apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-11-03 11:02:01 +08:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

[benchmark] support lite light api & optimize benchmark flags (#1950)

Browse Source 
* [benchmark] support lite light api & optimize benchmark flags

* [backend] remove un-used option member

* [backend] remove un-used option member
This commit is contained in:
DefTruth
2023-05-17 18:51:32 +08:00
committed by GitHub
parent 09ec386e8d
commit 1f66d5d9d3
6 changed files with 198 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
benchmark/cpp/benchmark.cc
View File

@@ -33,16 +33,11 @@ DEFINE_string(tensors, "tensor_a.txt:tensor_b.txt",
"The paths to dumped tensors.");
DEFINE_bool(mem, false, "Whether to force to collect memory info.");
DEFINE_int32(interval, -1, "Sampling interval for collect memory info.");
DEFINE_string(model_file, "UNKNOWN",
"Optional, set specific model file,"
"eg, model.pdmodel, model.onnx");
DEFINE_string(params_file, "",
"Optional, set specific params file,"
"eg, model.pdiparams.");
DEFINE_string(model_format, "PADDLE",
"Optional, set specific model format,"
"eg, PADDLE/ONNX/RKNN/TORCHSCRIPT/SOPHGO");
DEFINE_bool(disable_mkldnn, false, "disable mkldnn for paddle backend");
DEFINE_string(optimized_model_dir, "", "Set optimized model dir for lite backend.");
#if defined(ENABLE_BENCHMARK)
static std::vector<int64_t> GetInt64Shape(const std::vector<int>& shape) {
@@ -117,15 +112,25 @@ static void RuntimeProfiling(int argc, char* argv[]) {
auto model_format = fastdeploy::ModelFormat::PADDLE;
if (FLAGS_model_file != "UNKNOWN") {
// Set model file/param/format via command line
model_file = FLAGS_model + sep + FLAGS_model_file;
params_file = FLAGS_model + sep + FLAGS_params_file;
if (FLAGS_model != "") {
model_file = FLAGS_model + sep + FLAGS_model_file;
params_file = FLAGS_model + sep + FLAGS_params_file;
} else {
model_file = FLAGS_model_file;
params_file = FLAGS_params_file;
}
model_format = GetModelFormat(FLAGS_model_format);
if (model_format == fastdeploy::ModelFormat::PADDLE &&
FLAGS_params_file == "") {
std::cout << "[ERROR] params_file can not be empty for PADDLE"
if (config_info["backend"] != "lite") {
std::cout << "[ERROR] params_file can not be empty for PADDLE"
<< " format, Please, set your custom params_file manually."
<< std::endl;
return;
return;
} else {
std::cout << "[INFO] Will using the lite light api for: "
<< model_file << std::endl;
}
}
} else {
// Set model file/param/format via model dir (only support
@@ -140,6 +145,16 @@ static void RuntimeProfiling(int argc, char* argv[]) {
option.SetModelPath(model_file, params_file, model_format);
// Set opt model dir
if (config_info["backend"] == "lite") {
if (FLAGS_optimized_model_dir != "") {
option.paddle_lite_option.optimized_model_dir =
FLAGS_optimized_model_dir;
} else {
option.paddle_lite_option.optimized_model_dir = FLAGS_model;
}
}
// Get input shapes/names/dtypes
std::vector<std::vector<int32_t>> input_shapes =
benchmark::ResultManager::GetInputShapes(FLAGS_shapes);

18
benchmark/cpp/flags.h
View File

@@ -33,6 +33,12 @@ DEFINE_string(config_path, "config.txt", "Path of benchmark config.");
DEFINE_int32(warmup, -1, "Number of warmup for profiling.");
DEFINE_int32(repeat, -1, "Number of repeats for profiling.");
DEFINE_int32(xpu_l3_cache, -1, "Size xpu l3 cache for profiling.");
DEFINE_string(model_file, "UNKNOWN",
"Optional, set specific model file,"
"eg, model.pdmodel, model.onnx");
DEFINE_string(params_file, "",
"Optional, set specific params file,"
"eg, model.pdiparams.");
static void PrintUsage() {
std::cout << "Usage: infer_demo --model model_path --image img_path "
@@ -50,8 +56,12 @@ static void PrintBenchmarkInfo(std::unordered_map<std::string,
std::vector<std::string> model_names;
fastdeploy::benchmark::Split(FLAGS_model, model_names, sep);
if (model_names.empty()) {
std::cout << "Directory of the inference model is invalid!!!" << std::endl;
return;
if (FLAGS_model_file != "UNKNOWN") {
model_names.push_back(FLAGS_model_file);
} else {
std::cout << "[WARNING] Directory of the inference model is empty!!!"
<< std::endl;
}
}
// Save benchmark info
int warmup = std::stoi(config_info["warmup"]);
@@ -65,7 +75,9 @@ static void PrintBenchmarkInfo(std::unordered_map<std::string,
std::stringstream ss;
ss.precision(3);
ss << "\n======= Model Info =======\n";
ss << "model_name: " << model_names[model_names.size() - 1] << std::endl;
if (!model_names.empty()) {
ss << "model_name: " << model_names[model_names.size() - 1] << std::endl;
}
ss << "profile_mode: " << config_info["profile_mode"] << std::endl;
if (config_info["profile_mode"] == "runtime") {
ss << "include_h2d_d2h: " << config_info["include_h2d_d2h"] << std::endl;