diff --git a/torch_pruning/pruner/algorithms/metapruner.py b/torch_pruning/pruner/algorithms/metapruner.py
index b4b936b..9e895be 100644
--- a/torch_pruning/pruner/algorithms/metapruner.py
+++ b/torch_pruning/pruner/algorithms/metapruner.py
@@ -419,8 +419,8 @@ def _prune(self) -> typing.Generator:
                     if group is None: continue
                 ch_groups = self._get_channel_groups(group)
                 imp = self.estimate_importance(group) # raw importance score
-                group_size = len(imp) // ch_groups
                 if imp is None: continue 
+                group_size = len(imp) // ch_groups
                 if ch_groups > 1: # layers with dimension grouping, such as GroupConv, GroupNorm, Multi-head attention, etc.
                     # We average importance across groups here. For example:
                     # imp = [1, 2, 3, 4, 5, 6] with ch_groups=2.