Update modeling_aimv2.py

modeling_aimv2.py

@@ -105,7 +105,7 @@ class AIMv2ViTPreprocessor(nn.Module):
         tokens = self.patchifier(x)
         pos_embed = get_sincos_pos_embed(
             H // self.patch_h, W // self.patch_w, embed_dim=self.embed_dim
-        )
+        ).to(tokens.device)
         tokens = tokens + pos_embed
         return tokens