diff --git a/dsv4/kernels/gemm/fused_swiglu.py b/dsv4/kernels/gemm/fused_swiglu.py
index 4040666c..61fcffeb 100644
--- a/dsv4/kernels/gemm/fused_swiglu.py
+++ b/dsv4/kernels/gemm/fused_swiglu.py
@@ -1285,6 +1285,10 @@ class FusedSwiGLUScaledGroupedGemmKernel:
         # ── Optional: NVFP4 per-expert global scales ──
         global_scale_a: Optional[cute.Tensor],
         global_scale_b: Optional[cute.Tensor],
+        # ── Fused SwiGLU epilogue outputs (replaces out when fused_swiglu=True) ──
+        fp4_out: Optional[cute.Tensor] = None,
+        sf_out: Optional[cute.Tensor] = None,
+        l2_global_scale: Optional[cute.Tensor] = None,
     ):
         """
         GPU device kernel for MoE Scaled Grouped GEMM with block scaling.