diff --git a/unsloth/kernels/fast_lora.py b/unsloth/kernels/fast_lora.py
index e24fa48dcc..52cbf47488 100644
--- a/unsloth/kernels/fast_lora.py
+++ b/unsloth/kernels/fast_lora.py
@@ -123,16 +123,16 @@ class LoRA_MLP(torch.autograd.Function):
         g  = g .view(-1, g .shape[-1])
         dtype = X.dtype
 
-        DW = matmul_lora(dY, downW.t(), downW_quant, downB, downA, downS)
-        se = torch.nn.functional.sigmoid(e)
-        f = e * se
-        h = f * g
-        df = se * (1 - f) + f
-        DW_f   = DW * f
-        DW_dfg = DW * df * g
         # DW = matmul_lora(dY, downW.t(), downW_quant, downB, downA, downS)
-        # DW, e, g = swiglu_DWf_DW_dfg_kernel(DW, e, g)
-        # h, DW_f, DW_dfg = DW, e, g
+        # se = torch.nn.functional.sigmoid(e)
+        # f = e * se
+        # h = f * g
+        # df = se * (1 - f) + f
+        # DW_f   = DW * f
+        # DW_dfg = DW * df * g
+        DW = matmul_lora(dY, downW.t(), downW_quant, downB, downA, downS)
+        DW, e, g = swiglu_DWf_DW_dfg_kernel(DW, e, g)
+        h, DW_f, DW_dfg = DW, e, g
 
         # Down projection LoRA weights
         d_downA = h.t() @ (dY @ downB.t())