Update app.py

app.py

@@ -59,30 +59,37 @@ def extract_features_from_seq(sequence_list):
 # --- 4. 核心预测函数 ---
 def predict(sequence_input):
     if model is None:
-        return {"错误": "模型未能加载或初始化失败，请检查后台日志"}
+        # 如果模型加载失败，可以提前抛出错误
+        raise gr.Error("模型未能加载或初始化失败，请检查后台日志。")
     
     if not sequence_input or not isinstance(sequence_input, str):
-        return {"错误": "请输入有效的生物序列"}
+        # 对于无效输入，也直接抛出错误
+        raise gr.Error("请输入有效的生物序列。")
     
     cleaned_sequence = sequence_input.strip().upper()
     sequence_list = [cleaned_sequence]
 
-    try:
-        # !!! 在这里调用了上面的函数 !!!
-        x1_np, x2_np = extract_features_from_seq(sequence_list)
-    except Exception as e:
-        # 如果特征提取失败（包括 NameError），会在这里捕获
-        return {f"特征提取失败": str(e)}
-
+    # !!! 移除这里的 try...except !!!
+    # 让任何可能发生的错误自然地被Gradio捕获
+    x1_np, x2_np = extract_features_from_seq(sequence_list)
+    
+    # 将 NumPy 数组转换为 PyTorch 张量
     tensor_x1 = torch.tensor(x1_np).to(device)
     tensor_x2 = torch.tensor(x2_np).to(device)
 
+    # 模型预测
     with torch.no_grad():
         outputs = model(tensor_x1, tensor_x2)
     
+    # 计算概率
     probabilities = torch.sigmoid(outputs).squeeze().cpu().numpy()
     
+    # 准备输出结果
     labels = ["类别 A (a)", "类别 C (c)", "类别 M (m)", "类别 S (s)"]
+    # 确保即使只有一个序列，结果也能正确处理
+    if probabilities.ndim == 0: # 如果只有一个输出
+        probabilities = [probabilities]
+    
     result = {label: float(prob) for label, prob in zip(labels, probabilities)}
     
     return result