Correct the forward call

model.py

@@ -13,7 +13,7 @@ class CSDModel(PreTrainedModel):
 
     @torch.inference_mode()
     def forward(self, pixel_values):
-        features = self.backbone(pixel_values)
+        features = self.backbone(pixel_values, return_dict=False)[1]
         style_embeds = self.out_style(features)
         content_embeds = self.out_content(features)
         return features, style_embeds, content_embeds