diff --git a/GeneralRelativity/Interpolation.py b/GeneralRelativity/Interpolation.py
index dc2e6f4..bc8f530 100644
--- a/GeneralRelativity/Interpolation.py
+++ b/GeneralRelativity/Interpolation.py
@@ -151,6 +151,8 @@ def __init__(
         learnable: bool = False,
         align_grids_with_lower_dim_values: bool = False,
         dtype: Type[torch.dtype] = torch.float,
+        device: str = 'cpu',
+
     ):
         """
         Initialize the Interp class.
@@ -161,6 +163,8 @@ def __init__(
         num_channels (int): Number of channels in the input tensor.
         learnable (bool): If True, the interpolation parameters are learnable.
         align_grids_with_lower_dim_values (bool): If True, aligns grid points with lower-dimensional values.
+        dtype (torch.dtype): dtype of weights, defaults to float
+        device (str): defaults to 'cpu'
         """
         self.num_points = num_points
         self.max_degree = max_degree
@@ -214,7 +218,7 @@ def __init__(
             # Create a convolutional kernel with zeros
             kernel = torch.zeros(
                 (num_channels, 1, kernel_size, kernel_size, kernel_size), dtype=dtype
-            )
+            ).to(device)
 
             # Find the minimum index for displacements to adjust kernel indexing
             min_index = torch.min(displacements)
@@ -228,7 +232,7 @@ def __init__(
 
             conv_layer = nn.Conv3d(
                 num_channels, num_channels, kernel_size, groups=num_channels, bias=False
-            )
+            ).to(device)
             conv_layer.weight = nn.Parameter(kernel)
             conv_layer.weight.requires_grad = learnable
             self.conv_layers.append(conv_layer)