Rename MultiTensorAdam To FusedAdam (#50449)

* rename multi_tensor_adam to fused_adam

* fix some bugs

* fix CI coverage

* rename test_fused_adam.py

* fix some bug

* add test_fused_adam_op.py

* fix some bugs

* fix fused_adam_op.cc

* fix CI bugs

* fix CI bug

* fix CI bug

paddle/fluid/operators/optimizers/multi_tensor_adam_op.cc → paddle/fluid/operators/fused/fused_adam_op.cc

@@ -22,19 +22,34 @@ namespace operators {
 
 using Tensor = phi::DenseTensor;
 
-class MultiTensorAdamOp : public framework::OperatorWithKernel {
+class FusedAdamOp : public framework::OperatorWithKernel {
  public:
   using framework::OperatorWithKernel::OperatorWithKernel;
 
   phi::KernelKey GetExpectedKernelType(
-      const framework::ExecutionContext& ctx) const override {
+      const framework::ExecutionContext &ctx) const override {
     auto param_dtype =
         framework::OperatorWithKernel::IndicateVarDataType(ctx, "Params");
     return phi::KernelKey(param_dtype, ctx.GetPlace());
   }
+
+  phi::KernelKey GetKernelTypeForVar(
+      const std::string &var_name,
+      const phi::DenseTensor &tensor,
+      const phi::KernelKey &expected_kernel_type) const override {
+    if (var_name == "Beta1Pows" || var_name == "Beta2Pows" ||
+        var_name == "SkipUpdate") {
+      return phi::KernelKey(phi::Backend::ALL_BACKEND,
+                            expected_kernel_type.layout(),
+                            expected_kernel_type.dtype());
+    } else {
+      return phi::KernelKey(
+          tensor.place(), tensor.layout(), expected_kernel_type.dtype());
+    }
+  }
 };
 
-class MultiTensorAdamOpMaker : public framework::OpProtoAndCheckerMaker {
+class FusedAdamOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
   void Make() override {
     AddInput("Params", "(Tensor) Input parameters").AsDuplicable();
@@ -144,13 +159,13 @@ $$
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-DECLARE_INFER_SHAPE_FUNCTOR(multi_tensor_adam,
-                            MultiTensorAdamInferShapeFunctor,
-                            PD_INFER_META(phi::MultiTensorAdamInferMeta));
+DECLARE_INFER_SHAPE_FUNCTOR(fused_adam,
+                            FusedAdamInferShapeFunctor,
+                            PD_INFER_META(phi::FusedAdamInferMeta));
 REGISTER_OPERATOR(
-    multi_tensor_adam,
-    ops::MultiTensorAdamOp,
-    ops::MultiTensorAdamOpMaker,
+    fused_adam,
+    ops::FusedAdamOp,
+    ops::FusedAdamOpMaker,
     paddle::framework::EmptyGradOpMaker<paddle::framework::OpDesc>,
     paddle::framework::EmptyGradOpMaker<paddle::imperative::OpBase>,
-    MultiTensorAdamInferShapeFunctor);
+    FusedAdamInferShapeFunctor);

paddle/fluid/pybind/eager_generator.h

@@ -166,7 +166,7 @@ std::map<std::string, std::set<std::string>> op_ins_map = {
       "Beta1Pow",
       "Beta2Pow",
       "MasterParam"}},
-    {"multi_tensor_adam",
+    {"fused_adam",
      {"Params",
       "Grads",
       "LearningRate",
@@ -332,7 +332,7 @@ std::map<std::string, std::set<std::string>> op_outs_map = {
       "Beta1PowOut",
       "Beta2PowOut",
       "MasterParamOut"}},
-    {"multi_tensor_adam",
+    {"fused_adam",
      {"ParamsOut",
       "Moments1Out",
       "Moments2Out",
@@ -400,7 +400,7 @@ std::map<std::string, std::set<std::string>> op_passing_outs_map = {
       "Beta1PowOut",
       "Beta2PowOut",
       "MasterParamOut"}},
-    {"multi_tensor_adam",
+    {"fused_adam",
      {"ParamsOut",
       "Moments1Out",
       "Moments2Out",

