mirrors/PaddleClas
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/PaddleClas.git synced 2025-06-03 21:55:06 +08:00
Code Issues Projects Releases Wiki Activity

Merge pull request #1744 from mmglove/develop

Browse Source 
fix benchmark
This commit is contained in:
Walter 2022-03-09 18:55:28 +08:00 committed by GitHub
commit eee3ca70fb
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
10 changed files with 36 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
test_tipc/benchmark_train.sh
View File

@ -142,7 +142,6 @@ else
batch_size=${params_list[1]}
batch_size=`echo ${batch_size} | tr -cd "[0-9]" `
precision=${params_list[2]}
# run_process_type=${params_list[3]}
run_mode=${params_list[3]}
device_num=${params_list[4]}
IFS=";"
@ -167,10 +166,9 @@ for batch_size in ${batch_size_list[*]}; do
gpu_id=$(set_gpu_id $device_num)
if [ ${#gpu_id} -le 1 ];then
run_process_type="SingleP"
log_path="$SAVE_LOG/profiling_log"
mkdir -p $log_path
log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}_${device_num}_profiling"
log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_profiling"
func_sed_params "$FILENAME" "${line_gpuid}" "0" # sed used gpu_id
# set profile_option params
tmp=`sed -i "${line_profile}s/.*/${profile_option}/" "${FILENAME}"`
@ -186,8 +184,8 @@ for batch_size in ${batch_size_list[*]}; do
speed_log_path="$SAVE_LOG/index"
mkdir -p $log_path
mkdir -p $speed_log_path
log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}_${device_num}_log"
speed_log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}_${device_num}_speed"
log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_log"
speed_log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_speed"
func_sed_params "$FILENAME" "${line_profile}" "null" # sed profile_id as null
cmd="bash test_tipc/test_train_inference_python.sh ${FILENAME} benchmark_train > ${log_path}/${log_name} 2>&1 "
echo $cmd
@ -198,13 +196,12 @@ for batch_size in ${batch_size_list[*]}; do
eval "cat ${log_path}/${log_name}"
# parser log
_model_name="${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}"
_model_name="${model_name}_bs${batch_size}_${precision}_${run_mode}"
cmd="${python} ${BENCHMARK_ROOT}/scripts/analysis.py --filename ${log_path}/${log_name} \
--speed_log_file '${speed_log_path}/${speed_log_name}' \
--model_name ${_model_name} \
--base_batch_size ${batch_size} \
--run_mode ${run_mode} \
--run_process_type ${run_process_type} \
--fp_item ${precision} \
--keyword ips: \
--skip_steps 2 \
@ -218,13 +215,12 @@ for batch_size in ${batch_size_list[*]}; do
else
IFS=";"
unset_env=`unset CUDA_VISIBLE_DEVICES`
run_process_type="MultiP"
log_path="$SAVE_LOG/train_log"
speed_log_path="$SAVE_LOG/index"
mkdir -p $log_path
mkdir -p $speed_log_path
log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}_${device_num}_log"
speed_log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}_${device_num}_speed"
log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_log"
speed_log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_speed"
func_sed_params "$FILENAME" "${line_gpuid}" "$gpu_id" # sed used gpu_id
func_sed_params "$FILENAME" "${line_profile}" "null" # sed --profile_option as null
cmd="bash test_tipc/test_train_inference_python.sh ${FILENAME} benchmark_train > ${log_path}/${log_name} 2>&1 "
@ -235,14 +231,13 @@ for batch_size in ${batch_size_list[*]}; do
export model_run_time=$((${job_et}-${job_bt}))
eval "cat ${log_path}/${log_name}"
# parser log
_model_name="${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}"
_model_name="${model_name}_bs${batch_size}_${precision}_${run_mode}"
cmd="${python} ${BENCHMARK_ROOT}/scripts/analysis.py --filename ${log_path}/${log_name} \
--speed_log_file '${speed_log_path}/${speed_log_name}' \
--model_name ${_model_name} \
--base_batch_size ${batch_size} \
--run_mode ${run_mode} \
--run_process_type ${run_process_type} \
--fp_item ${precision} \
--keyword ips: \
--skip_steps 2 \

3
test_tipc/static/ResNet50/N1C1/ResNet50_bs256_fp16_SingleP_DP.sh → test_tipc/static/ResNet50/N1C1/ResNet50_bs256_fp16_DP.sh
View File

@ -1,7 +1,6 @@
model_item=ResNet50
bs_item=256
fp_item=fp16
run_process_type=SingleP
run_mode=DP
device_num=N1C1
max_epochs=1
@ -10,4 +9,4 @@ num_workers=8
# get data
bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
# run
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

5
test_tipc/static/ResNet50/N1C1/ResNet50_bs256_fp32_SingleP_DP.sh → test_tipc/static/ResNet50/N1C1/ResNet50_bs256_fp32_DP.sh
View File

@ -1,7 +1,6 @@
model_item=ResNet50
bs_item=256
fp_item=fp32
run_process_type=SingleP
run_mode=DP
device_num=N1C1
max_epochs=1
@ -10,8 +9,8 @@ num_workers=8
# get data
bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
# run
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
# run profiling
sleep 10;
export PROFILING=true
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

3
test_tipc/static/ResNet50/N1C1/ResNet50_bs64_fp16_SingleP_DP.sh → test_tipc/static/ResNet50/N1C1/ResNet50_bs64_fp16_DP.sh
View File

@ -1,7 +1,6 @@
model_item=ResNet50
bs_item=64
fp_item=fp16
run_process_type=SingleP
run_mode=DP
device_num=N1C1
max_epochs=1
@ -10,4 +9,4 @@ num_workers=8
# get data
bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
# run
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

5
test_tipc/static/ResNet50/N1C1/ResNet50_bs64_fp32_SingleP_DP.sh → test_tipc/static/ResNet50/N1C1/ResNet50_bs64_fp32_DP.sh
View File

@ -1,7 +1,6 @@
model_item=ResNet50
bs_item=64
fp_item=fp32
run_process_type=SingleP
run_mode=DP
device_num=N1C1
max_epochs=1
@ -10,8 +9,8 @@ num_workers=8
# get data
bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
# run
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
# run profiling
sleep 10;
export PROFILING=true
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

3
test_tipc/static/ResNet50/N1C8/ResNet50_bs256_fp16_MultiP_DP.sh → test_tipc/static/ResNet50/N1C8/ResNet50_bs256_fp16_DP.sh
View File

@ -1,7 +1,6 @@
model_item=ResNet50
bs_item=256
fp_item=fp16
run_process_type=MultiP
run_mode=DP
device_num=N1C8
max_epochs=1
@ -10,4 +9,4 @@ num_workers=8
# get data
bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
# run
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

3
test_tipc/static/ResNet50/N1C8/ResNet50_bs256_fp32_MultiP_DP.sh → test_tipc/static/ResNet50/N1C8/ResNet50_bs256_fp32_DP.sh
View File

@ -1,7 +1,6 @@
model_item=ResNet50
bs_item=256
fp_item=fp32
run_process_type=MultiP
run_mode=DP
device_num=N1C8
max_epochs=1
@ -10,4 +9,4 @@ num_workers=8
# get data
bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
# run
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

3
test_tipc/static/ResNet50/N1C8/ResNet50_bs64_fp16_MultiP_DP.sh → test_tipc/static/ResNet50/N1C8/ResNet50_bs64_fp16_DP.sh
View File

@ -1,7 +1,6 @@
model_item=ResNet50
bs_item=64
fp_item=fp16
run_process_type=MultiP
run_mode=DP
device_num=N1C8
max_epochs=1
@ -10,4 +9,4 @@ num_workers=8
# get data
bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
# run
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

3
test_tipc/static/ResNet50/N1C8/ResNet50_bs64_fp32_MultiP_DP.sh → test_tipc/static/ResNet50/N1C8/ResNet50_bs64_fp32_DP.sh
View File

@ -1,7 +1,6 @@
model_item=ResNet50
bs_item=64
fp_item=fp32
run_process_type=MultiP
run_mode=DP
device_num=N1C8
max_epochs=1
@ -10,4 +9,4 @@ num_workers=8
# get data
bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
# run
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

