diff --git a/diffsynth/vram_management/layers.py b/diffsynth/vram_management/layers.py
index 6f6f768..3b6e2bb 100644
--- a/diffsynth/vram_management/layers.py
+++ b/diffsynth/vram_management/layers.py
@@ -75,6 +75,9 @@ def enable_vram_management_recursively(model: torch.nn.Module, module_map: dict,
     for name, module in model.named_children():
         for source_module, target_module in module_map.items():
             if isinstance(module, source_module):
+                if "rope_embedder" in name or "patch_embedding" in name:
+                    continue
+
                 num_param = sum(p.numel() for p in module.parameters())
                 if max_num_param is not None and total_num_param + num_param > max_num_param:
                     module_config_ = overflow_module_config
diff --git a/wanvideo/modules/model.py b/wanvideo/modules/model.py
index e56689f..bb82ddd 100644
--- a/wanvideo/modules/model.py
+++ b/wanvideo/modules/model.py
@@ -115,7 +115,7 @@ def rope_apply(x, grid_sizes, freqs):
             freqs[0][:f].view(*f_size).expand(f, h, w, -1),
             freqs[1][:h].view(*h_size).expand(f, h, w, -1),
             freqs[2][:w].view(*w_size).expand(f, h, w, -1)
-        ], dim=-1).reshape(seq_len, 1, -1)
+        ], dim=-1).reshape(seq_len, 1, -1).to(x.device)
 
         @torch.compiler.disable()
         def view_as_real_no_compile(x_i):