Update app.py

app.py

@@ -1,63 +1,72 @@
 import gradio as gr
 from llama_cpp import Llama
+from huggingface_hub import hf_hub_download
 
-# --- 1. SETUP: Load the Model ---
-model_id = "XY26/dual-frame-mistral-7"
-filename = "mistral-7b-v0.3.Q4_K_M.gguf"
+# 1. Download the Model
+print("⬇️ Downloading model...")
+model_path = hf_hub_download(
+    repo_id="XY26/dual-frame-mistral-7",
+    filename="mistral-7b-v0.3.Q4_K_M.gguf"
+)
 
-# Load the model (CPU Optimized)
+# 2. Load the Engine
+print("⚙️ Loading engine...")
 llm = Llama(
-    model_path=gr.utils.download_from_hub(model_id, filename),
-    n_ctx=2048,      # Context window
-    n_threads=2      # CPU threads (Keep low for free tier)
+    model_path=model_path,
+    n_ctx=4096,        # Context window
+    n_threads=2,       # CPU threads
+    verbose=False
 )
 
-# --- 2. THE SMART LOGIC (The Router) ---
-def is_opinion_question(text):
-    """
-    Simple keyword detection to guess if a user wants an opinion or a fact.
-    """
-    triggers = ["should", "opinion", "think", "good", "bad", "pros", "cons", "benefit", "risk", "impact", "effect", "analyze", "perspectives"]
-    text_lower = text.lower()
-    return any(trigger in text_lower for trigger in triggers)
-
-def generate_response(message, history):
+def smart_response(message, history):
+    messages = []
     
-    # DECISION TIME: Fact vs Opinion
-    if is_opinion_question(message):
-        # MODE A: DUAL FRAMING (Inject the special prompt)
-        print(f"Detected OPINION: {message}")
-        system_prompt = """Below is an instruction that describes a task. Write a response that appropriately completes the request.
+    # --- SMART ROUTER LOGIC ---
+    # We check if the user is asking for an opinion/debate vs a simple fact
+    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
+        system_content = """Below is an instruction that describes a task. Write a response that appropriately completes the request.
 ### Instruction:
 {}
 ### Response:
 """
-        prompt = system_prompt.format(message)
+        # 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:
-        # MODE B: FACTUAL (Inject a strict 'normal' prompt)
-        print(f"Detected FACT: {message}")
-        system_prompt = """[INST] You are a helpful assistant. Answer the following question concisely and accurately. Do not use positive/negative framing headers.
-Question: {} [/INST]"""
-        prompt = system_prompt.format(message)
+        print(f"ℹ️ Detected FACT: {message}")
+        # Standard helpful assistant behavior
+        messages.append({"role": "system", "content": "You are a helpful assistant. Answer concisely."})
+        messages.append({"role": "user", "content": message})
 
-    # Generate
-    output = llm(
-        prompt, 
-        max_tokens=512, 
-        stop=["</s>", "[/INST]"], 
-        echo=False
+    # --- GENERATION ---
+    stream = llm.create_chat_completion(
+        messages=messages,
+        max_tokens=1024,
+        stream=True,
+        temperature=0.7
     )
     
-    return output['choices'][0]['text']
+    partial_message = ""
+    for chunk in stream:
+        delta = chunk['choices'][0]['delta']
+        if 'content' in delta:
+            partial_message += delta['content']
+            yield partial_message
 
-# --- 3. THE INTERFACE ---
-interface = gr.ChatInterface(
-    fn=generate_response,
+# 3. Launch Interface
+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 = ?"],
 )
 
 if __name__ == "__main__":
-    interface.launch()
+    demo.queue().launch(server_name="0.0.0.0", server_port=7860)