Update Save/Load Interface to 2.0 (#55836)

Update Save/Load Interface to 2.0

paddle/cinn/auto_schedule/tests/performance_comparison_test.cc

@@ -55,6 +55,7 @@ DEFINE_string(resnet50_model_dir,
 DEFINE_int32(evaluate_knobs,
              -1,
              "the options to control which schedule tests will be run.");
+DECLARE_double(cinn_infer_model_version);
 
 namespace cinn {
 namespace auto_schedule {
@@ -353,6 +354,7 @@ TEST_F(PerformanceTester, Gather) {
 // paddle model test
 TEST_F(PerformanceTester, ResNet50) {
   CHECK_NE(FLAGS_resnet50_model_dir, "");
+  FLAGS_cinn_infer_model_version = 1.0;
   std::unordered_map<std::string, std::vector<int64_t>> feeds = {
       {"inputs", {batch_size, 3, 224, 224}}};
   Evaluate(cinn::frontend::PaddleModelConvertor(common::DefaultNVGPUTarget())

paddle/cinn/frontend/interpreter.h

@@ -43,7 +43,7 @@ class Interpreter final {
    */
   void LoadPaddleModel(const std::string& model_dir,
                        const Target& target,
-                       bool params_combined = false,
+                       bool params_combined,
                        const std::string& model_name = "");
 
   /**

paddle/cinn/frontend/interpreter_test.cc

@@ -24,11 +24,11 @@ namespace cinn::frontend {
 
 TEST(Interpreter, basic) {
   Interpreter executor({"A"}, {{1, 30}});
-  executor.LoadPaddleModel(FLAGS_model_dir, common::DefaultTarget());
+  executor.LoadPaddleModel(FLAGS_model_dir, common::DefaultTarget(), true);
   executor.Run();
   // fc_0.tmp_2 is eliminated by OpFusion, so here
   // change to get tenor of the out variable
-  executor.GetTensor("fc_0.tmp_2");
+  executor.GetTensor("save_infer_model/scale_0.tmp_0");
 }
 
 }  // namespace cinn::frontend

paddle/cinn/frontend/paddle/model_parser.cc

@@ -245,12 +245,8 @@ void LoadModelPb(const std::string &model_dir,
   VLOG(3) << "param_file is: " << param_file;
   // Load model
   VLOG(4) << "Start load model program...";
-  std::string prog_path = model_dir + "/__model__";
-  std::string param_file_temp = param_file;
-  if (combined) {
-    // prog_path = model_file;
-    param_file_temp = model_dir + "/params";
-  }
+  std::string prog_path = model_dir + model_file;
+  std::string param_file_temp = model_dir + param_file;
   framework_proto::ProgramDesc pb_proto_prog =
       *LoadProgram(prog_path, model_from_memory);
   pb::ProgramDesc pb_prog(&pb_proto_prog);

paddle/cinn/frontend/paddle/model_parser_test.cc

@@ -24,7 +24,7 @@ namespace cinn::frontend::paddle {
 TEST(LoadModelPb, naive_model) {
   hlir::framework::Scope scope;
   cpp::ProgramDesc program_desc;
-  LoadModelPb(FLAGS_model_dir, "__model__", "", &scope, &program_desc, false);
+  LoadModelPb(FLAGS_model_dir, "/__model__", "", &scope, &program_desc, false);
 
   ASSERT_EQ(program_desc.BlocksSize(), 1UL);
 

paddle/cinn/frontend/paddle_model_convertor.cc

@@ -27,6 +27,8 @@
 #include "paddle/cinn/frontend/var_type_utils.h"
 #include "paddle/cinn/hlir/op/use_ops.h"
 
+DECLARE_double(cinn_infer_model_version);
+
 namespace cinn {
 namespace frontend {
 
@@ -123,14 +125,25 @@ Program PaddleModelConvertor::LoadModel(
     bool is_combined,
     const std::unordered_map<std::string, std::vector<int64_t>>& feed) {
   paddle::cpp::ProgramDesc program_desc;
-  paddle::LoadModelPb(model_dir,
-                      "__model__",
-                      "",
-                      scope_.get(),
-                      &program_desc,
-                      is_combined,
-                      false,
-                      target_);
+  if (FLAGS_cinn_infer_model_version < 2.0) {
+    paddle::LoadModelPb(model_dir,
+                        "/__model__",
+                        "/params",
+                        scope_.get(),
+                        &program_desc,
+                        is_combined,
+                        false,
+                        target_);
+  } else {
+    paddle::LoadModelPb(model_dir,
+                        ".pdmodel",
+                        ".pdiparams",
+                        scope_.get(),
+                        &program_desc,
+                        is_combined,
+                        false,
+                        target_);
+  }
   CHECK_EQ(program_desc.BlocksSize(), 1)
       << "CINN can only support the model with a single block";
   auto* block_desc = program_desc.GetBlock<paddle::cpp::BlockDesc>(0);

paddle/cinn/frontend/paddle_model_to_program.cc

@@ -21,6 +21,8 @@
 #include "paddle/cinn/frontend/paddle/pb/program_desc.h"
 #include "paddle/cinn/hlir/framework/node.h"
 
