diff --git a/imperative/python/megengine/optimizer/adadelta.py b/imperative/python/megengine/optimizer/adadelta.py
index 60b18593ae0100717d0d2f9b01752f5c4ac67292..1c321d2118d7f632e4969aca96520b34012f982a 100644
--- a/imperative/python/megengine/optimizer/adadelta.py
+++ b/imperative/python/megengine/optimizer/adadelta.py
@@ -84,7 +84,7 @@ class Adadelta(Optimizer):
             step += c1
             grad = param.grad
             if weight_decay != 0.0:
-                grad += param * _weight_decay
+                grad = grad + param * _weight_decay
 
             square_avg = states["square_avg"]
             acc_delta = states["acc_delta"]
diff --git a/imperative/python/megengine/optimizer/adagrad.py b/imperative/python/megengine/optimizer/adagrad.py
index 9b30907727fe0154a3211a2e98a34ab3a3413dde..c983c7916bfde2cb57ef4f4b8c95de4828690b1c 100644
--- a/imperative/python/megengine/optimizer/adagrad.py
+++ b/imperative/python/megengine/optimizer/adagrad.py
@@ -82,7 +82,7 @@ class Adagrad(Optimizer):
             step += c1
             grad = param.grad
             if weight_decay != 0.0:
-                grad += param * _weight_decay
+                grad = grad + param * _weight_decay
 
             square_avg = states["square_avg"]
             square_avg += grad ** c2
diff --git a/imperative/python/megengine/optimizer/adam.py b/imperative/python/megengine/optimizer/adam.py
index 4bd7bea6a25d7a1c971c8d0550262e48fb490202..40d5eec5bf24e28ffb59ed0a4215e71272fd16e1 100644
--- a/imperative/python/megengine/optimizer/adam.py
+++ b/imperative/python/megengine/optimizer/adam.py
@@ -85,7 +85,7 @@ class Adam(Optimizer):
 
             grad = param.grad
             if weight_decay != 0.0:
-                grad += param * _weight_decay
+                grad = grad + param * _weight_decay
 
             states = self._state[param]
 
diff --git a/imperative/python/megengine/optimizer/sgd.py b/imperative/python/megengine/optimizer/sgd.py
index 95e5867c51a06cf01618f730cb480fa9f3dfc784..5ed256d209d6d4890525101c86b3a7cad6076220 100644
--- a/imperative/python/megengine/optimizer/sgd.py
+++ b/imperative/python/megengine/optimizer/sgd.py
@@ -72,7 +72,7 @@ class SGD(Optimizer):
 
             grad = param.grad
             if weight_decay != 0.0:
-                grad += param * _weight_decay
+                grad = grad + param * _weight_decay
 
             if inplace_mode:
                 if momentum: