修改GradientClipCallback中parameter的存储方式，防止在torch 1.5版本中报错

fastNLP/core/callback.py

@@ -464,7 +464,10 @@ class GradientClipCallback(Callback):
             self.clip_fun = nn.utils.clip_grad_value_
         else:
             raise ValueError("Only supports `norm` or `value` right now.")
-        self.parameters = list(parameters)
+        if parameters is not None:
+            self.parameters = list(parameters)
+        else:
+            self.parameters = None
         self.clip_value = clip_value
     
     def on_backward_end(self):