diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py b/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
index e8de57269..dcd227a5b 100644
--- a/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
@@ -1441,7 +1441,7 @@ class FeedforwardModule(nn.Module):
                                                   channel_dim=-1,
                                                   min_positive=0.3,
                                                   max_positive=1.0,
-                                                  min_abs=1.0,
+                                                  min_abs=0.75,
                                                   max_abs=5.0,
                                                   min_prob=0.25)
         self.activation = SwooshL()
@@ -1493,8 +1493,8 @@ class NonlinAttentionModule(nn.Module):
             hidden_channels, channel_dim=-1,
             min_positive=ScheduledFloat((0.0, 0.25), (20000.0, 0.05)),
             max_positive=ScheduledFloat((0.0, 0.75), (20000.0, 0.95)),
-            min_abs=0.75,
-            max_abs=ScheduledFloat((0.0, 2.5), (8000.0, 5.0), default=1.0),
+            min_abs=0.5,
+            max_abs=5.0,
         )
         self.tanh = nn.Tanh()
 
@@ -1637,7 +1637,7 @@ class ConvolutionModule(nn.Module):
             bottleneck_dim, channel_dim=1,
             min_positive=ScheduledFloat((0.0, 0.1), (8000.0, 0.05)),
             max_positive=1.0,
-            min_abs=ScheduledFloat((0.0, 0.2), (20000.0, 1.0)),
+            min_abs=ScheduledFloat((0.0, 0.2), (20000.0, 0.75)),
             max_abs=10.0,
         )