[easy] Fix .sizes() call in saved_variable.cpp for nested tensor (pyt…

…orch#83356)

Small fix so that TestMultipleDispatch in the above PR will throw the correct error when using an inplace operation on a saved nested input

Pull Request resolved: pytorch#83356
Approved by: https://github.com/albanD

torch/csrc/autograd/saved_variable.cpp

@@ -159,7 +159,12 @@ Variable SavedVariable::unpack(std::shared_ptr<Node> saved_for) const {
       message
           << "one of the variables needed for gradient computation has been "
              "modified by an inplace operation: ["
-          << data_.toString() << " " << data_.sizes() << "]";
+          << data_.toString() << " ";
+      if (data_.is_nested()) {
+        message << data_._nested_tensor_size() << "]";
+      } else {
+        message << data_.sizes() << "]";
+      }
       if (grad_fn) {
         message << ", which is output " << output_nr_ << " of "
                 << grad_fn->name() << ",";