Update app.py

app.py

@@ -3,7 +3,7 @@ from llama_cpp import Llama
 
 app = Flask(__name__)
 
-llm = Llama(model_path="./yuna-ai-v3-q3_k_m.gguf")
+llm = Llama(model_path="./yuna-ai-v3-q3_k_m.gguf", verbose=False)
 
 @app.route('/')
 def index():
@@ -13,12 +13,13 @@ def index():
 def generate():
     user_message = request.json['message']
     output = llm(
-        f"Q: {user_message}\nA:",
-        max_tokens=32,
-        stop=["Q:", "\n"],
+        f"Yuki: {user_message}\nYuna:",
+        max_tokens=16,
+        stop=["Yuki:", "Yuna:", "\n"],
         echo=False
     )
-    return jsonify({'response': output})
+    
+    return jsonify({'response': output['choices'][0]['text']})
 
 if __name__ == '__main__':
     app.run(host='0.0.0.0', port=7860)