diff --git a/cv/classification/inceptionv3/tensorflow/get_num_devices.sh b/cv/classification/inceptionv3/tensorflow/get_num_devices.sh
index 7c6036a715270f99169e8cc384ba8ae12af26a60..a9c3708955c785f98159cfa872157538f6aee1a8 100644
--- a/cv/classification/inceptionv3/tensorflow/get_num_devices.sh
+++ b/cv/classification/inceptionv3/tensorflow/get_num_devices.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright (c) 2023, Shanghai Iluvatar CoreX Semiconductor Co., Ltd.
+# Copyright (c) 2023-2024, Shanghai Iluvatar CoreX Semiconductor Co., Ltd.
 # All Rights Reserved.
 #
 #    Licensed under the Apache License, Version 2.0 (the "License"); you may
@@ -19,8 +19,8 @@ if [ -n "$devices"  ]; then
     _devices=(${devices//,/ })
     num_devices=${#_devices[@]}
 else
-    num_devices=8
-    export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
+    num_devices=2
+    export CUDA_VISIBLE_DEVICES=0,1
     echo "Not found CUDA_VISIBLE_DEVICES, set nproc_per_node = ${num_devices}"
 fi
 export IX_NUM_CUDA_VISIBLE_DEVICES=${num_devices}
diff --git a/cv/classification/inceptionv3/tensorflow/run_train_distributed_imagenette.sh b/cv/classification/inceptionv3/tensorflow/run_train_distributed_imagenette.sh
index 1787fb081eb448fe832764edeb9fb0fc606123ba..1abf5d2c05bc628985e1e32ede54848e3829e366 100644
--- a/cv/classification/inceptionv3/tensorflow/run_train_distributed_imagenette.sh
+++ b/cv/classification/inceptionv3/tensorflow/run_train_distributed_imagenette.sh
@@ -25,7 +25,8 @@ export TF_CPP_MIN_LOG_LEVEL=1
 
 i=0
 model="alexnet"
-for arg in "$@"; do
+for arg in "$@"
+do
     if [ $i -eq 0 ]; then
         model=$arg
         let i++
@@ -40,11 +41,12 @@ for arg in "$@"; do
 done
 echo "## Training model: ${model}"
 
+
 : ${BATCH_SIZE:=16}
 # TRAIN_EPOCHS=10
 # optional optimizer: momentum, rmsprop, momentum, sgd
 OPTIMIZER=momentum
-DATE=$(date +%Y%m%d%H%M%S)
+DATE=`date +%Y%m%d%H%M%S`
 
 LOG_DIR="logs/${model}_distributed"
 DATA_DIR=./imagenette
@@ -56,7 +58,8 @@ mkdir -p ${BASE_DIR}
 rm -rf ${TRAIN_DIR}
 
 EXIT_STATUS=0
-check_status() {
+check_status()
+{
     if ((${PIPESTATUS[0]} != 0)); then
         EXIT_STATUS=1
     fi
@@ -66,7 +69,7 @@ check_status() {
 # Prepare devices
 #################################################
 devices=$CUDA_VISIBLE_DEVICES
-if [ -n "$devices" ]; then
+if [ -n "$devices"  ]; then
     devices=(${devices//,/ })
     num_devices=${#devices[@]}
 else
@@ -86,7 +89,8 @@ fi
 #################################################
 worker_hosts=""
 i=0
-for device in "${devices[@]}"; do
+for device in "${devices[@]}";
+do
     if [ "$i" == "0" ]; then
         let i++
         continue
@@ -102,13 +106,13 @@ echo "worker_hosts: ${worker_hosts}"
 #################################################
 trap ctrl_c INT
 function ctrl_c() {
-    echo "*** Trapped CTRL-C, killing process running background"
-    for pid in "${pid_list[@]}"; do
-        echo "Killing pid ${pid}"
-        kill ${pid}
-        wait ${pid}
-    done
-    exit 0
+  echo "*** Trapped CTRL-C, killing process running background"
+  for pid in "${pid_list[@]}"; do
+    echo "Killing pid ${pid}"
+    kill ${pid}
+    wait ${pid}
+  done
+  exit 0
 }
 
