.

meta_ui.py

@@ -149,7 +149,7 @@ def chat(
     # threading.Thread(target=worker).start()
     thread = threading.Thread(target=worker, daemon=True) 
     thread.start()      
-    yield "⏳ 正在生成中...", log_state
+    # yield "⏳ 正在生成中...", log_state
 
 
     while True:
@@ -169,24 +169,15 @@ def chat(
         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
+        # yield result, log_state       # 第一次真正把模型回复丢给前端
+        if api_suffix == "/v1/chat/completions":
+            history.append({"role": "assistant", "content": result["text"]})
+            yield result["text"], history
+        else:
+            yield result["text"], log_state
+
         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:
     gr.Markdown(f"## 💬 调试界面  \n权重 **{MODEL_PATH.name}**")