initialize loc-spec params from gaussian

src/ml_downscaling_emulator/score_sde_pytorch_hja22/models/location_params.py

@@ -5,7 +5,7 @@ class LocationParams(torch.nn.Module):
     def __init__(self, n_channels, size) -> None:
         super().__init__()
 
-        self.params = torch.nn.Parameter(torch.zeros(n_channels, size, size))
+        self.params = torch.nn.Parameter(torch.randn(n_channels, size, size))
 
     def forward(self, cond):
         batch_size = cond.shape[0]