Update app.py

app.py

@@ -150,7 +150,18 @@ def generate_chat_response(messages, max_tokens=512, temperature=0.7):
 
         del inputs, outputs
         gc.collect()
+        
+        response = tokenizer.decode(generated_ids, skip_special_tokens=True)
+        response = response.split("<|im_end|>")[0].strip()
+
+        # 强制加前缀，方便判断客户端是否真的收到了内容
+        response = "【OpenClaw 测试回复】\n\n" + response
 
+        if len(response) < 30:
+            response = "（模型输出内容较短） " + response
+
+        logger.info(f"生成内容长度: {len(response)} 字符，前50字: {response[:50]}")
+        
         return {"text": response}
 
     except Exception as e: