Update app.py

app.py

@@ -1,104 +1,90 @@
 import os
 import gradio as gr
 import spaces
-import fitz  # PyMuPDF
-import docx  # python-docx
+import fitz
+import docx
 from huggingface_hub import InferenceClient
 
-# 1. 引擎配置 (MiniMax-M2.5 229B)
-model_id = "MiniMaxAI/MiniMax-M2.5" 
-client = InferenceClient(model_id, token=os.getenv("HF_TOKEN"))
+# 1. 强制显式调用 Token
+hf_token = os.getenv("HF_TOKEN")
+model_id = "MiniMaxAI/MiniMax-M2.5"
+
+# 如果没有 Token，界面会直接提示
+if not hf_token:
+    print("❌ 警告：未检测到 HF_TOKEN，请在 Settings -> Secrets 中添加")
+
+client = InferenceClient(model_id, token=hf_token)
 
-# 2. 全能文件解析解析器
 def extract_text(file_path):
     if not file_path: return ""
     try:
         ext = str(file_path).lower()
-        # PDF 解析
         if ext.endswith(".pdf"):
             doc = fitz.open(file_path)
-            text = "".join([page.get_text() for page in doc])
-            doc.close()
-            return text[:15000] # 适当截取，保证稳定性
-        # Word 解析
+            return "".join([page.get_text() for page in doc])[:15000]
         elif ext.endswith(".docx"):
             doc = docx.Document(file_path)
             return "\n".join([p.text for p in doc.paragraphs])[:15000]
-        # 纯文本解析
         elif ext.endswith(".txt"):
             with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
                 return f.read()[:15000]
-        return f"[暂不支持直接解析此格式: {os.path.basename(file_path)}]"
-    except Exception as e:
-        return f"[解析失败: {str(e)}]"
+        return ""
+    except:
+        return ""
 
 @spaces.GPU(duration=60)
 def chat_fn(message, history):
-    # 强制历史记录清洗 (MiniMax API 对此非常敏感)
+    # 彻底压扁历史记录为标准 API 格式
     clean_messages = []
     for turn in history:
-        # 确保 content 永远是字符串
-        u_content = turn.get("content", "")
-        if not isinstance(u_content, str): u_content = str(u_content)
-        clean_messages.append({"role": turn.get("role", "user"), "content": u_content})
+        clean_messages.append({"role": turn["role"], "content": str(turn["content"])})
     
-    # 提取当前输入
     prompt_text = message.get("text", "")
     files = message.get("files", [])
     
-    # 处理文件逻辑
-    if files and len(files) > 0:
+    if files:
         f = files[0]
         file_path = f.get("path") if isinstance(f, dict) else f
         context = extract_text(file_path)
         if context:
-            prompt_text = f"【参考文档】:\n{context}\n\n【指令】: {prompt_text}"
+            prompt_text = f"【参考文档】:\n{context}\n\n【指令说明】: 请基于以上内容回答：{prompt_text}"
         
     clean_messages.append({"role": "user", "content": prompt_text})
 
     response = ""
     try:
-        # 增加超时和异常保护
-        stream = client.chat_completion(
+        # 增加 headers 确保 Token 穿透
+        for msg in client.chat_completion(
             clean_messages, 
             max_tokens=2048, 
             stream=True, 
-            temperature=0.7
-        )
-        
-        for msg in stream:
-            # 严格检查 API 返回结构
-            if hasattr(msg, 'choices') and len(msg.choices) > 0:
-                delta = msg.choices[0].delta
-                if hasattr(delta, 'content') and delta.content:
-                    response += delta.content
-                    yield response
+            temperature=0.7,
+            headers={"Authorization": f"Bearer {hf_token}"} if hf_token else None
+        ):
+            if msg.choices[0].delta.content:
+                response += msg.choices[0].delta.content
+                yield response
         
         if response:
-            response += "\n\n---\n> **OpenIPOS MegaNode | MiniMax-M2.5 极速逻辑引擎**"
-            yield response
+            yield response + "\n\n---\n> **OpenIPOS MegaNode | MiniMax-M2.5 极速逻辑节点**"
             
     except Exception as e:
-        yield f"⚠️ 调度波动: {str(e)}。提示：由于模型发布初期访问量大，API 偶尔会返回 Error，请稍后重试。"
+        # 如果是权限问题，这里会给出明确提示
+        err_msg = str(e)
+        if "401" in err_msg or "token" in err_msg.lower():
+            yield "❌ 授权失败：请检查 Space Settings 中的 HF_TOKEN 是否正确配置。"
+        else:
+            yield f"⚠️ 调度波动: {err_msg}。请尝试刷新重试。"
 
-# 3. 统一品牌装修
 with gr.Blocks(fill_height=True) as demo:
-    with gr.Row():
-        gr.HTML("""
-            <div style="display: flex; align-items: center; gap: 20px; padding: 10px;">
-                <img src="https://s3.bmp.ovh/2026/03/30/rIAGKg0O.png" style="height: 60px; width: auto;">
-                <div>
-                    <h1 style="margin: 0; font-size: 24px;">OpenIPOS MegaNode</h1>
-                    <p style="margin: 0; color: #666;">中文精品人工智能算力出口 (MiniMax-M2.5)</p>
-                </div>
-            </div>
-        """)
-    
-    gr.Markdown("---")
-    # 开启多模态支持
+    gr.HTML("""
+        <div style="display: flex; align-items: center; gap: 20px; padding: 10px;">
+            <img src="https://s3.bmp.ovh/2026/03/30/rIAGKg0O.png" style="height: 60px;">
+            <div><h1 style="margin: 0; font-size: 24px;">OpenIPOS MegaNode</h1>
+            <p style="margin: 0; color: #666;">中文精品人工智能算力出口 (MiniMax-M2.5)</p></div>
+        </div>
+    """)
     gr.ChatInterface(chat_fn, multimodal=True)
-    
-    gr.Markdown("---")
     gr.Markdown("© 2026 OpenIPOS Global Limited. 算力驱动：Hugging Face Inference Providers")
 
 demo.launch(theme=gr.themes.Soft())