chat

app.py

@@ -116,12 +116,14 @@ def build_llm_prompt(system: str, context: list[str], user_question: str) -> str
     conversation.append({"role": "user", "content": user_question.strip()})
 
     # 套用 LLaMA-style prompt 格式
-    return tokenizer.apply_chat_template(
+    input_token = tokenizer.apply_chat_template(
         conversation,
         tokenize=False,
         add_generation_prompt=False
     )
 
+    return tokenizer.decode(input_token)
+
 # ---------- 4. Gradio playground (same UI as before) --------------------------
 def store_doc(doc_text: str,user_id="demo",chunk_size=DEFAULT_CHUNK_SIZE,chunk_overlap=DEFAULT_CHUNK_OVERLAP):
     try: