From f906f849812060fb225dbd1c379279a418db8c1b Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Tue, 28 Sep 2021 02:28:25 +0000 Subject: [PATCH] opt benchmark --- benchmark/run_det.sh | 16 +++--- ppocr/utils/profiler.py | 110 ++++++++++++++++++++++++++++++++++++++++ tools/program.py | 1 + 3 files changed, 119 insertions(+), 8 deletions(-) create mode 100644 ppocr/utils/profiler.py diff --git a/benchmark/run_det.sh b/benchmark/run_det.sh index c94af85c..cf1abab0 100644 --- a/benchmark/run_det.sh +++ b/benchmark/run_det.sh @@ -1,27 +1,27 @@ # 提供可稳定复现性能的脚本，默认在标准docker环境内py37执行： paddlepaddle/paddle:latest-gpu-cuda10.1-cudnn7 paddle=2.1.2 py=37 # 执行目录：需说明 -cd PaddleOCR +#cd PaddleOCR # 1 安装该模型需要的依赖 (如需开启优化策略请注明) python3.7 -m pip install -r requirements.txt # 2 拷贝该模型需要数据、预训练模型 -wget -p ./tain_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar && cd train_data && tar xf icdar2015.tar && cd ../ -wget -p ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_pretrained.pdparams +#wget -p ./tain_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar && cd train_data && tar xf icdar2015.tar && cd ../ +#wget -p ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_pretrained.pdparams # 3 批量运行（如不方便批量，1，2需放到单个模型中） model_mode_list=(det_mv3_db det_r50_vd_east) fp_item_list=(fp32) -bs_list=(256 128) +bs_list=(4 8) for model_mode in ${model_mode_list[@]}; do for fp_item in ${fp_item_list[@]}; do for bs_item in ${bs_list[@]}; do echo "index is speed, 1gpus, begin, ${model_name}" run_mode=sp - CUDA_VISIBLE_DEVICES=0 bash benchmark/run_benchmark.sh ${run_mode} ${bs_item} ${fp_item} 10 ${model_mode} # (5min) + CUDA_VISIBLE_DEVICES=3 bash benchmark/run_benchmark_det.sh ${run_mode} ${bs_item} ${fp_item} 10 ${model_mode} # (5min) sleep 60 echo "index is speed, 8gpus, run_mode is multi_process, begin, ${model_name}" - run_mode=mp - CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 bash benchmark/run_benchmark.sh ${run_mode} ${bs_item} ${fp_item} 10 ${model_mode} - sleep 60 + #run_mode=mp + #CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 bash benchmark/run_benchmark.sh ${run_mode} ${bs_item} ${fp_item} 10 ${model_mode} + #sleep 60 done done done diff --git a/ppocr/utils/profiler.py b/ppocr/utils/profiler.py new file mode 100644 index 00000000..c4e28bc6 --- /dev/null +++ b/ppocr/utils/profiler.py @@ -0,0 +1,110 @@ +# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve. +# +# 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 sys +import paddle + +# A global variable to record the number of calling times for profiler +# functions. It is used to specify the tracing range of training steps. +_profiler_step_id = 0 + +# A global variable to avoid parsing from string every time. +_profiler_options = None + + +class ProfilerOptions(object): + ''' + Use a string to initialize a ProfilerOptions. + The string should be in the format: "key1=value1;key2=value;key3=value3". + For example: + "profile_path=model.profile" + "batch_range=[50, 60]; profile_path=model.profile" + "batch_range=[50, 60]; tracer_option=OpDetail; profile_path=model.profile" + ProfilerOptions supports following key-value pair: + batch_range - a integer list, e.g. [100, 110]. + state - a string, the optional values are 'CPU', 'GPU' or 'All'. + sorted_key - a string, the optional values are 'calls', 'total', + 'max', 'min' or 'ave. + tracer_option - a string, the optional values are 'Default', 'OpDetail', + 'AllOpDetail'. + profile_path - a string, the path to save the serialized profile data, + which can be used to generate a timeline. + exit_on_finished - a boolean. + ''' + + def __init__(self, options_str): + assert isinstance(options_str, str) + + self._options = { + 'batch_range': [10, 20], + 'state': 'All', + 'sorted_key': 'total', + 'tracer_option': 'Default', + 'profile_path': '/tmp/profile', + 'exit_on_finished': True + } + self._parse_from_string(options_str) + + def _parse_from_string(self, options_str): + for kv in options_str.replace(' ', '').split(';'): + key, value = kv.split('=') + if key == 'batch_range': + value_list = value.replace('[', '').replace(']', '').split(',') + value_list = list(map(int, value_list)) + if len(value_list) >= 2 and value_list[0] >= 0 and value_list[ + 1] > value_list[0]: + self._options[key] = value_list + elif key == 'exit_on_finished': + self._options[key] = value.lower() in ("yes", "true", "t", "1") + elif key in [ + 'state', 'sorted_key', 'tracer_option', 'profile_path' + ]: + self._options[key] = value + + def __getitem__(self, name): + if self._options.get(name, None) is None: + raise ValueError( + "ProfilerOptions does not have an option named %s." % name) + return self._options[name] + + +def add_profiler_step(options_str=None): + ''' + Enable the operator-level timing using PaddlePaddle's profiler. + The profiler uses a independent variable to count the profiler steps. + One call of this function is treated as a profiler step. + + Args: + profiler_options - a string to initialize the ProfilerOptions. + Default is None, and the profiler is disabled. + ''' + if options_str is None: + return + + global _profiler_step_id + global _profiler_options + + if _profiler_options is None: + _profiler_options = ProfilerOptions(options_str) + + if _profiler_step_id == _profiler_options['batch_range'][0]: + paddle.utils.profiler.start_profiler( + _profiler_options['state'], _profiler_options['tracer_option']) + elif _profiler_step_id == _profiler_options['batch_range'][1]: + paddle.utils.profiler.stop_profiler(_profiler_options['sorted_key'], + _profiler_options['profile_path']) + if _profiler_options['exit_on_finished']: + sys.exit(0) + + _profiler_step_id += 1 diff --git a/tools/program.py b/tools/program.py index c89d7e3c..d941e717 100755 --- a/tools/program.py +++ b/tools/program.py @@ -31,6 +31,7 @@ from ppocr.utils.stats import TrainingStats from ppocr.utils.save_load import save_model from ppocr.utils.utility import print_dict from ppocr.utils.logging import get_logger +from ppocr.utils import profiler from ppocr.data import build_dataloader import numpy as np -- GitLab