Update utils/model.py

utils/model.py

@@ -427,8 +427,8 @@ class OwlViTForClassification(nn.Module):
 
             device_ = txt_embeds.device
             position_ids = position_ids.to(device_)
-            text_inputs_parts["position_ids"] = position_ids
-            position_embedding = position_embedding.repeat(text_embeds.size(0), 1, 1)
+            position_embedding = position_embedding.repeat(position_ids.size(0), 1, 1)
+            text_inputs_parts["position_ids"] = position_embedding
             print(f"pos + emb: {(txt_embeds + position_embedding).shape}")
             text_embeds_parts = self.owlvit.text_model.get_text_features(**text_inputs_parts)