Update modeling_llava_qwen2.py

modeling_llava_qwen2.py

@@ -535,13 +535,13 @@ class SigLipVisionTower(nn.Module):
         if type(images) is list:
             image_features = []
             for image in images:
-                image_forward_out = self.vision_tower(image.unsqueeze(0),
+                image_forward_out = self.vision_tower(image.to(device=self.device, dtype=self.dtype).unsqueeze(0),
                                                       output_hidden_states=True)
                 image_feature = image_forward_out.hidden_states[-1].to(image.dtype)
                 assert image_features.shape[-2] == 729
                 image_features.append(image_feature)
         else:
-            image_forward_outs = self.vision_tower(images,
+            image_forward_outs = self.vision_tower(images.to(device=self.device, dtype=self.dtype),
                                                    output_hidden_states=True)
             image_features = image_forward_outs.hidden_states[-1].to(images.dtype)
             assert image_features.shape[-2] == 729
@@ -682,9 +682,9 @@ class LlavaMetaForCausalLM(ABC):
             image_features = self.encode_images(concat_images)
             split_sizes = [image.shape[0] for image in images]
             image_features = torch.split(image_features, split_sizes, dim=0)
-            image_features = [x.flatten(0, 1) for x in image_features]
+            image_features = [x.flatten(0, 1).to(self.device) for x in image_features]
         else:
-            image_features = self.encode_images(images)
+            image_features = self.encode_images(images).to(self.device)
 
         # Let's just add dummy tensors if they do not exist,
         # it is a headache to deal with None all the time.