diff --git a/ranger21/ranger21.py b/ranger21/ranger21.py
index 88968fe..a8ab0a5 100644
--- a/ranger21/ranger21.py
+++ b/ranger21/ranger21.py
@@ -735,13 +735,16 @@ def step(self, closure=None):
                 if self.use_adabelief:
                     grad_ma.mul_(beta1).add_(grad, alpha=1 - beta1)
                     grad_residual = grad - grad_ma
-                    variance_ma_belief.mul_(beta2).addcmul(
+                    variance_ma_belief.mul_(beta2).addcmul_(
                         grad_residual, grad_residual, value=1 - beta2
                     )
                 # print(f"upper loop grad = {grad.shape}")
                 variance_ma.mul_(beta2).addcmul_(grad, grad, value=1 - beta2)
                 # print(f"variance_ma, grad adjusted")
-                variance_ma_debiased = variance_ma / bias_correction2
+                if self.use_adabelief:
+                    variance_ma_debiased = variance_ma_belief / bias_correction2
+                else:
+                    variance_ma_debiased = variance_ma / bias_correction2
 
                 variance_ma_sum += variance_ma_debiased.sum()
                 # print(f"variance_ma_sum = {variance_ma_sum}")
@@ -909,7 +912,8 @@ def step(self, closure=None):
                         # Maintains the maximum of all 2nd moment running avg. till now
                         torch.max(max_variance_ma, variance_ma, out=variance_ma)
                         # Use the max. for normalizing running avg. of gradient
-                        denom = (variance_ma.sqrt() / math.sqrt(bias_correction2)).add_(
+                        variance_source = variance_ma_belief if self.use_adabelief else variance_ma
+                        denom = (variance_source.sqrt() / math.sqrt(bias_correction2)).add_(
                             group["eps"]
                         )