change inference_mode to no_grad to avoid errors

image_segmentor.py

@@ -13,7 +13,7 @@ class ImageSegmentor:
         self.image_processor = AutoImageProcessor.from_pretrained("openmmlab/upernet-convnext-small")
         self.image_segmentor = UperNetForSemanticSegmentation.from_pretrained("openmmlab/upernet-convnext-small")
 
-    @torch.inference_mode()
+    @torch.no_grad()
     def __call__(self, image: np.ndarray, **kwargs) -> PIL.Image.Image:
         detect_resolution = kwargs.pop("detect_resolution", 512)
         image_resolution = kwargs.pop("image_resolution", 512)

model.py

@@ -126,7 +126,7 @@ class Model:
             image=control_image,
         ).images
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @spaces.GPU(enable_queue=True)
     def process_canny(
         self,
@@ -171,7 +171,7 @@ class Model:
         ]
         return [control_image] * num_images + results + conditions_of_generated_imgs
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @spaces.GPU(enable_queue=True)
     def process_softedge(
         self,
@@ -238,7 +238,7 @@ class Model:
         ]
         return [control_image] * num_images + results + conditions_of_generated_imgs
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @spaces.GPU(enable_queue=True)
     def process_segmentation(
         self,
@@ -292,7 +292,7 @@ class Model:
         ]
         return [control_image] * num_images + results + conditions_of_generated_imgs
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @spaces.GPU(enable_queue=True)
     def process_depth(
         self,
@@ -345,7 +345,7 @@ class Model:
         ]
         return [control_image] * num_images + results + conditions_of_generated_imgs
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @spaces.GPU(enable_queue=True)
     def process_lineart(
         self,