提交 46532c44 编写于 作者: D dongshuilong

fix train list bug for benchmark

上级 13a29e1a
...@@ -179,6 +179,22 @@ for batch_size in ${batch_size_list[*]}; do ...@@ -179,6 +179,22 @@ for batch_size in ${batch_size_list[*]}; do
func_sed_params "$FILENAME" "${line_epoch}" "$epoch" func_sed_params "$FILENAME" "${line_epoch}" "$epoch"
gpu_id=$(set_gpu_id $device_num) gpu_id=$(set_gpu_id $device_num)
# if bs is big, then copy train_list.txt to generate more train log
# There are 5w image in train_list. And the train log printed interval is 10 iteration.
# At least 25 log number would be good to calculate ips for benchmark system.
# So the copy number for train_list is as follows:
total_batch_size=`echo $[$batch_size*${device_num:1:1}*${device_num:3:3}]`
copy_num=`echo $[$total_batch_size/200]`
if [ $copy_num -gt 1 ];then
cd dataset/ILSVRC2012
rm -rf train_list.txt
for ((i=1; i <=$copy_num; i++));do
cat val_list.txt >> train_list.txt
done
cd ../../
fi
if [ ${#gpu_id} -le 1 ];then if [ ${#gpu_id} -le 1 ];then
log_path="$SAVE_LOG/profiling_log" log_path="$SAVE_LOG/profiling_log"
mkdir -p $log_path mkdir -p $log_path
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册