Fix copies between DETR and DETA (#29037)

src/transformers/models/deta/modeling_deta.py

@@ -627,7 +627,8 @@ class DetaMultiscaleDeformableAttention(nn.Module):
 
     def _reset_parameters(self):
         nn.init.constant_(self.sampling_offsets.weight.data, 0.0)
-        thetas = torch.arange(self.n_heads, dtype=torch.int64).float() * (2.0 * math.pi / self.n_heads)
+        default_dtype = torch.get_default_dtype()
+        thetas = torch.arange(self.n_heads, dtype=torch.int64).to(default_dtype) * (2.0 * math.pi / self.n_heads)
         grid_init = torch.stack([thetas.cos(), thetas.sin()], -1)
         grid_init = (
             (grid_init / grid_init.abs().max(-1, keepdim=True)[0])