diff --git a/paddle/fluid/operators/fused/fused_gemm_epilogue_op.cu b/paddle/fluid/operators/fused/fused_gemm_epilogue_op.cu
index a6a136391de2514cadaf7ce9b06788e97929d3b9..6e9bd009a7de4c25ab16fff956f41523779d7cfb 100644
--- a/paddle/fluid/operators/fused/fused_gemm_epilogue_op.cu
+++ b/paddle/fluid/operators/fused/fused_gemm_epilogue_op.cu
@@ -130,5 +130,6 @@ REGISTER_OP_CUDA_KERNEL(
     ops::FusedGemmEpilogueGradKernel<phi::GPUContext, double>,
     ops::FusedGemmEpilogueGradKernel<phi::GPUContext,
                                      paddle::platform::float16>,
-    ops::FusedGemmEpilogueKernel<phi::GPUContext, paddle::platform::bfloat16>);
+    ops::FusedGemmEpilogueGradKernel<phi::GPUContext,
+                                     paddle::platform::bfloat16>);
 #endif