+DECLARE_double(cinn_infer_model_version);
+
 namespace cinn {
 namespace frontend {
 using utils::Join;
@@ -740,14 +742,25 @@ Variable PaddleModelToProgram::GetVar(const std::string& name) {
 std::unique_ptr<Program> PaddleModelToProgram::operator()(
     const std::string& model_dir, bool is_combined) {
   paddle::cpp::ProgramDesc program_desc;
-  paddle::LoadModelPb(model_dir,
-                      "__model__",
-                      "",
-                      scope_,
-                      &program_desc,
-                      is_combined,
-                      false,
-                      target_);
+  if (FLAGS_cinn_infer_model_version < 2.0) {
+    paddle::LoadModelPb(model_dir,
+                        "/__model__",
+                        "/params",
+                        scope_,
+                        &program_desc,
+                        is_combined,
+                        false,
+                        target_);
+  } else {
+    paddle::LoadModelPb(model_dir,
+                        ".pdmodel",
+                        ".pdiparams",
+                        scope_,
+                        &program_desc,
+                        is_combined,
+                        false,
+                        target_);
+  }
   CHECK_EQ(program_desc.BlocksSize(), 1)
       << "CINN can only support the model with a single block";
   auto* block_desc = program_desc.GetBlock<paddle::cpp::BlockDesc>(0);

paddle/cinn/runtime/flags.cc

@@ -33,6 +33,7 @@ DEFINE_bool(cinn_cudnn_deterministic,
 #endif
 
 using ::GFLAGS_NAMESPACE::BoolFromEnv;
+using ::GFLAGS_NAMESPACE::DoubleFromEnv;
 using ::GFLAGS_NAMESPACE::Int32FromEnv;
 using ::GFLAGS_NAMESPACE::Int64FromEnv;
 using ::GFLAGS_NAMESPACE::StringFromEnv;
@@ -180,6 +181,11 @@ DEFINE_int32(cinn_error_message_level,
              "Specify the level of printing error message in the schedule."
              "0 means short, 1 means detailed.");
 
+DEFINE_double(cinn_infer_model_version,
+              DoubleFromEnv("FLAGS_cinn_infer_model_version", 2.0),
+              "Paddle has different model format in inference model. We use "
+              "a flag to load different versions.");
+
 namespace cinn {
 namespace runtime {
 

test/cinn/CMakeLists.txt

@@ -91,7 +91,8 @@ if(WITH_CUDNN)
     COMMAND
       ${CMAKE_COMMAND} -E env
       PYTHONPATH=${CMAKE_BINARY_DIR}:${CMAKE_BINARY_DIR}/python/cinn:$ENV{PYTHONPATH}
-      python3 ${CMAKE_CURRENT_SOURCE_DIR}/test_resnet18.py
+      FLAGS_cinn_infer_model_version=1.0 python3
+      ${CMAKE_CURRENT_SOURCE_DIR}/test_resnet18.py
       "${CMAKE_BINARY_DIR}/third_party/ResNet18" "${WITH_GPU}"
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 
@@ -100,7 +101,8 @@ if(WITH_CUDNN)
     COMMAND
       ${CMAKE_COMMAND} -E env
       PYTHONPATH=${CMAKE_BINARY_DIR}:${CMAKE_BINARY_DIR}/python/cinn:$ENV{PYTHONPATH}
-      python3 ${CMAKE_CURRENT_SOURCE_DIR}/test_mobilenetv2.py
+      FLAGS_cinn_infer_model_version=1.0 python3
+      ${CMAKE_CURRENT_SOURCE_DIR}/test_mobilenetv2.py
       "${CMAKE_BINARY_DIR}/third_party/MobileNetV2" "${WITH_GPU}"
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 
@@ -109,7 +111,8 @@ if(WITH_CUDNN)
     COMMAND
       ${CMAKE_COMMAND} -E env
       PYTHONPATH=${CMAKE_BINARY_DIR}:${CMAKE_BINARY_DIR}/python/cinn:$ENV{PYTHONPATH}
-      python3 ${CMAKE_CURRENT_SOURCE_DIR}/test_efficientnet.py
+      FLAGS_cinn_infer_model_version=1.0 python3
+      ${CMAKE_CURRENT_SOURCE_DIR}/test_efficientnet.py
       "${CMAKE_BINARY_DIR}/third_party/EfficientNet" "${WITH_GPU}"
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 
@@ -118,7 +121,8 @@ if(WITH_CUDNN)
     COMMAND
       ${CMAKE_COMMAND} -E env
       PYTHONPATH=${CMAKE_BINARY_DIR}:${CMAKE_BINARY_DIR}/python/cinn:$ENV{PYTHONPATH}
-      python3 ${CMAKE_CURRENT_SOURCE_DIR}/test_mobilenetv1.py
+      FLAGS_cinn_infer_model_version=1.0 python3
+      ${CMAKE_CURRENT_SOURCE_DIR}/test_mobilenetv1.py
       "${CMAKE_BINARY_DIR}/third_party/MobilenetV1" "${WITH_GPU}"
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 
@@ -127,7 +131,8 @@ if(WITH_CUDNN)
     COMMAND
       ${CMAKE_COMMAND} -E env
       PYTHONPATH=${CMAKE_BINARY_DIR}:${CMAKE_BINARY_DIR}/python/cinn:$ENV{PYTHONPATH}
-      python3 ${CMAKE_CURRENT_SOURCE_DIR}/test_resnet50.py
+      FLAGS_cinn_infer_model_version=1.0 python3
+      ${CMAKE_CURRENT_SOURCE_DIR}/test_resnet50.py
       "${CMAKE_BINARY_DIR}/third_party/ResNet50" "${WITH_GPU}"
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 
@@ -136,7 +141,8 @@ if(WITH_CUDNN)
     COMMAND
       ${CMAKE_COMMAND} -E env
       PYTHONPATH=${CMAKE_BINARY_DIR}:${CMAKE_BINARY_DIR}/python/cinn:$ENV{PYTHONPATH}
-      python3 ${CMAKE_CURRENT_SOURCE_DIR}/test_squeezenet.py
+      FLAGS_cinn_infer_model_version=1.0 python3
+      ${CMAKE_CURRENT_SOURCE_DIR}/test_squeezenet.py
       "${CMAKE_BINARY_DIR}/third_party/SqueezeNet" "${WITH_GPU}"
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 
@@ -145,8 +151,9 @@ if(WITH_CUDNN)
     COMMAND
       ${CMAKE_COMMAND} -E env
       PYTHONPATH=${CMAKE_BINARY_DIR}:${CMAKE_BINARY_DIR}/python/cinn:$ENV{PYTHONPATH}
-      python3 ${CMAKE_CURRENT_SOURCE_DIR}/test_paddle_model_convertor.py --path
-      "${CMAKE_BINARY_DIR}/third_party/resnet_model"
+      FLAGS_cinn_infer_model_version=1.0 python3
+      ${CMAKE_CURRENT_SOURCE_DIR}/test_paddle_model_convertor.py --path
+      "${CMAKE_BINARY_DIR}/third_party/resnet_model_1"
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 endif()
 

