Update app.py

app.py

@@ -1,17 +1,17 @@
 import os
 import gradio as gr
-import spaces
 import fitz
 import docx
 from huggingface_hub import InferenceClient
 
-# 1. 引擎配置：Qwen-2.5-72B-Instruct (目前最稳的 70B+ 免费通道)
+# 1. 引擎配置：Qwen-2.5-72B-Instruct (全能稳定版)
 model_id = "Qwen/Qwen2.5-72B-Instruct"
 hf_token = os.getenv("HF_TOKEN")
 
 client = InferenceClient(
     model=model_id, 
     token=hf_token,
+    # 增加超时时间设置，防止翻译长文时断连
     headers={"Authorization": f"Bearer {hf_token}"} if hf_token else None
 )
 
@@ -32,11 +32,10 @@ def extract_text(file_path):
     except:
         return ""
 
-@spaces.GPU(duration=60)
+# 注意：这里移除了 @spaces.GPU，因为 API 模式下它反而会因超时导致报错
 def chat_fn(message, history):
     clean_messages = []
-    # Qwen 专用的系统提示词
-    clean_messages.append({"role": "system", "content": "You are a senior intellectual property expert and professional translator. Provide high-quality analysis."})
+    clean_messages.append({"role": "system", "content": "You are a professional patent and legal translation expert. Provide accurate translations."})
     
     for turn in history:
         clean_messages.append({"role": turn["role"], "content": str(turn["content"])})
@@ -55,12 +54,12 @@ def chat_fn(message, history):
 
     response = ""
     try:
-        # Qwen-72B 专用调用
+        # 增加流式输出
         stream = client.chat_completion(
             clean_messages, 
             max_tokens=4096, 
             stream=True, 
-            temperature=0.6
+            temperature=0.4
         )
         
         for msg in stream:
@@ -71,10 +70,10 @@ def chat_fn(message, history):
                     yield response
         
         if response:
-            yield response + "\n\n---\n> **OpenIPOS MegaNode | Qwen-2.5 全能专业节点**"
+            yield response + "\n\n---\n> **OpenIPOS MegaNode | Qwen-2.5 专业算力出口**"
             
     except Exception as e:
-        yield f"⚠️ 算力调度提示: {str(e)}。由于万亿级模型访问量大，若持续报错请刷新重试。"
+        yield f"⚠️ 算力调度提示: {str(e)}。由于翻译长文档耗时较长，请分段进行或刷新重试。"
 
 # 3. 统一品牌装修
 with gr.Blocks(fill_height=True) as demo:
@@ -88,6 +87,6 @@ with gr.Blocks(fill_height=True) as demo:
         </div>
     """)
     gr.ChatInterface(chat_fn, multimodal=True)
-    gr.Markdown("© 2026 OpenIPOS Global Limited. 算力驱动：Hugging Face Inference Providers")
+    gr.Markdown("© 2026 OpenIPOS Global Limited. 算力驱动：Novita API Cluster")
 
 demo.launch(theme=gr.themes.Soft())