Updated schema of NumToTensor clean pattern

Author: gs-olive

core/lowering/passes/device_casting.cpp

@@ -41,7 +41,7 @@ void UnpackAndCastNumToTensor(std::shared_ptr<torch::jit::Graph>& graph) {
   // 0D Tensors are initialized on cpu, and need to be casted to CUDA
   // to avoid device mismatch issues
   std::string num_to_tensor_clean_pattern = R"IR(
-    graph(%1: int):
+    graph(%1: Scalar):
       %2: Tensor = prim::NumToTensor(%1)
       %device: Device = prim::Constant[value="cuda"]()
       %dtype: NoneType = prim::Constant()