test/cinn/fake_model/naive_mul.py

@@ -14,7 +14,7 @@
 
 
 import paddle
-from paddle import fluid, static
+from paddle import static
 
 size = 30
 paddle.enable_static()
@@ -37,5 +37,5 @@ loss = exe = static.Executor(cpu)
 
 exe.run(static.default_startup_program())
 
-fluid.io.save_inference_model("./naive_mul_model", [a.name], [a1], exe)
+static.io.save_inference_model("./naive_mul_model", [a], [a1], exe)
 print('res is : ', a1.name)

test/cinn/fake_model/naive_multi_fc.py

@@ -17,7 +17,7 @@ A fake model with multiple FC layers to test CINN on a more complex model.
 
 
 import paddle
-from paddle import fluid, static
+from paddle import static
 
 size = 64
 num_layers = 6
@@ -54,5 +54,5 @@ loss = exe = static.Executor(cpu)
 
 exe.run(static.default_startup_program())
 
-fluid.io.save_inference_model("./multi_fc_model", [a.name], [fc_out], exe)
+static.io.save_inference_model("./multi_fc_model", [a], [fc_out], exe)
 print('res', fc_out.name)

test/cinn/fake_model/resnet_model.py

@@ -47,7 +47,8 @@ exe = static.Executor(cpu)
 
 exe.run(static.default_startup_program())
 
+static.io.save_inference_model("./resnet_model", [resnet_input], [temp7], exe)
 fluid.io.save_inference_model(
-    "./resnet_model", [resnet_input.name], [temp7], exe
+    "./resnet_model_1", [resnet_input.name], [temp7], exe
 )
 print('res', temp7.name)

test/cinn/test_resnet.py

@@ -39,7 +39,9 @@ class TestLoadResnetModel(unittest.TestCase):
         self.x_shape = [1, 160, 7, 7]
 
     def get_paddle_inference_result(self, data):
-        config = fluid.core.AnalysisConfig(self.model_dir)
+        config = fluid.core.AnalysisConfig(
+            self.model_dir + ".pdmodel", self.model_dir + ".pdiparams"
+        )
         config.disable_gpu()
         config.switch_ir_optim(False)
         self.paddle_predictor = fluid.core.create_paddle_predictor(config)
@@ -51,11 +53,11 @@ class TestLoadResnetModel(unittest.TestCase):
         np.random.seed(0)
         x_data = np.random.random(self.x_shape).astype("float32")
         self.executor = Interpreter(["resnet_input"], [self.x_shape])
-        self.executor.load_paddle_model(self.model_dir, self.target, False)
+        self.executor.load_paddle_model(self.model_dir, self.target, True)
         a_t = self.executor.get_tensor("resnet_input")
         a_t.from_numpy(x_data, self.target)
 
-        out = self.executor.get_tensor("relu_0.tmp_0")
+        out = self.executor.get_tensor("save_infer_model/scale_0.tmp_0")
         out.from_numpy(np.zeros(out.shape(), dtype='float32'), self.target)
 
         self.executor.run()