Update app.py

app.py

@@ -4,8 +4,7 @@ import fitz
 import docx
 from huggingface_hub import InferenceClient
 
-# 1. 引擎配置：GPT-OSS-120B (基于 MoE 架构的高阶逻辑重器)
-# 采用 fastest 自动分流模式，确保响应最灵敏
+# 1. 引擎配置：GPT-OSS-120B (通用重型算力)
 model_id = "openai/gpt-oss-120b"
 hf_token = os.getenv("HF_TOKEN")
 
@@ -36,8 +35,8 @@ def extract_text(file_path):
 
 def chat_fn(message, history):
     clean_messages = []
-    # 设定为全能通用 AI 助手身份
-    clean_messages.append({"role": "system", "content": "You are a highly capable AI assistant powered by GPT-OSS-120B. Provide professional, detailed, and accurate responses."})
+    # 设定通用大师身份
+    clean_messages.append({"role": "system", "content": "You are a professional AI assistant. Please output responses in standard Markdown format (including tables, bold text, etc.)."})
     
     for turn in history:
         clean_messages.append({"role": turn["role"], "content": str(turn["content"])})
@@ -50,7 +49,7 @@ def chat_fn(message, history):
         file_path = f.get("path") if isinstance(f, dict) else f
         context = extract_text(file_path)
         if context:
-            prompt_text = f"【参考背景】:\n{context}\n\n【用户指令】: {prompt_text}"
+            prompt_text = f"【参考背景文档】:\n{context}\n\n【用户指令】: {prompt_text}"
         
     clean_messages.append({"role": "user", "content": prompt_text})
 
@@ -61,21 +60,33 @@ def chat_fn(message, history):
             clean_messages, 
             max_tokens=4096, 
             stream=True, 
-            temperature=0.5
+            temperature=0.4
         )
         
         for msg in stream:
             if hasattr(msg, 'choices') and len(msg.choices) > 0:
                 delta_content = msg.choices[0].delta.content
                 if delta_content:
-                    response += str(delta_content)
+                    # --- 核心修复：针对性清洗 JSON 列表格式数据 ---
+                    # 120B 经常返回 [{'text': '...'}]，我们需要把里面的文字抠出来
+                    content = ""
+                    if isinstance(delta_content, list):
+                        for item in delta_content:
+                            if isinstance(item, dict) and 'text' in item:
+                                content += item['text']
+                    elif isinstance(delta_content, dict):
+                        content = delta_content.get("text", "")
+                    else:
+                        content = str(delta_content)
+                    
+                    response += content
                     yield response
         
         if response:
             yield response + "\n\n---\n> **OpenIPOS MegaNode | GPT-OSS-120B 通用算力出口**"
             
     except Exception as e:
-        yield f"⚠️ 算力调度提示: {str(e)}。由于 120B 模型负载较大，请刷新重试。"
+        yield f"⚠️ 算力调度提示: {str(e)}。由于 120B 模型算力强劲但协议特殊，请尝试重新发送指令。"
 
 # 3. 统一品牌装修
 with gr.Blocks(fill_height=True) as demo: