diff --git a/meta_ui.py b/meta_ui.py
index fd4d700..2c7fc65 100644
--- a/meta_ui.py
+++ b/meta_ui.py
@@ -115,23 +115,44 @@ def chat(
         out = backend(user, samp, api_suffix)
         result_q.put(out)
 
-    threading.Thread(target=worker).start()
+    # threading.Thread(target=worker).start()
+    thread = threading.Thread(target=worker, daemon=True) 
     yield "⏳ 正在生成中...", log_state
 
+
     while True:
+        # ⚠️ 线程已结束且队列已空 → 直接 return 让生成器终止
+        if not thread.is_alive() and result_q.empty():
+            return  # ← 新增这一行
+
         try:
             result = result_q.get(timeout=0.1)
-            # ★ 不论 /generate（纯 str）还是 /v1/completions（dict），都转成 {"text": ...}
-            if isinstance(result, str):
-                result = {"text": result}
-            elif not isinstance(result, dict) or "text" not in result:
-                result = {"text": str(result)}
-
-            yield result, log_state
-            break  
         except Empty:
             continue
 
+        # 统一格式
+        if isinstance(result, str):
+            result = {"text": result}
+        elif not isinstance(result, dict) or "text" not in result:
+            result = {"text": str(result)}
+
+        yield result, log_state       # 第一次真正把模型回复丢给前端
+        return                       # ← 把旧的 break 换成 return
+
+    # while True:
+    #     try:
+    #         result = result_q.get(timeout=0.1)
+    #         # ★ 不论 /generate（纯 str）还是 /v1/completions（dict），都转成 {"text": ...}
+    #         if isinstance(result, str):
+    #             result = {"text": result}
+    #         elif not isinstance(result, dict) or "text" not in result:
+    #             result = {"text": str(result)}
+
+    #         yield result, log_state
+    #         break  
+    #     except Empty:
+    #         continue
+
 
 # ────────────────── Gradio UI ──────────────────
 with gr.Blocks(title="调试界面") as demo: