Update modeling_chatglm.py

modeling_chatglm.py

@@ -667,6 +667,9 @@ class GLMTransformer(torch.nn.Module):
                     if len(presents) == 0:
                         presents = kv_cache
                     else:
+                        # Ensure both tensors are on the same device
+                        if presents.device != kv_cache.device:
+                            presents = presents.to(kv_cache.device)
                         presents = torch.cat((presents, kv_cache), dim=0)
 
         if output_hidden_states: