Sync Benchmark Sample#
This sample demonstrates how to estimate performance of a model using Synchronous Inference Request API. It makes sense to use synchronous inference only in latency oriented scenarios. Models with static input shapes are supported. This sample does not have other configurable command-line arguments. Feel free to modify sample’s source code to try out different options. Before using the sample, refer to the following requirements:
The sample accepts any file format supported by
core.read_model
.The sample has been validated with: the yolo-v3-tf and face-detection-0200 models.
To build the sample, use instructions available at Build the Sample Applications section in “Get Started with Samples” guide.
How It Works#
The sample compiles a model for a given device, randomly generates input data, performs synchronous inference multiple times for a given number of seconds. Then, it processes and reports performance results.
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
# Copyright (C) 2022 Intel Corporation
# SPDX-License-Identifier: Apache-2.0
import logging as log
import statistics
import sys
from time import perf_counter
import numpy as np
import openvino as ov
from openvino.runtime import get_version
from openvino.runtime.utils.types import get_dtype
def fill_tensor_random(tensor):
dtype = get_dtype(tensor.element_type)
rand_min, rand_max = (0, 1) if dtype == bool else (np.iinfo(np.uint8).min, np.iinfo(np.uint8).max)
# np.random.uniform excludes high: add 1 to have it generated
if np.dtype(dtype).kind in ['i', 'u', 'b']:
rand_max += 1
rs = np.random.RandomState(np.random.MT19937(np.random.SeedSequence(0)))
if 0 == tensor.get_size():
raise RuntimeError("Models with dynamic shapes aren't supported. Input tensors must have specific shapes before inference")
tensor.data[:] = rs.uniform(rand_min, rand_max, list(tensor.shape)).astype(dtype)
def main():
log.basicConfig(format='[ %(levelname)s ] %(message)s', level=log.INFO, stream=sys.stdout)
log.info('OpenVINO:')
log.info(f"{'Build ':.<39} {get_version()}")
device_name = 'CPU'
if len(sys.argv) == 3:
device_name = sys.argv[2]
elif len(sys.argv) != 2:
log.info(f'Usage: {sys.argv[0]} <path_to_model> <device_name>(default: CPU)')
return 1
# Optimize for latency. Most of the devices are configured for latency by default,
# but there are exceptions like GNA
latency = {'PERFORMANCE_HINT': 'LATENCY'}
# Create Core and use it to compile a model.
# Select the device by providing the name as the second parameter to CLI.
# Using MULTI device is pointless in sync scenario
# because only one instance of openvino.runtime.InferRequest is used
core = ov.Core()
compiled_model = core.compile_model(sys.argv[1], device_name, latency)
ireq = compiled_model.create_infer_request()
# Fill input data for the ireq
for model_input in compiled_model.inputs:
fill_tensor_random(ireq.get_tensor(model_input))
# Warm up
ireq.infer()
# Benchmark for seconds_to_run seconds and at least niter iterations
seconds_to_run = 10
niter = 10
latencies = []
start = perf_counter()
time_point = start
time_point_to_finish = start + seconds_to_run
while time_point < time_point_to_finish or len(latencies) < niter:
ireq.infer()
iter_end = perf_counter()
latencies.append((iter_end - time_point) * 1e3)
time_point = iter_end
end = time_point
duration = end - start
# Report results
fps = len(latencies) / duration
log.info(f'Count: {len(latencies)} iterations')
log.info(f'Duration: {duration * 1e3:.2f} ms')
log.info('Latency:')
log.info(f' Median: {statistics.median(latencies):.2f} ms')
log.info(f' Average: {sum(latencies) / len(latencies):.2f} ms')
log.info(f' Min: {min(latencies):.2f} ms')
log.info(f' Max: {max(latencies):.2f} ms')
log.info(f'Throughput: {fps:.2f} FPS')
if __name__ == '__main__':
main()
// Copyright (C) 2022 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include <string>
#include <vector>
// clang-format off
#include "openvino/openvino.hpp"
#include "samples/args_helper.hpp"
#include "samples/common.hpp"
#include "samples/latency_metrics.hpp"
#include "samples/slog.hpp"
// clang-format on
using Ms = std::chrono::duration<double, std::ratio<1, 1000>>;
int main(int argc, char* argv[]) {
try {
slog::info << "OpenVINO:" << slog::endl;
slog::info << ov::get_openvino_version();
std::string device_name = "CPU";
if (argc == 3) {
device_name = argv[2];
} else if (argc != 2) {
slog::info << "Usage : " << argv[0] << " <path_to_model> <device_name>(default: CPU)" << slog::endl;
return EXIT_FAILURE;
}
// Optimize for latency. Most of the devices are configured for latency by default,
// but there are exceptions like GNA
ov::AnyMap latency{{ov::hint::performance_mode.name(), ov::hint::PerformanceMode::LATENCY}};
// Create ov::Core and use it to compile a model.
