run_distribute_train.sh 2.7 KB
Newer Older
Z
zhaoting 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
#!/bin/bash
# Copyright 2020 Huawei Technologies Co., Ltd
#
# 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.
# ============================================================================

Z
zhaoting 已提交
17
echo "=============================================================================================================="
Z
zhaoting 已提交
18
echo "Please run the scipt as: "
19
echo "sh run_distribute_train.sh DEVICE_NUM EPOCH_SIZE LR DATASET RANK_TABLE_FILE PRE_TRAINED PRE_TRAINED_EPOCH_SIZE"
Z
zhaoting 已提交
20
echo "for example: sh run_distribute_train.sh 8 500 0.2 coco /data/hccl.json /opt/ssd-300.ckpt(optional) 200(optional)"
Z
zhaoting 已提交
21
echo "It is better to use absolute path."
C
chengxianbin 已提交
22 23
echo "================================================================================================================="

Z
zhaoting 已提交
24
if [ $# != 5 ] && [ $# != 7 ]
C
chengxianbin 已提交
25
then
Z
zhaoting 已提交
26
    echo "Usage: sh run_distribute_train.sh [DEVICE_NUM] [EPOCH_SIZE] [LR] [DATASET] \
27
[RANK_TABLE_FILE] [PRE_TRAINED](optional) [PRE_TRAINED_EPOCH_SIZE](optional)"
C
chengxianbin 已提交
28 29
    exit 1
fi
Z
zhaoting 已提交
30 31

# Before start distribute train, first create mindrecord files.
C
chengxianbin 已提交
32 33
BASE_PATH=$(cd "`dirname $0`" || exit; pwd)
cd $BASE_PATH/../ || exit
34
python train.py --only_create_dataset=True
Z
zhaoting 已提交
35 36 37 38 39

echo "After running the scipt, the network runs in the background. The log will be generated in LOGx/log.txt"

export RANK_SIZE=$1
EPOCH_SIZE=$2
Z
zhaoting 已提交
40 41 42 43
LR=$3
DATASET=$4
PRE_TRAINED=$6
PRE_TRAINED_EPOCH_SIZE=$7
44
export RANK_TABLE_FILE=$5
Z
zhaoting 已提交
45 46 47 48 49 50

for((i=0;i<RANK_SIZE;i++))
do
    export DEVICE_ID=$i
    rm -rf LOG$i
    mkdir ./LOG$i
C
chengxianbin 已提交
51 52
    cp ./*.py ./LOG$i
    cp -r ./src ./LOG$i
Z
zhaoting 已提交
53 54 55 56
    cd ./LOG$i || exit
    export RANK_ID=$i
    echo "start training for rank $i, device $DEVICE_ID"
    env > env.log
Z
zhaoting 已提交
57
    if [ $# == 5 ]
C
chengxianbin 已提交
58
    then
Z
zhaoting 已提交
59
        python train.py  \
60
        --distribute=True  \
Z
zhaoting 已提交
61
        --lr=$LR \
C
chengxianbin 已提交
62 63 64 65 66 67
        --dataset=$DATASET \
        --device_num=$RANK_SIZE  \
        --device_id=$DEVICE_ID  \
        --epoch_size=$EPOCH_SIZE > log.txt 2>&1 &
    fi

Z
zhaoting 已提交
68
    if [ $# == 7 ]
C
chengxianbin 已提交
69
    then
Z
zhaoting 已提交
70
        python train.py  \
71
        --distribute=True  \
Z
zhaoting 已提交
72
        --lr=$LR \
C
chengxianbin 已提交
73 74 75 76 77 78 79 80
        --dataset=$DATASET \
        --device_num=$RANK_SIZE  \
        --device_id=$DEVICE_ID  \
        --pre_trained=$PRE_TRAINED \
        --pre_trained_epoch_size=$PRE_TRAINED_EPOCH_SIZE \
        --epoch_size=$EPOCH_SIZE > log.txt 2>&1 &
    fi

Z
zhaoting 已提交
81 82
    cd ../
done