diff --git a/bnpm/ca2p_preprocessing.py b/bnpm/ca2p_preprocessing.py
index b5b18af..9f51b08 100644
--- a/bnpm/ca2p_preprocessing.py
+++ b/bnpm/ca2p_preprocessing.py
@@ -442,7 +442,6 @@ def trace_quality_metrics(
         v1=F.T,
         v2=Fneu.T,
         center=True,
-        device=device,
     )
 
     # F_baseline = torch.quantile(F, percentile_baseline/100, dim=1, keepdim=True)
diff --git a/bnpm/similarity.py b/bnpm/similarity.py
index c97ceb7..b30ff41 100644
--- a/bnpm/similarity.py
+++ b/bnpm/similarity.py
@@ -381,7 +381,7 @@ def pairwise_orthogonalization_torch_helper(v1, v2, center:bool=True):
     EVR_total_weighted = torch.nansum(v1_var * EVR) / torch.sum(v1_var)
     EVR_total_unweighted = torch.nanmean(EVR)
     return v1_orth, EVR, EVR_total_weighted, EVR_total_unweighted
-def pairwise_orthogonalization_torch(v1, v2, center:bool=True, device='cpu'):
+def pairwise_orthogonalization_torch(v1, v2, center:bool=True):
     """
     Orthogonalizes columns of v2 off of the columns of v1
      and returns the orthogonalized v1 and the explained
@@ -425,13 +425,14 @@ def pairwise_orthogonalization_torch(v1, v2, center:bool=True, device='cpu'):
     """
     if isinstance(v1, np.ndarray):
         v1 = torch.from_numpy(v1)
+        if isinstance(v2, np.ndarray):
+            v2 = torch.from_numpy(v2)
+        else:
+            raise ValueError("v2 must be a numpy array if v1 is a numpy array")
         return_numpy = True
-    else:
+    elif isinstance(v1, torch.Tensor):
         return_numpy = False
-    if isinstance(v2, np.ndarray):
-        v2 = torch.from_numpy(v2)
-    v1 = v1.to(device)
-    v2 = v2.to(device)
+        
     v1_orth, EVR, EVR_total_weighted, EVR_total_unweighted = pairwise_orthogonalization_torch_helper(v1, v2, center=center)
     if return_numpy:
         v1_orth = v1_orth.cpu().numpy()