final

app.py

@@ -90,17 +90,14 @@ def generate(
     max_new_tokens: int=DEFAULT_MAX_NEW_TOKENS,
 ) -> Iterator[str]:
 
-    print(chat_history)
-
     # limit to last 3 turns
     start_idx = max(0, len(chat_history) - 3)
     recent_history = chat_history[start_idx:]
 
     # build list of messages
     messages = []
-    for user_msg, model_msg in recent_history:
-        messages.append({"role": "user", "content": user_msg})
-        messages.append({"role": "model", "content": model_msg})
+    for rh in recent_history:
+        messages.append({"role": rh["role"], "content": rh["content"]})
     messages.append({"role": "user", "content": message})
 
     input_ids = torch.tensor([tokenizer.apply_chat_template(