Update app.py

app.py

@@ -33,7 +33,8 @@ def hf_chat(system_prompt, user_text, max_tokens=320):
         output = model.generate(
             **inputs,
             max_new_tokens=max_tokens,
-            do_sample=False
+            do_sample=False,
+            use_cache=True
         )
 
     generated = output[0][inputs["input_ids"].shape[1]:]
@@ -44,7 +45,6 @@ def hf_chat(system_prompt, user_text, max_tokens=320):
 
     return text.strip()
 
-
 # ==============================
 # MEZAYA AI
 # ==============================
@@ -130,7 +130,6 @@ SSOT:
 
         return response, ssot
 
-
 agent = MezayaAI()
 
 ALLOWED_ROLES = [
@@ -160,16 +159,14 @@ def respond(message, history, state, role_override):
 
     reply, state = agent.process(message, state, role_override=role_override)
 
-    # Old Gradio Chatbot format: list of tuples
-    history.append((message, reply))
+    history.append({"role": "user", "content": message})
+    history.append({"role": "assistant", "content": reply})
 
     return "", history, state
 
-
 def reset():
     return "", [], agent.new_session()
 
-
 # ==============================
 # UI
 # ==============================