diff --git a/paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc b/paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc
index e1857d399b6ce5753aff2945655ba93ce6dd5cf7..e4719cbcf0e14fe65d9b46a96d70fd6ecda8ef53 100644
--- a/paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc
+++ b/paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc
@@ -70,7 +70,6 @@ static std::set<std::string> OpsNeedSetOutputDtypeWhenRegisterPhiKernel = {
     "merged_adam",
     "momentum",
     "multiclass_nms3",
-    "multinomial",
     "nanmedian",
     "sync_batch_norm_grad",
     "unique",
diff --git a/paddle/phi/kernels/cpu/multinomial_kernel.cc b/paddle/phi/kernels/cpu/multinomial_kernel.cc
index 1e31b4649fbcc8fd85b6d471679f0094edd51eb2..d1f713db00bd56ad10cb6556eb0ec92cb9743984 100644
--- a/paddle/phi/kernels/cpu/multinomial_kernel.cc
+++ b/paddle/phi/kernels/cpu/multinomial_kernel.cc
@@ -45,4 +45,6 @@ void MultinomialKernel(const Context& dev_ctx,
 }  // namespace phi
 
 PD_REGISTER_KERNEL(
-    multinomial, CPU, ALL_LAYOUT, phi::MultinomialKernel, float, double) {}
+    multinomial, CPU, ALL_LAYOUT, phi::MultinomialKernel, float, double) {
+  kernel->OutputAt(0).SetDataType(phi::DataType::INT64);
+}
diff --git a/paddle/phi/kernels/gpu/multinomial_kernel.cu b/paddle/phi/kernels/gpu/multinomial_kernel.cu
index 2e6e95f4c232ec7281944a1594e8bd302275a8c9..8b1d960527fd7dc784ad2a88202f202c93627142 100644
--- a/paddle/phi/kernels/gpu/multinomial_kernel.cu
+++ b/paddle/phi/kernels/gpu/multinomial_kernel.cu
@@ -296,6 +296,8 @@ PD_REGISTER_KERNEL(multinomial,  // cuda_only
                    ALL_LAYOUT,
                    phi::MultinomialKernel,
                    float,
-                   double) {}
+                   double) {
+  kernel->OutputAt(0).SetDataType(phi::DataType::INT64);
+}
 
 #endif