From 061a0f2dcadd5b3275e81a006c596672c49b45fb Mon Sep 17 00:00:00 2001
From: Kakaru <97896816+KakaruHayate@users.noreply.github.com>
Date: Sat, 15 Nov 2025 14:32:10 +0800
Subject: [PATCH] Keep AtanGLU behavior unchanged during eval

---
 modules/commons/common_layers.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/modules/commons/common_layers.py b/modules/commons/common_layers.py
index ae27427c..0012b99c 100644
--- a/modules/commons/common_layers.py
+++ b/modules/commons/common_layers.py
@@ -154,7 +154,10 @@ def forward(self, x):
         # out, gate = x.chunk(2, dim=self.dim)
         # Using torch.split instead of chunk for ONNX export compatibility.        
         out, gate = torch.split(x, x.size(self.dim) // 2, dim=self.dim)
-        return ATanGLUFunction.apply(out, gate)
+        if self.training:
+            return ATanGLUFunction.apply(out, gate)
+        else:
+            return out * torch.atan(gate)
         
         
 class KaimingNormalConv1d(torch.nn.Conv1d):