33
test_tipc/static/ResNet50/benchmark_common/run_benchmark.sh
View File

@ -1,23 +1,22 @@
#!/usr/bin/env bash
# Test training benchmark for a model.
# Usage：bash run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num}
# Usage：bash run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num}
function _set_params(){
model_item=${1:-"model_item"} # (必选) 模型 item
base_batch_size=${2:-"2"} # (必选) 如果是静态图单进程，则表示每张卡上的BS，需在训练时*卡数
fp_item=${3:-"fp32"} # (必选) fp32|fp16
run_process_type=${4:-"SingleP"} # (必选) 单进程 SingleP|多进程 MultiP
run_mode=${5:-"DP"} # (必选) MP模型并行|DP数据并行|PP流水线并行|混合并行DP1-MP1-PP1|DP1-MP4-PP1
device_num=${6:-"N1C1"} # (必选) 使用的卡数量，N1C1|N1C8|N4C32 （4机32卡）
run_mode=${4:-"DP"} # (必选) MP模型并行|DP数据并行|PP流水线并行|混合并行DP1-MP1-PP1|DP1-MP4-PP1
device_num=${5:-"N1C1"} # (必选) 使用的卡数量，N1C1|N1C8|N4C32 （4机32卡）
profiling=${PROFILING:-"false"} # (必选) Profiling 开关，默认关闭，通过全局变量传递
model_repo="PaddleClas" # (必选) 模型套件的名字
speed_unit="samples/sec" # (必选)速度指标单位
skip_steps=10 # (必选)解析日志，跳过模型前几个性能不稳定的step
keyword="ips:" # (必选)解析日志，筛选出性能数据所在行的关键字
convergence_key="loss:" # (可选)解析日志，筛选出收敛数据所在行的关键字 如：convergence_key="loss:"
max_epochs=${7:-"1"} # （可选）需保证模型执行时间在5分钟内，需要修改代码提前中断的直接提PR 合入套件；或使用max_epoch参数
num_workers=${8:-"4"} # (可选)
max_epochs=${6:-"1"} # （可选）需保证模型执行时间在5分钟内，需要修改代码提前中断的直接提PR 合入套件；或使用max_epoch参数
num_workers=${7:-"4"} # (可选)
# 以下为通用执行命令，无特殊可不用修改
model_name=${model_item}_bs${base_batch_size}_${fp_item}_${run_process_type}_${run_mode} # (必填) 且格式不要改动,与竞品名称对齐
model_name=${model_item}_bs${base_batch_size}_${fp_item}_${run_mode} # (必填) 且格式不要改动,与竞品名称对齐
device=${CUDA_VISIBLE_DEVICES//,/ }
arr=(${device})
num_gpu_devices=${#arr[*]}
@ -48,13 +47,19 @@ function _train(){
train_cmd="${config_file} -o DataLoader.Train.sampler.batch_size=${base_batch_size} -o Global.epochs=${max_epochs} -o DataLoader.Train.loader.num_workers=${num_workers} ${profiling_config} -o Global.eval_during_train=False"
# 以下为通用执行命令，无特殊可不用修改
case ${run_process_type} in
SingleP)
train_cmd="python ppcls/static/train.py ${train_cmd}";;
MultiP)
train_cmd="python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 ppcls/static/train.py ${train_cmd}";;
*) echo "choose run_process_type(SingleP or MultiP)"; exit 1;
case ${run_mode} in
DP) if [[ ${device_num} = "N1C1" ]];then
echo "run ${run_mode} ${device_num}"
train_cmd="python ppcls/static/train.py ${train_cmd}"
else
rm -rf ./mylog
train_cmd="python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 ppcls/static/train.py ${train_cmd}"
fi
;;
DP1-MP1-PP1) echo "run run_mode: DP1-MP1-PP1" ;;
*) echo "choose run_mode "; exit 1;
esac
echo "train_cmd: ${train_cmd} log_file: ${log_file}"
timeout 5m ${train_cmd} > ${log_file} 2>&1
if [ $? -ne 0 ];then
@ -63,7 +68,7 @@ function _train(){
echo -e "${model_name}, SUCCESS"
fi
# kill -9 `ps -ef|grep 'python'|awk '{print $2}'`
if [ ${run_process_type} = "MultiP" -a -d mylog ]; then
if [ ${device_num} != "N1C1" -a -d mylog ]; then
rm ${log_file}
cp mylog/workerlog.0 ${log_file}
fi