paddle/phi/api/yaml/legacy_ops.yaml

@@ -781,6 +781,17 @@
   data_transform :
     skip_transform : x
 
+- op : fused_adam_
+  args : (Tensor[] params, Tensor[] grads, Tensor learning_rate, Tensor[] moments1, Tensor[] moments2, Tensor[] beta1_pows, Tensor[] beta2_pows, Tensor[] master_params, Tensor skip_update, Scalar beta1, Scalar beta2, Scalar epsilon, int chunk_size, float weight_decay, bool use_adamw, bool multi_precision, bool use_global_beta_pow)
+  output : Tensor[](params_out){params.size()}, Tensor[](moments1_out){params.size()}, Tensor[](moments2_out){params.size()}, Tensor[](beta1_pows_out){params.size()}, Tensor[](beta2_pows_out){params.size()}, Tensor[](master_params_out){params.size()}
+  infer_meta :
+    func : FusedAdamInferMeta
+  kernel :
+    func : fused_adam
+    data_type : params
+  optional : skip_update, master_params
+  inplace : (params -> params_out), (moments1 -> moments1_out), (moments2 -> moments2_out), (beta1_pows -> beta1_pows_out), (beta2_pows -> beta2_pows_out), (master_params -> master_params_out)
+
 - op : gather
   args : (Tensor x, Tensor index, Scalar(int) axis=0)
   output : Tensor(out)
@@ -1237,17 +1248,6 @@
   optional : master_param
   inplace : (param -> param_out), (velocity -> velocity_out), (master_param -> master_param_out)
 
-- op : multi_tensor_adam_
-  args : (Tensor[] params, Tensor[] grads, Tensor learning_rate, Tensor[] moments1, Tensor[] moments2, Tensor[] beta1_pows, Tensor[] beta2_pows, Tensor[] master_params, Tensor skip_update, Scalar beta1, Scalar beta2, Scalar epsilon, int chunk_size, float weight_decay, bool use_adamw, bool multi_precision, bool use_global_beta_pow)
-  output : Tensor[](params_out){params.size()}, Tensor[](moments1_out){params.size()}, Tensor[](moments2_out){params.size()}, Tensor[](beta1_pows_out){params.size()}, Tensor[](beta2_pows_out){params.size()}, Tensor[](master_params_out){params.size()}
-  infer_meta :
-    func : MultiTensorAdamInferMeta
-  kernel :
-    func : multi_tensor_adam
-    data_type : params
-  optional : skip_update, master_params
-  inplace : (params -> params_out), (moments1 -> moments1_out), (moments2 -> moments2_out), (beta1_pows -> beta1_pows_out), (beta2_pows -> beta2_pows_out), (master_params -> master_params_out)
-
 - op : multiclass_nms3
   args : (Tensor bboxes, Tensor scores, Tensor rois_num, float score_threshold, int nms_top_k, int keep_top_k, float nms_threshold=0.3, bool normalized=true, float nms_eta=1.0, int background_label=0)
   output : Tensor(out), Tensor(index), Tensor(nms_rois_num)

paddle/phi/infermeta/multiary.cc

@@ -2982,7 +2982,7 @@ void YoloLossInferMeta(const MetaTensor& x,
   gt_match_mask->set_dtype(x.dtype());
 }
 
-void MultiTensorAdamInferMeta(
+void FusedAdamInferMeta(
     const std::vector<const MetaTensor*>& params,
     const std::vector<const MetaTensor*>& grads,
     const MetaTensor& learning_rate,

paddle/phi/infermeta/multiary.h

@@ -533,7 +533,7 @@ void YoloLossInferMeta(const MetaTensor& x,
                        MetaTensor* objectness_mask,
                        MetaTensor* gt_match_mask);
 
