    n_threads=2,
     verbose=False
 )
 
@@ -22,26 +22,24 @@ def smart_response(message, history):
     messages = []
     
     # --- SMART ROUTER LOGIC ---
-    # We check if the user is asking for an opinion/debate vs a simple fact
+    # Check if user wants an Opinion (Dual Frame) or Fact (Standard)
     triggers = ["should", "opinion", "think", "good", "bad", "pros", "cons", "benefit", "risk", "impact"]
     is_opinion = any(t in message.lower() for t in triggers)
 
     if is_opinion:
         print(f"🧠 Detected OPINION: {message}")
-        # Force the Dual Framing prompt
+        # Inject the Fine-Tuned Prompt Format
         system_content = """Below is an instruction that describes a task. Write a response that appropriately completes the request.
 ### Instruction:
 {}
 ### Response:
 """
-        # For this model format, we inject the instruction into the system prompt structure or user message
-        # We will format it as a single user message for simplicity with this model
         final_prompt = system_content.format(message)
         messages.append({"role": "user", "content": final_prompt})
         
     else:
         print(f"ℹ️ Detected FACT: {message}")
-        # Standard helpful assistant behavior
+        # Standard Prompt
         messages.append({"role": "system", "content": "You are a helpful assistant. Answer concisely."})
         messages.append({"role": "user", "content": message})
 
@@ -50,7 +48,9 @@ def smart_response(message, history):
         messages=messages,
         max_tokens=1024,
         stream=True,
-        temperature=0.7
+        temperature=0.7,
+        # CRITICAL FIX: Stop tokens prevent the AI from talking to itself
+        stop=["</s>", "User:", "###", "Instruction:", "<|endoftext|>"]
     )
     
     partial_message = ""
@@ -64,8 +64,7 @@ def smart_response(message, history):
 demo = gr.ChatInterface(
     fn=smart_response,
     title="Dual-Mind Mistral 🧠",
-    description="Ask a factual question (e.g., 'Capital of France') for a normal answer.\nAsk a complex question (e.g., 'Should I use AI?') to get Dual Perspectives.",
-    examples=["What is the capital of France?", "Should schools ban smartphones?", "Is nuclear energy safe?", "1 + 1 = ?"],
+    description="Ask a factual question (e.g., 'Capital of France') for a normal answer.\nAsk a complex question (e.g., 'Should I use AI?') to get Dual Perspectives."
 )
 
 if __name__ == "__main__":