Update modeling_gpt2vision.py

modeling_gpt2vision.py

@@ -71,7 +71,7 @@ class GPT2Vision(PreTrainedModel):
         # Convert the image to a tensor and add a batch dimension
         image_tensor = self.vision_encoder.image_transform(image).unsqueeze(0).to(self.device)
         with torch.no_grad():
-            img_features = self.vision_encoder(image_tensor).last_hidden_state
+            img_features = self.vision_encoder(image_tensor,device="cuda")
         img_embs = self.mlp(img_features)
         
         # Tokenize the question