diff --git a/src/scvi/external/decipher/_model.py b/src/scvi/external/decipher/_model.py
index ed7c8a5193..56a34bcfa5 100644
--- a/src/scvi/external/decipher/_model.py
+++ b/src/scvi/external/decipher/_model.py
@@ -199,7 +199,7 @@ def compute_imputed_gene_expression(
             mu = self.module.decoder_z_to_x(z_loc)
             mu = F.softmax(mu, dim=-1)
             library_size = x.sum(axis=-1, keepdim=True)
-            imputed_gene_expr = (library_size * mu).detach().cpu().numpy()
+            imputed_gene_expr = (library_size * mu.detach().cpu()).numpy()
             imputed_gene_expression_batches.append(imputed_gene_expr)
         imputed_gene_expression = np.concatenate(imputed_gene_expression_batches, axis=0)
 
@@ -300,7 +300,7 @@ def compute_gene_patterns(
         t_points = trajectory.trajectory_latent
         t_times = trajectory.trajectory_time
 
-        t_points = torch.FloatTensor(t_points)
+        t_points = torch.FloatTensor(t_points).to(self.module.device)
         z_mean, z_scale = self.module.decoder_v_to_z(t_points)
         z_scale = F.softplus(z_scale)
 
@@ -308,11 +308,12 @@ def compute_gene_patterns(
 
         gene_patterns = {}
         gene_patterns["mean"] = (
-            F.softmax(self.module.decoder_z_to_x(z_mean), dim=-1).detach().numpy() * l_scale
+            F.softmax(self.module.decoder_z_to_x(z_mean), dim=-1).detach().cpu().numpy() * l_scale
         )
 
         gene_expression_samples = (
-            F.softmax(self.module.decoder_z_to_x(z_samples), dim=-1).detach().numpy() * l_scale
+            F.softmax(self.module.decoder_z_to_x(z_samples), dim=-1).detach().cpu().numpy()
+            * l_scale
         )
         gene_patterns["q25"] = np.quantile(gene_expression_samples, 0.25, axis=0)
         gene_patterns["q75"] = np.quantile(gene_expression_samples, 0.75, axis=0)