benchmark_train.sh 4.3 KB
Newer Older
1 2 3 4 5
#!/bin/bash
source test_tipc/common_func.sh

# run benchmark sh 
# Usage:
L
LDOUBLEV 已提交
6
# bash run_benchmark_train.sh config.txt params
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44

function func_parser_params(){
    strs=$1
    IFS="="
    array=(${strs})
    tmp=${array[1]}
    echo ${tmp}
}

function func_sed_params(){
    filename=$1
    line=$2
    param_value=$3

    # cmd="sed -n '${line}p' $filename"
    # params=`eval $cmd`
    params=`sed -n "${line}p" $filename`
    # params=`cmd`
    #params=$($cmd)
    IFS=":"
    array=(${params})
    key=${array[0]}
    value=${array[1]}
    if [[ $value =~ 'benchmark_train' ]];then
        IFS='='
        _val=(${value})
        param_value="${_val[0]}=${param_value}"
    fi
    new_params="${key}:${param_value}"
    IFS=";"
    cmd="sed -i '${line}s/.*/${new_params}/' '${filename}'"
    # echo $cmd
    eval $cmd
}

function set_gpu_id(){
    string=$1
    _str=${string:1:6}
L
LDOUBLEV 已提交
45
    IFS="C"
46
    arr=(${_str})
L
LDOUBLEV 已提交
47
    M=${arr[0]}
48
    P=${arr[1]}
L
LDOUBLEV 已提交
49 50
    gn=`expr $P - 1`
    gpu_num=`expr $gn / $M`
51 52 53 54
    seq=`seq -s "," 0 $gpu_num`
    echo $seq
}

L
LDOUBLEV 已提交
55 56 57 58 59 60 61
function get_repo_name(){
    IFS=";"
    cur_dir=$(pwd)
    IFS="/"
    arr=(${cur_dir})
    echo ${arr[-1]}
}
62 63


L
LDOUBLEV 已提交
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159
# FILENAME=$1
# # MODE be one of ['benchmark_train']
# MODE=$2
# params=$3
IFS="\n"

FILENAME="test_tipc/configs/det_mv3_db_v2.0/train_benchmark.txt"
MODE="benchmark_train"
params="dynamic_bs8_fp32_SingleP_DP_N1C4"

# parser params from input: modeltype_bs${bs_item}_${fp_item}_${run_process_type}_${run_mode}_${device_num}
IFS="_"
params_list=(${params})
model_type=${params_list[0]}
batch_size=${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[4]}
device_num=${params_list[5]}
device_num_copy=$device_num
IFS=";"

echo $precision
# sed batchsize and precision
func_sed_params "$FILENAME" "6" "$precision"
func_sed_params "$FILENAME" "9" "$batch_size"

# parser params from train_benchmark.txt
dataline=`cat $FILENAME`
# parser params
IFS=$'\n'
lines=(${dataline})
model_name=$(func_parser_value "${lines[1]}")

# 获取benchmark_params所在的行数
line_num=`grep -n "benchmark_params" $FILENAME  | cut -d ":" -f 1`
echo $line_num # debug
echo ${lines[$line_num]} # debug
# for train log parser
line_num=`expr $line_num + 3`
ips_unit_value=$(func_parser_value "${lines[line_num]}")

line_num=`expr $line_num + 1`
skip_steps_value=$(func_parser_value "${lines[line_num]}")

echo $skip_steps_value "haha"

line_num=`expr $line_num + 1`
keyword_value=$(func_parser_value "${lines[line_num]}")

line_num=`expr $line_num + 1`
convergence_key_value=$(func_parser_value "${lines[line_num]}")

line_num=`expr $line_num + 1`
flags_value=$(func_parser_value "${lines[line_num]}")

# echo "device_num: $device_num"
# gpu_id=$(set_gpu_id "N4C32")
# echo "gpuid: $gpu_id"

gpu_id=$(set_gpu_id $device_num)
echo "gpu_id: $gpu_id"
repo_name=$(get_repo_name )
echo "repo_name: ${repo_name}"

# train_log_file=${run_log_path}/${model_repo}_${model_name}_${device_num}_log
# profiling_log_file=${profiling_log_path}/${model_repo}_${model_name}_${device_num}_profiling
SAVE_LOG="benchmark_log"

if [ ${#gpu_id} -le 1 ];then
    echo "single_gpu train"
    log_path="$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"
    echo $log_name
    func_sed_params "$FILENAME" "4" "0"  # sed used gpu_id 
    cmd="bash test_tipc/test_train_inference_python.sh benchmark_train > ${log_path}/${log_name} 2>&1 "
    echo $cmd
    # without profile
    log_path="$SAVE_LOG/train_log"
    mkdir -p $log_path
    log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}_${device_num}_log"
    func_sed_params "$FILENAME" "13" "null"  # sed used gpu_id 
    cmd="bash test_tipc/test_train_inference_python.sh benchmark_train > ${log_path}/${log_name} 2>&1 "
    echo $cmd
else
    echo "multi_gpu training"
    log_path="$SAVE_LOG/train_log"
    mkdir -p $log_path
    log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_process_type}_${run_mode}_${device_num}_log"
    func_sed_params "$FILENAME" "4" "$gpu_id"  # sed used gpu_id 
    func_sed_params "$FILENAME" "13" "$null"  # sed --profile_option as null
    cmd="bash test_tipc/test_train_inference_python.sh benchmark_train > ${log_path}/${log_name} 2>&1 "
    echo $cmd
fi
160 161