diff --git a/nodes.py b/nodes.py
index 062aafe..094d8e9 100644
--- a/nodes.py
+++ b/nodes.py
@@ -10,7 +10,7 @@ import comfy.model_management
 import comfy.model_patcher
 import folder_paths
 
-from .ops import GGMLTensor, GGMLOps, move_patch_to_cuda
+from .ops import GGMLTensor, GGMLOps, move_patch_to_device
 from .dequant import dequantize_tensor
 
 # Add a custom keys for files ending in .gguf
@@ -108,8 +108,7 @@ class GGUFModelPatcher(comfy.model_patcher.ModelPatcher):
         qtype = getattr(weight, "tensor_type", None)
         if qtype not in (None, gguf.GGMLQuantizationType.F32, gguf.GGMLQuantizationType.F16):
             out_weight = weight.to(device_to)
-            if self.patch_on_device:
-                patches = move_patch_to_cuda(patches, self.load_device)
+            patches = move_patch_to_device(patches, self.load_device if self.patch_on_device else self.offload_device)
             out_weight.patches.append((calculate_weight, patches, key))
         else:
             inplace_update = self.weight_inplace_update or inplace_update
diff --git a/ops.py b/ops.py
index bc17c3a..82a3e8e 100644
--- a/ops.py
+++ b/ops.py
@@ -91,7 +91,7 @@ class GGMLLayer(torch.nn.Module):
         patch_list = []
         device = tensor.device
         for function, patches, key in getattr(tensor, "patches", []):
-            patch_list += move_patch_to_cuda(patches, device)
+            patch_list += move_patch_to_device(patches, device)
 
         # dequantize tensor while patches load
         weight = dequantize_tensor(tensor, dtype, self.dequant_dtype)
@@ -149,12 +149,12 @@ class GGMLOps(comfy.ops.manual_cast):
                 input, weight, self.padding_idx, self.max_norm, self.norm_type, self.scale_grad_by_freq, self.sparse
             ).to(dtype=output_dtype)
 
-def move_patch_to_cuda(item, device):
+def move_patch_to_device(item, device):
     if isinstance(item, torch.Tensor):
         return item.to(device, non_blocking=True)
     elif isinstance(item, tuple):
-        return tuple(move_patch_to_cuda(x, device) for x in item)
+        return tuple(move_patch_to_device(x, device) for x in item)
     elif isinstance(item, list):
-        return [move_patch_to_cuda(x, device) for x in item]
+        return [move_patch_to_device(x, device) for x in item]
     else:
         return item