Update app.py

app.py

@@ -82,18 +82,11 @@ def build_system_prompt():
     # 把强约束和人设合并为 system prompt
     return f"{SYSTEM_SAFETY.strip()}\n\n=== Persona ===\n{PERSONA.strip()}"
 
-def build_prompt(history, user_msg):
+def build_prompt(history_msgs, user_msg):
     messages = [{"role": "system", "content": build_system_prompt()}]
-
-    # 先注入 few-shot（固定风格与边界）
-    messages.extend(FEW_SHOTS)
-
-    # 再接最近对话，避免超过上下文
-    recent = history[-4:] if len(history) > 4 else history
-    for u, a in recent:
-        if u: messages.append({"role": "user", "content": u})
-        if a: messages.append({"role": "assistant", "content": a})
-
+    messages.extend(FEW_SHOTS)                     # 先注入 few-shot
+    tail = history_msgs[-8:] if len(history_msgs) > 8 else history_msgs
+    messages.extend(tail)                          # 最近几条历史（messages 形式）
     messages.append({"role": "user", "content": user_msg})
     prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
     return prompt
@@ -108,37 +101,39 @@ GEN_KW = dict(
     eos_token_id=tokenizer.eos_token_id
 )
 
-def stream_chat(history, user_msg):
+def stream_chat(history_msgs, user_msg):
     if not user_msg or not user_msg.strip():
-        yield history
+        yield history_msgs
         return
 
-    # 输入侧轻过滤
     if violates(user_msg):
-        reply = SAFE_REPLACEMENT
-        yield history + [[user_msg, reply]]
+        yield history_msgs + [
+            {"role":"user","content": user_msg},
+            {"role":"assistant","content": SAFE_REPLACEMENT},
+        ]
         return
 
-    prompt = build_prompt(history, user_msg)
+    prompt = build_prompt(history_msgs, user_msg)
     inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-
-    streamer = TextIteratorStreamer(
-        tokenizer, skip_prompt=True, skip_special_tokens=True
-    )
+    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
 
     gen_kwargs = dict(**inputs, streamer=streamer, **GEN_KW)
-    thread = Thread(target=model.generate, kwargs=gen_kwargs)
-    thread.start()
+    Thread(target=model.generate, kwargs=gen_kwargs).start()
 
     reply = ""
     for new_text in streamer:
         reply += new_text
-        # 输出侧轻过滤：一旦命中，立即替换为安全话术
         if violates(reply):
             reply = SAFE_REPLACEMENT
-            yield history + [[user_msg, reply]]
+            yield history_msgs + [
+                {"role":"user","content": user_msg},
+                {"role":"assistant","content": reply},
+            ]
             return
-        yield history + [[user_msg, reply]]
+        yield history_msgs + [
+            {"role":"user","content": user_msg},
+            {"role":"assistant","content": reply},
+        ]
 
 # ======== Gradio UI（移动端友好） ========
 CSS = """
@@ -148,7 +143,8 @@ footer{ display:none !important; }
 
 with gr.Blocks(css=CSS, theme=gr.themes.Soft()) as demo:
     gr.Markdown("### 💋 御姐聊天 · Mobile Web\n温柔撩人，但始终优雅有分寸。")
-    chat = gr.Chatbot(height=520, bubble_full_width=False, show_copy_button=True)
+    #chat = gr.Chatbot(height=520, bubble_full_width=False, show_copy_button=True)
+    chat = gr.Chatbot(type="messages", height=520, show_copy_button=True)
     with gr.Row():
         msg = gr.Textbox(placeholder="想跟姐姐聊点什么？（回车发送）", autofocus=True)
         send = gr.Button("发送", variant="primary")
@@ -160,4 +156,4 @@ with gr.Blocks(css=CSS, theme=gr.themes.Soft()) as demo:
     msg.submit(stream_chat, [chat, msg], [chat]); msg.submit(lambda:"", None, msg)
     send.click(stream_chat, [chat, msg], [chat]); send.click(lambda:"", None, msg)
 
-demo.queue().launch()
+demo.queue().launch(share=True)