-void MultiTensorAdamInferMeta(
+void FusedAdamInferMeta(
     const std::vector<const MetaTensor*>& params,
     const std::vector<const MetaTensor*>& grads,
     const MetaTensor& learning_rate,

paddle/phi/kernels/cpu/multi_tensor_adam_kernel.cc → paddle/phi/kernels/cpu/fused_adam_kernel.cc

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/kernels/multi_tensor_adam_kernel.h"
+#include "paddle/phi/kernels/fused_adam_kernel.h"
 #include <vector>
 
 #include "paddle/phi/core/kernel_registry.h"
@@ -29,7 +29,7 @@ static paddle::optional<DenseTensor> TensorPtrToOptionalTensor(
 }
 
 template <typename T, typename Context>
-void MultiTensorAdamKernel(
+void FusedAdamKernel(
     const Context& dev_ctx,
     const std::vector<const DenseTensor*>& params,
     const std::vector<const DenseTensor*>& grads,
@@ -157,9 +157,5 @@ void MultiTensorAdamKernel(
 
 }  // namespace phi
 
-PD_REGISTER_KERNEL(multi_tensor_adam,
-                   CPU,
-                   ALL_LAYOUT,
-                   phi::MultiTensorAdamKernel,
-                   float,
-                   double) {}
+PD_REGISTER_KERNEL(
+    fused_adam, CPU, ALL_LAYOUT, phi::FusedAdamKernel, float, double) {}

paddle/phi/kernels/multi_tensor_adam_kernel.h → paddle/phi/kernels/fused_adam_kernel.h

@@ -20,7 +20,7 @@
 namespace phi {
 
 template <typename T, typename Context>
-void MultiTensorAdamKernel(
+void FusedAdamKernel(
     const Context &dev_ctx,
     const std::vector<const DenseTensor *> &params,
     const std::vector<const DenseTensor *> &grads,

paddle/phi/kernels/gpu/multi_tensor_adam_kernel.cu → paddle/phi/kernels/gpu/fused_adam_kernel.cu

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/kernels/multi_tensor_adam_kernel.h"
+#include "paddle/phi/kernels/fused_adam_kernel.h"
 #include <vector>
 #include "glog/logging.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
@@ -31,9 +31,9 @@ namespace phi {
 // https://github.com/NVIDIA/apex
 
 template <typename T, bool CPUBetaPows /*=true*/>
-struct MultiTensorAdamBetaPowInfo {
+struct FusedAdamBetaPowInfo {
   using MPDType = typename phi::dtype::MPTypeTrait<T>::Type;
-  MultiTensorAdamBetaPowInfo(const MPDType* beta1pow, const MPDType* beta2pow) {
+  FusedAdamBetaPowInfo(const MPDType* beta1pow, const MPDType* beta2pow) {
     beta1pow_ = *beta1pow;
     beta2pow_ = *beta2pow;
   }
@@ -48,9 +48,9 @@ struct MultiTensorAdamBetaPowInfo {
 };
 
 template <typename T>
-struct MultiTensorAdamBetaPowInfo<T, /*CPUBetaPows=*/false> {
+struct FusedAdamBetaPowInfo<T, /*CPUBetaPows=*/false> {
   using MPDType = typename phi::dtype::MPTypeTrait<T>::Type;
-  MultiTensorAdamBetaPowInfo(const MPDType* beta1pow, const MPDType* beta2pow) {
+  FusedAdamBetaPowInfo(const MPDType* beta1pow, const MPDType* beta2pow) {
     beta1pow_ = beta1pow;
     beta2pow_ = beta2pow;
   }
@@ -73,13 +73,13 @@ template <typename T,
           int N,
           int MaxTensorSize,
           int MaxBlockSize>
-struct MultiTensorAdamFunctor {
+struct FusedAdamFunctor {
   __device__ __forceinline__ void operator()(
       int chunk_size,
       const funcs::TensorAndBlockInfo<N, MaxTensorSize, MaxBlockSize>& t_info,
       MT beta1,
       MT beta2,
-      MultiTensorAdamBetaPowInfo<T, IsCPUBetaPow> beta_pow,
+      FusedAdamBetaPowInfo<T, IsCPUBetaPow> beta_pow,
       MT epsilon,
       const MT* learning_rate,
       MT decay) const {
@@ -261,7 +261,7 @@ static int GetVecSizeFromTensors(const std::vector<TensorT*>& tensors,
 }
 
