run_12node_48gpu_resnet50_OOO_Backprop.sh

#!/bin/bash

# set up Model
export MODEL_SIZE=50
export BATCH_SIZE=128
export REVERSE_FIRST_K=30
export NUM_TRAINING_STEP=40

# set up cluster (assign IP of MASTER_HOST)
export MASTER_HOST=127.0.0.1
export MASTER_PORT=1234

# NODE_HOST_LIST[0] should be the IP of MASTER_HOST
NODE_HOST_LIST=( )
NODE_HOST_LIST[0]=NODE_IP
NODE_HOST_LIST[1]=NODE_IP
NODE_HOST_LIST[2]=NODE_IP
NODE_HOST_LIST[3]=NODE_IP
NODE_HOST_LIST[4]=NODE_IP
NODE_HOST_LIST[5]=NODE_IP
NODE_HOST_LIST[6]=NODE_IP
NODE_HOST_LIST[7]=NODE_IP
NODE_HOST_LIST[8]=NODE_IP
NODE_HOST_LIST[9]=NODE_IP
NODE_HOST_LIST[10]=NODE_IP
NODE_HOST_LIST[11]=NODE_IP

export NODE_HOSTS_STRING=${NODE_HOST_LIST[@]}

export NUM_NODE=${#NODE_HOST_LIST[@]}

# set up cluster-setting
export NUM_WORKER=48
export NUM_SERVER=12

export NUM_WORKER_PER_NODE=`expr $NUM_WORKER / $NUM_NODE`
export NUM_SERVER_PER_NODE=`expr $NUM_SERVER / $NUM_NODE`

export DEBUG_PRINT=0
export DEBUG_C_PRINT=0

# set up network
export DOCKER_IMAGE="hanyangbdsl/ooo_backprop_data_parallel:latest"
export SSH_KEY_PATH="SSH_KEY_PATH"
export SSH_ID="ACCOUNT_ID"
export DMLC_INTERFACE="NETWORK INTERFACE"

../../expr/data_par/scripts/multi_node/run.sh preset