// Select the device by providing the name as the second parameter to CLI.
// Using MULTI device is pointless in sync scenario
// because only one instance of ov::InferRequest is used
ov::Core core;
ov::CompiledModel compiled_model = core.compile_model(argv[1], device_name, latency);
ov::InferRequest ireq = compiled_model.create_infer_request();
// Fill input data for the ireq
for (const ov::Output<const ov::Node>& model_input : compiled_model.inputs()) {
fill_tensor_random(ireq.get_tensor(model_input));
}
// Warm up
ireq.infer();
// Benchmark for seconds_to_run seconds and at least niter iterations
std::chrono::seconds seconds_to_run{10};
size_t niter = 10;
std::vector<double> latencies;
latencies.reserve(niter);
auto start = std::chrono::steady_clock::now();
auto time_point = start;
auto time_point_to_finish = start + seconds_to_run;
while (time_point < time_point_to_finish || latencies.size() < niter) {
ireq.infer();
auto iter_end = std::chrono::steady_clock::now();
latencies.push_back(std::chrono::duration_cast<Ms>(iter_end - time_point).count());
time_point = iter_end;
}
auto end = time_point;
double duration = std::chrono::duration_cast<Ms>(end - start).count();
// Report results
slog::info << "Count: " << latencies.size() << " iterations" << slog::endl
<< "Duration: " << duration << " ms" << slog::endl
<< "Latency:" << slog::endl;
size_t percent = 50;
LatencyMetrics{latencies, "", percent}.write_to_slog();
slog::info << "Throughput: " << double_to_string(latencies.size() * 1000 / duration) << " FPS" << slog::endl;
} catch (const std::exception& ex) {
slog::err << ex.what() << slog::endl;
return EXIT_FAILURE;
}
return EXIT_SUCCESS;
}
You can see the explicit description of each sample step at Integration Steps section of “Integrate OpenVINO™ Runtime with Your Application” guide.
Running#
python sync_benchmark.py <path_to_model> <device_name>(default: CPU)
sync_benchmark <path_to_model> <device_name>(default: CPU)
To run the sample, you need to specify a model. You can get a model specific for your inference task from one of model repositories, such as TensorFlow Zoo, HuggingFace, or TensorFlow Hub.
Example#
Download a pre-trained model.
You can convert it by using:
import openvino as ov ov_model = ov.convert_model('./models/googlenet-v1') # or, when model is a Python model object ov_model = ov.convert_model(googlenet-v1)
ovc ./models/googlenet-v1
Perform benchmarking, using the
googlenet-v1
model on aCPU
:python sync_benchmark.py googlenet-v1.xml
sync_benchmark googlenet-v1.xml
Sample Output#
The application outputs performance results.
[ INFO ] OpenVINO:
[ INFO ] Build ................................. <version>
[ INFO ] Count: 2333 iterations
[ INFO ] Duration: 10003.59 ms
[ INFO ] Latency:
[ INFO ] Median: 3.90 ms
[ INFO ] Average: 4.29 ms
[ INFO ] Min: 3.30 ms
[ INFO ] Max: 10.11 ms
[ INFO ] Throughput: 233.22 FPS
The application outputs performance results.
[ INFO ] OpenVINO:
[ INFO ] Build ................................. <version>
[ INFO ] Count: 992 iterations
[ INFO ] Duration: 15009.8 ms
[ INFO ] Latency:
[ INFO ] Median: 14.00 ms
[ INFO ] Average: 15.13 ms
[ INFO ] Min: 9.33 ms
[ INFO ] Max: 53.60 ms
[ INFO ] Throughput: 66.09 FPS