 template <typename T, typename Context>
-void MultiTensorAdamKernel(
+void FusedAdamKernel(
     const Context& dev_ctx,
     const std::vector<const DenseTensor*>& params,
     const std::vector<const DenseTensor*>& grads,
@@ -365,9 +365,9 @@ void MultiTensorAdamKernel(
     constexpr int kMaxTensorSize = __multi_precision ? 48 : 60;              \
     constexpr int kMaxBlockSize = __multi_precision ? 320 : 320;             \
     constexpr int kBlockSize = 512;                                          \
-    MultiTensorAdamBetaPowInfo<T, __is_cpu_betapow> beta_pow_info(           \
+    FusedAdamBetaPowInfo<T, __is_cpu_betapow> beta_pow_info(                 \
         beta1_pow_first->data<MPDType>(), beta2_pow_first->data<MPDType>()); \
-    MultiTensorAdamFunctor<T,                                                \
+    FusedAdamFunctor<T,                                                      \
                      MPDType,                                                \
                      __vec_size,                                             \
                      __multi_precision,                                      \
@@ -487,10 +487,10 @@ void MultiTensorAdamKernel(
 
 }  // namespace phi
 
-PD_REGISTER_KERNEL(multi_tensor_adam,
+PD_REGISTER_KERNEL(fused_adam,
                    GPU,
                    ALL_LAYOUT,
-                   phi::MultiTensorAdamKernel,
+                   phi::FusedAdamKernel,
                    phi::dtype::float16,
                    float,
                    double) {

paddle/phi/ops/compat/multi_tensor_adam_sig.cc → paddle/phi/ops/compat/fused_adam_sig.cc

@@ -18,22 +18,21 @@
 
 namespace phi {
 
-KernelSignature MultiTensorAdamOpArgumentMapping(
-    const ArgumentMappingContext& ctx) {
+KernelSignature FusedAdamOpArgumentMapping(const ArgumentMappingContext& ctx) {
   paddle::small_vector<const char*> in_names = {"Params",
                                                 "Grads",
                                                 "LearningRate",
                                                 "Moments1",
                                                 "Moments2",
-                                                "Beta1Pow",
-                                                "Beta2Pow",
+                                                "Beta1Pows",
+                                                "Beta2Pows",
                                                 "MasterParams",
                                                 "SkipUpdate"};
   paddle::small_vector<const char*> out_names = {"ParamsOut",
                                                  "Moments1Out",
                                                  "Moments2Out",
-                                                 "Beta1PowOut",
-                                                 "Beta2PowOut",
+                                                 "Beta1PowsOut",
+                                                 "Beta2PowsOut",
                                                  "MasterParamsOut"};
   paddle::small_vector<const char*> attr_names = {"beta1",
                                                   "beta2",
@@ -44,7 +43,7 @@ KernelSignature MultiTensorAdamOpArgumentMapping(
                                                   "multi_precision",
                                                   "use_global_beta_pow"};
 
-  return KernelSignature("multi_tensor_adam",
+  return KernelSignature("fused_adam",
                          std::move(in_names),
                          std::move(attr_names),
                          std::move(out_names));
@@ -52,5 +51,4 @@ KernelSignature MultiTensorAdamOpArgumentMapping(
 
