Update app.py

app.py

@@ -61,10 +61,14 @@ def evaluate(problem, response):
         step_rewards = derive_step_rewards(rewards, reward_flags)
 
         # 确保返回的是有效的JSON字符串
-        return json.dumps(step_rewards[0].tolist())
+        if isinstance(step_rewards[0], torch.Tensor):
+            return json.dumps(step_rewards[0].cpu().numpy().tolist())
+        elif isinstance(step_rewards[0], np.ndarray):
+            return json.dumps(step_rewards[0].tolist())
+        else:
+            return json.dumps(list(step_rewards[0]))  # 转换为列表
     except Exception as e:
         return json.dumps({"error": str(e)})
-
 # 创建Gradio界面
 iface = gr.Interface(
     fn=evaluate,