diff --git a/network/Conv2dApproximation.py b/network/Conv2dApproximation.py
index 9276d4f..c37bdb7 100644
--- a/network/Conv2dApproximation.py
+++ b/network/Conv2dApproximation.py
@@ -1,11 +1,14 @@
 import torch
 import math
 
-from network.CPP.PyMultiApp import MultiApp
+from network.PyMultiplicationApproximationCPU import MultiplicationApproximationCPU
+from network.PyMultiplicationApproximationGPU import MultiplicationApproximationGPU
 
 global_multiapp_gpu_setting: list[torch.Tensor] = []
 global_multiapp_size: list[torch.Tensor] = []
-global_multiapp_cpp: list[MultiApp] = []
+global_multiapp_cpp: list[
+    MultiplicationApproximationCPU | MultiplicationApproximationGPU
+] = []
 
 
 class Conv2dApproximation(torch.nn.Module):
@@ -77,7 +80,11 @@ class Conv2dApproximation(torch.nn.Module):
 
         global_multiapp_gpu_setting.append(torch.tensor([0]))
         global_multiapp_size.append(torch.tensor([0, 0, 0, 0]))
-        global_multiapp_cpp.append(MultiApp())
+        if device == torch.device("cpu"):
+            global_multiapp_cpp.append(MultiplicationApproximationCPU())
+        else:
+            global_multiapp_cpp.append(MultiplicationApproximationGPU())
+
         self.multiapp_gpu_setting_position = len(global_multiapp_gpu_setting) - 1
         self.multiapp_cpp_position = len(global_multiapp_cpp) - 1