 }  // namespace phi
 
-PD_REGISTER_ARG_MAPPING_FN(multi_tensor_adam,
-                           phi::MultiTensorAdamOpArgumentMapping);
+PD_REGISTER_ARG_MAPPING_FN(fused_adam, phi::FusedAdamOpArgumentMapping);

paddle/phi/tests/kernels/CMakeLists.txt

@@ -77,8 +77,8 @@ if(WITH_GPU)
     SRCS test_auto_tune.cu
     DEPS gtest)
   cc_test(
-    test_multi_tensor_adam_kernel
-    SRCS test_multi_tensor_adam_kernel.cc
+    test_fused_adam_kernel
+    SRCS test_fused_adam_kernel.cc
     DEPS gtest phi phi_api_utils)
 elseif(WITH_ROCM)
   hip_test(

paddle/phi/tests/kernels/test_multi_tensor_adam_kernel.cc → paddle/phi/tests/kernels/test_fused_adam_kernel.cc

@@ -32,8 +32,8 @@
 #include "paddle/phi/kernels/cast_kernel.h"
 #include "paddle/phi/kernels/elementwise_subtract_kernel.h"
 #include "paddle/phi/kernels/full_kernel.h"
+#include "paddle/phi/kernels/fused_adam_kernel.h"
 #include "paddle/phi/kernels/gaussian_kernel.h"
-#include "paddle/phi/kernels/multi_tensor_adam_kernel.h"
 #include "paddle/phi/kernels/reduce_max_kernel.h"
 
 namespace phi {
@@ -179,9 +179,9 @@ struct AdamInfo {
         GenerateConstantTensorVectors<MT, Context>(*ctx, one_shapes, beta2);
   }
 
-  void Update(bool use_multi_tensor, const std::vector<DenseTensor> &grads) {
-    if (use_multi_tensor) {
-      UpdateWithMultiTensorAdam(grads);
+  void Update(bool use_fused, const std::vector<DenseTensor> &grads) {
+    if (use_fused) {
+      UpdateWithFusedAdam(grads);
     } else {
       for (size_t j = 0; j < params.size(); ++j) {
         if (use_adamw) {
@@ -226,7 +226,7 @@ struct AdamInfo {
   }
 
  private:
-  void UpdateWithMultiTensorAdam(const std::vector<DenseTensor> &grads) {
+  void UpdateWithFusedAdam(const std::vector<DenseTensor> &grads) {
     auto param_metas = ToMetaTensorVector(params);
     auto grad_metas = ToMetaTensorVector(grads);
     auto master_param_metas = ToMetaTensorVector(master_params);
@@ -235,15 +235,15 @@ struct AdamInfo {
     auto beta1_pow_metas = ToMetaTensorVector(beta1_pows);
     auto beta2_pow_metas = ToMetaTensorVector(beta2_pows);
 
-    MultiTensorAdamInferMeta(
-        ToConstMetaTensorPtrVector(param_metas),
+    FusedAdamInferMeta(ToConstMetaTensorPtrVector(param_metas),
                        ToConstMetaTensorPtrVector(grad_metas),
                        learning_rate,
                        ToConstMetaTensorPtrVector(moment1_metas),
                        ToConstMetaTensorPtrVector(moment2_metas),
                        ToConstMetaTensorPtrVector(beta1_pow_metas),
                        ToConstMetaTensorPtrVector(beta2_pow_metas),
-        multi_precision ? paddle::make_optional(
+                       multi_precision
+                           ? paddle::make_optional(
                                  ToConstMetaTensorPtrVector(master_param_metas))
                            : paddle::none,
                        MetaTensor(),
@@ -262,7 +262,7 @@ struct AdamInfo {
                        ToMutableMetaTensorPtrVector(beta2_pow_metas),
                        ToMutableMetaTensorPtrVector(master_param_metas));
 
-    MultiTensorAdamKernel<T, Context>(
+    FusedAdamKernel<T, Context>(
         *ctx,
         ToConstTensorPtrVector(params),
         ToConstTensorPtrVector(grads),
@@ -395,7 +395,7 @@ auto MaxDiff(const Context &ctx,
 }
 
 template <typename T, typename PlaceType>
-void TestMultiTensorAdamBase(const std::vector<std::vector<int64_t>> &shapes,
+void TestFusedAdamBase(const std::vector<std::vector<int64_t>> &shapes,
                        float atol,
                        bool use_adamw,
                        bool multi_precision = false,
@@ -448,29 +448,28 @@ static auto GenerateRandomShapes(size_t n, uint64_t low, uint64_t high) {
   return shapes;
 }
 
-TEST(multi_tensor_adam, test_fp32_cpu) {
+TEST(fused_adam, test_fp32_cpu) {
   auto shapes = GenerateRandomShapes(30, 10, 20);
   float atol = 0.0f;
   for (auto use_adamw : {false, true}) {
-    TestMultiTensorAdamBase<float, CPUPlace>(shapes, atol, use_adamw);
+    TestFusedAdamBase<float, CPUPlace>(shapes, atol, use_adamw);
   }
 }
 
 #ifdef PADDLE_WITH_CUDA
-TEST(multi_tensor_adam, test_fp32_gpu) {
+TEST(fused_adam, test_fp32_gpu) {
   auto shapes = GenerateRandomShapes(40, 0, 2 << 18);
   float atol = 0.0f;
   for (auto use_adamw : {false, true}) {
-    TestMultiTensorAdamBase<float, GPUPlace>(shapes, atol, use_adamw);
+    TestFusedAdamBase<float, GPUPlace>(shapes, atol, use_adamw);
   }
 }
 
-TEST(multi_tensor_adam, test_fp16_gpu) {
+TEST(fused_adam, test_fp16_gpu) {
   auto shapes = GenerateRandomShapes(40, 0, 2 << 18);
   float atol = 5e-3f;
   for (auto use_adamw : {false, true}) {
-    TestMultiTensorAdamBase<dtype::float16, GPUPlace>(
-        shapes, atol, use_adamw, true);
+    TestFusedAdamBase<dtype::float16, GPUPlace>(shapes, atol, use_adamw, true);
   }
 }
 #endif

python/paddle/fluid/tests/unittests/test_fused_adam_op.py

+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import unittest
+
+import numpy as np
+from op_test import OpTest
+
+import paddle
+
+
+def fused_adam_step(inputs, attributes, num):
+    '''
+    Simulate one step of the fused_adam optimizer
+    :param inputs: dict of inputs
+    :param attributes: dict of attributes
+    :return tuple: tuple of output params, moments1, moments2, beta1_pows, beta2_pows
+    '''
+    params = inputs['Params']
+    grads = inputs['Grads']
+    moments1 = inputs['Moments1']
+    moments2 = inputs['Moments2']
+    lr = inputs['LearningRate']
+    beta1_pows = inputs['Beta1Pows']
+    beta2_pows = inputs['Beta2Pows']
+
+    params_out = []
+    moments1_out = []
+    moments2_out = []
+    beta1_pows_out = []
+    beta2_pows_out = []
+
+    epsilon = attributes['epsilon']
+
+    if 'beta1' in attributes:
+        beta1 = attributes['beta1']
+    else:
+        beta1 = inputs['Beta1Tensor'][0][0]
+    if 'beta2' in attributes:
+        beta2 = attributes['beta2']
+    else:
+        beta2 = inputs['Beta2Tensor'][0][0]
+
+    for i in range(num):
+        moments1_out.append(beta1 * moments1[i][1] + (1 - beta1) * grads[i][1])
+        moments2_out.append(
+            beta2 * moments2[i][1] + (1 - beta2) * np.square(grads[i][1])
+        )
+        lr_t = lr * np.sqrt(1 - beta2_pows[i][1]) / (1 - beta1_pows[i][1])
+        params_out.append(
+            params[i][1]
+            - lr_t * (moments1_out[i] / (np.sqrt(moments2_out[i]) + epsilon))
+        )
+
+    for i in range(num):
+        beta1_pows_out.append(
+            np.array([beta1_pows[i][1]]).astype("float32") * beta1
+        )
+        beta2_pows_out.append(
+            np.array([beta2_pows[i][1]]).astype("float32") * beta2
+        )
+
+    return (
+        params_out,
+        moments1_out,
+        moments2_out,
+        beta1_pows_out,
+        beta2_pows_out,
+    )
+
+
+class TestFusedAdamOp(OpTest):
+    def setUp(self):
+
+        paddle.enable_static()
+
+        '''Test FusedAdam Op with supplied attributes'''
+        self.__class__.op_type = "fused_adam"
+
+        num = 10
+        inputs_list = [[0] * num] * 6
+        learning_rate = 0.004
+        beta1 = 0.78
+        beta2 = 0.836
+        epsilon = 1e-4
+        beta1_pow = beta1**10
+        beta2_pow = beta2**10
+
+        self.attrs = {
+            'epsilon': epsilon,
+            'beta1': beta1,
+            'beta2': beta2,
+            "chunk_size": 32 * 2048,
+        }
+
+        for i in range(num):
+
+            inputs_list[0][i] = np.random.uniform(-1, 1, (102, 105)).astype(
+                "float32"
+            )
+            inputs_list[1][i] = np.random.uniform(-1, 1, (102, 105)).astype(
+                "float32"
+            )
+            inputs_list[2][i] = np.random.uniform(-1, 1, (102, 105)).astype(
+                "float32"
+            )
+            inputs_list[3][i] = np.random.random((102, 105)).astype("float32")
+            inputs_list[4][i] = np.array([beta1_pow]).astype("float32")
+            inputs_list[5][i] = np.array([beta2_pow]).astype("float32")
+
+        self.inputs = {
+            'Params': [
+                ("params" + str(i), inputs_list[0][i]) for i in range(num)
+            ],
+            'Grads': [
+                ("grads" + str(i), inputs_list[1][i]) for i in range(num)
+            ],
+            'Moments1': [
+                ("moments1" + str(i), inputs_list[2][i]) for i in range(num)
+            ],
+            'Moments2': [
+                ("moments2" + str(i), inputs_list[3][i]) for i in range(num)
+            ],
+            'LearningRate': np.array([learning_rate]).astype("float32"),
+            'Beta1Pows': [
+                ("beta1_pows" + str(i), inputs_list[4][i]) for i in range(num)
+            ],
+            'Beta2Pows': [
+                ("beta2_pows" + str(i), inputs_list[5][i]) for i in range(num)
+            ],
+        }
+
+        (
+            params_out,
+            moments1_out,
+            moments2_out,
+            beta1_pows_out,
+            beta2_pows_out,
+        ) = fused_adam_step(self.inputs, self.attrs, num)
+
+        self.outputs = {
+            'Moments1Out': [
+                ("moments1_out" + str(i), moments1_out[i]) for i in range(num)
+            ],
+            'Moments2Out': [
+                ("moments2_out" + str(i), moments2_out[i]) for i in range(num)
+            ],
+            'ParamsOut': [
+                ("params_out" + str(i), params_out[i]) for i in range(num)
+            ],
+            'Beta1PowsOut': [
+                ("beta1_pows_out" + str(i), beta1_pows_out[i])
+                for i in range(num)
+            ],
+            'Beta2PowsOut': [
+                ("beta2_pows_out" + str(i), beta2_pows_out[i])
+                for i in range(num)
+            ],
+        }
+
+    def test_check_output(self):
+        paddle.enable_static()
+        if paddle.is_compiled_with_cuda():
+            self.check_output()
+
+
+if __name__ == "__main__":
+    paddle.enable_static()
+    unittest.main()