 #################################################
@@ -116,9 +120,10 @@ function ctrl_c() {
 #################################################
 
 pid_list=()
-last_device=$(expr ${num_devices} - 1)
+last_device=`expr ${num_devices} - 1`
 i=0
-for device in "${devices[@]}"; do
+for device in "${devices[@]}";
+do
     job_name="worker"
     if [ "${i}" == "0" ]; then
         job_name="ps"
@@ -127,26 +132,30 @@ for device in "${devices[@]}"; do
     if [ ${i} -le 1 ]; then
         task_index=0
     else
-        task_index=$(expr ${i} - 1)
+        task_index=`expr ${i} - 1`
     fi
 
     if [ "${i}" == "${last_device}" ]; then
-        CUDA_VISIBLE_DEVICES=${device} UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py --data_name=imagenette --data_dir=${DATA_DIR} \
-            --data_format=NCHW \
-            --optimizer=${OPTIMIZER} --datasets_use_prefetch=False --local_parameter_device=gpu --num_gpus=${num_devices} \
-            --batch_size=${BATCH_SIZE} --model=${model} \
-            --variable_update=distributed_replicated \
-            --job_name=${job_name} --ps_hosts=127.0.0.1:50000 --worker_hosts="${worker_hosts}" \
-            --train_dir=${TRAIN_DIR} --task_index=${task_index} --print_training_accuracy=True "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log
-        [[ ${PIPESTATUS[0]} == 0 ]] || exit
+        CUDA_VISIBLE_DEVICES=${device} UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py\
+         --data_name=imagenette --data_dir=${DATA_DIR}\
+         --data_format=NCHW \
+         --optimizer=${OPTIMIZER} --datasets_use_prefetch=False\
+         --local_parameter_device=gpu --num_gpus=${num_devices}\
+         --batch_size=${BATCH_SIZE} --model=${model} \
+         --variable_update=distributed_replicated \
+         --job_name=${job_name} --ps_hosts=127.0.0.1:50000 --worker_hosts="${worker_hosts}"\
+         --train_dir=${TRAIN_DIR} --task_index=${task_index} --print_training_accuracy=True "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log; [[ ${PIPESTATUS[0]} == 0 ]] || exit
         echo "Distributed training PID ($!) on device ${device} where job name = ${job_name}"
     else
-        CUDA_VISIBLE_DEVICES=${device} UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py --data_name=imagenette --data_dir=${DATA_DIR} \
-            --data_format=NCHW \
-            --optimizer=${OPTIMIZER} --datasets_use_prefetch=False --local_parameter_device=gpu --num_gpus=${num_devices} \
-            --batch_size=${BATCH_SIZE} --model=${model} \
-            --variable_update=distributed_replicated --job_name=${job_name} --ps_hosts=127.0.0.1:50000 --worker_hosts="${worker_hosts}" \
-            --train_dir=${TRAIN_DIR} --task_index=${task_index} --print_training_accuracy=True "${new_args[@]}" &
+        CUDA_VISIBLE_DEVICES=${device} UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py\
+         --data_name=imagenette --data_dir=${DATA_DIR}\
+         --data_format=NCHW \
+         --optimizer=${OPTIMIZER} --datasets_use_prefetch=False\
+         --local_parameter_device=gpu --num_gpus=${num_devices}\
+         --batch_size=${BATCH_SIZE} --model=${model}\
+         --variable_update=distributed_replicated\
+         --job_name=${job_name} --ps_hosts=127.0.0.1:50000 --worker_hosts="${worker_hosts}"\
+         --train_dir=${TRAIN_DIR} --task_index=${task_index} --print_training_accuracy=True "${new_args[@]}" &
         echo "Distributed training PID ($!) on device ${device} where job name = ${job_name} and task_index = ${task_index}"
     fi
     let i++
diff --git a/cv/classification/inceptionv3/tensorflow/run_train_inception3_imagenette.sh b/cv/classification/inceptionv3/tensorflow/run_train_inception3_imagenette.sh
index 12337021f8e73fc39f05e2e487812626730ca96c..9c44c920f824220dc29d47550e4e5d33aedc547f 100644
--- a/cv/classification/inceptionv3/tensorflow/run_train_inception3_imagenette.sh
+++ b/cv/classification/inceptionv3/tensorflow/run_train_inception3_imagenette.sh
@@ -23,7 +23,7 @@ export TF_CPP_MIN_LOG_LEVEL=1
 #TRAIN_EPOCHS=10
 # optional optimizer: momentum, rmsprop, momentum, sgd
 OPTIMIZER=rmsprop
-DATE=$(date +%Y%m%d%H%M%S)
+DATE=`date +%Y%m%d%H%M%S`
 
 LOG_DIR="logs/inception3"
 DATA_DIR=./imagenette
@@ -35,14 +35,16 @@ mkdir -p ${BASE_DIR}
 rm -rf ${TRAIN_DIR}
 
 EXIT_STATUS=0
-check_status() {
+check_status()
+{
     if ((${PIPESTATUS[0]} != 0)); then
         EXIT_STATUS=1
     fi
 }
 
 i=0
-for arg in "$@"; do
+for arg in "$@"
+do
     if [[ $arg =~ "--epoch" ]]; then
         new_args[$i]="--num_epochs"
     else
@@ -51,10 +53,14 @@ for arg in "$@"; do
     let i++
 done
 
-python3 -u tf_cnn_benchmarks.py --data_name=imagenette --data_dir=${DATA_DIR} \
-    --data_format=NCHW --batch_size=${BATCH_SIZE} \
-    --model=inception3 --optimizer=${OPTIMIZER} --num_gpus=1 --weight_decay=1e-4 --train_dir=${TRAIN_DIR} \
-    --eval_during_training_every_n_epochs=2 --num_eval_epochs=1 --datasets_use_caching --stop_at_top_1_accuracy=0.9 --num_intra_threads=1 --num_inter_threads=1 "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log
-[[ ${PIPESTATUS[0]} == 0 ]] || exit
+python3 -u tf_cnn_benchmarks.py\
+ --data_name=imagenette --data_dir=${DATA_DIR}\
+ --data_format=NCHW --batch_size=${BATCH_SIZE}\
+ --model=inception3 --optimizer=${OPTIMIZER} --num_gpus=1\
+ --weight_decay=1e-4 --train_dir=${TRAIN_DIR}\
+ --eval_during_training_every_n_epochs=2\
+ --num_eval_epochs=1 --datasets_use_caching\
+ --stop_at_top_1_accuracy=0.9\
+ --num_intra_threads=1 --num_inter_threads=1 "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log; [[ ${PIPESTATUS[0]} == 0 ]] || exit
 
 exit ${EXIT_STATUS}
diff --git a/cv/classification/inceptionv3/tensorflow/run_train_inception3_multigpu_imagenette.sh b/cv/classification/inceptionv3/tensorflow/run_train_inception3_multigpu_imagenette.sh
index c48d0cbfd97bf0571cc52dfa9a8a8686e5178b07..6a8938ec5f1daf3bbdb0ad57d3151518d2aa171e 100644
--- a/cv/classification/inceptionv3/tensorflow/run_train_inception3_multigpu_imagenette.sh
+++ b/cv/classification/inceptionv3/tensorflow/run_train_inception3_multigpu_imagenette.sh
@@ -23,7 +23,7 @@ export TF_CPP_MIN_LOG_LEVEL=1
 #TRAIN_EPOCHS=10
 # optional optimizer: momentum, rmsprop, momentum, sgd
 OPTIMIZER=rmsprop
-DATE=$(date +%Y%m%d%H%M%S)
+DATE=`date +%Y%m%d%H%M%S`
 
 LOG_DIR="logs/inception3_multigpu"
 DATA_DIR=./imagenette
@@ -35,14 +35,16 @@ mkdir -p ${BASE_DIR}
 rm -rf ${TRAIN_DIR}
 
 EXIT_STATUS=0
-check_status() {
+check_status()
+{
     if ((${PIPESTATUS[0]} != 0)); then
         EXIT_STATUS=1
     fi
 }
 
 i=0
-for arg in "$@"; do
+for arg in "$@"
+do
     if [[ $arg =~ "--epoch" ]]; then
         new_args[$i]="--num_epochs"
     else
@@ -53,11 +55,14 @@ done
 
 source ./get_num_devices.sh
 
-UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py --data_name=imagenette --data_dir=${DATA_DIR} \
-    --data_format=NCHW --batch_size=${BATCH_SIZE} \
-    --model=inception3 --optimizer=${OPTIMIZER} --num_gpus=${IX_NUM_CUDA_VISIBLE_DEVICES} \
-    --weight_decay=1e-4 --train_dir=${TRAIN_DIR} \
-    --eval_during_training_every_n_epochs=2 --num_eval_epochs=1 --datasets_use_caching --stop_at_top_1_accuracy=0.9 --all_reduce_spec=pscpu --num_intra_threads=1 --num_inter_threads=1 "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log
-[[ ${PIPESTATUS[0]} == 0 ]] || exit
+UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py\
+ --data_name=imagenette --data_dir=${DATA_DIR}\
+ --data_format=NCHW --batch_size=${BATCH_SIZE}\
+ --model=inception3 --optimizer=${OPTIMIZER} --num_gpus=${IX_NUM_CUDA_VISIBLE_DEVICES}\
+ --weight_decay=1e-4 --train_dir=${TRAIN_DIR}\
+ --eval_during_training_every_n_epochs=2\
+ --num_eval_epochs=1 --datasets_use_caching\
+ --stop_at_top_1_accuracy=0.9 --all_reduce_spec=pscpu\
+ --num_intra_threads=1 --num_inter_threads=1 "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log; [[ ${PIPESTATUS[0]} == 0 ]] || exit
 
 exit ${EXIT_STATUS}
diff --git a/cv/classification/resnet50/tensorflow/benchmark_cnn.py b/cv/classification/resnet50/tensorflow/benchmark_cnn.py
index 6f65ea69b46f479a649c81aaddc797f30809c1ae..7f6c1db4c5ee81194c43a463c5bfad1235b0a04f 100644
--- a/cv/classification/resnet50/tensorflow/benchmark_cnn.py
+++ b/cv/classification/resnet50/tensorflow/benchmark_cnn.py
@@ -31,9 +31,11 @@ import re
 import threading
 import time
 import traceback
+import sys
 
 from absl import flags as absl_flags
 import numpy as np
+import math
 
 import six
 from six.moves import xrange  # pylint: disable=redefined-builtin
@@ -881,6 +883,9 @@ def benchmark_one_step(sess,
     lossval = results['average_loss']
   else:
     lossval = 0.
+  if not math.isfinite(lossval):
+    print("Loss is {}, stopping training".format(lossval))
+    sys.exit(1)
   if image_producer is not None:
     image_producer.notify_image_consumption()
   train_time = time.time() - start_time
diff --git a/cv/classification/resnet50/tensorflow/run_train_distributed_imagenette.sh b/cv/classification/resnet50/tensorflow/run_train_distributed_imagenette.sh
index 98639e5c9f5656c7a46bcc5a1f00609c1170a3f9..f4f48223c71319eb1ca461986d6f81850b57b212 100644
--- a/cv/classification/resnet50/tensorflow/run_train_distributed_imagenette.sh
+++ b/cv/classification/resnet50/tensorflow/run_train_distributed_imagenette.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright (c) 2023, Shanghai Iluvatar CoreX Semiconductor Co., Ltd.
+# Copyright (c) 2023-2024, Shanghai Iluvatar CoreX Semiconductor Co., Ltd.
 # All Rights Reserved.
 #
 #    Licensed under the Apache License, Version 2.0 (the "License"); you may
@@ -14,7 +14,6 @@
 #    License for the specific language governing permissions and limitations
 #    under the License.
 
-
 bash ./get_imagenette.sh
 
 export TF_CUDNN_USE_AUTOTUNE=1
@@ -43,7 +42,7 @@ done
 echo "## Training model: ${model}"
 
 
-: ${BATCH_SIZE:=32}
+: ${BATCH_SIZE:=16}
 # TRAIN_EPOCHS=10
 # optional optimizer: momentum, rmsprop, momentum, sgd
 OPTIMIZER=momentum
@@ -111,6 +110,7 @@ function ctrl_c() {
   for pid in "${pid_list[@]}"; do
     echo "Killing pid ${pid}"
     kill ${pid}
+    wait ${pid}
   done
   exit 0
 }
diff --git a/cv/classification/vgg/tensorflow/get_num_devices.sh b/cv/classification/vgg/tensorflow/get_num_devices.sh
index 14d6c0a5a7dc9a1b93e5b389d715c36aee2aa618..a9c3708955c785f98159cfa872157538f6aee1a8 100644
--- a/cv/classification/vgg/tensorflow/get_num_devices.sh
+++ b/cv/classification/vgg/tensorflow/get_num_devices.sh
@@ -1,4 +1,5 @@
-# Copyright (c) 2023, Shanghai Iluvatar CoreX Semiconductor Co., Ltd.
+#!/bin/bash
+# Copyright (c) 2023-2024, Shanghai Iluvatar CoreX Semiconductor Co., Ltd.
 # All Rights Reserved.
 #
 #    Licensed under the Apache License, Version 2.0 (the "License"); you may
@@ -18,8 +19,8 @@ if [ -n "$devices"  ]; then
     _devices=(${devices//,/ })
     num_devices=${#_devices[@]}
 else
-    num_devices=8
-    export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
+    num_devices=2
+    export CUDA_VISIBLE_DEVICES=0,1
     echo "Not found CUDA_VISIBLE_DEVICES, set nproc_per_node = ${num_devices}"
 fi
-export IX_NUM_CUDA_VISIBLE_DEVICES=${num_devices}
\ No newline at end of file
+export IX_NUM_CUDA_VISIBLE_DEVICES=${num_devices}
diff --git a/cv/classification/vgg/tensorflow/run_train_distributed_imagenette.sh b/cv/classification/vgg/tensorflow/run_train_distributed_imagenette.sh
index 36b41ee8530f24ce9301138963981817ecceb4de..c7642b98633c5e291901347fd34bef03758bd635 100644
--- a/cv/classification/vgg/tensorflow/run_train_distributed_imagenette.sh
+++ b/cv/classification/vgg/tensorflow/run_train_distributed_imagenette.sh
@@ -25,7 +25,8 @@ export TF_CPP_MIN_LOG_LEVEL=1
 
 i=0
 model="alexnet"
-for arg in "$@"; do
+for arg in "$@"
+do
     if [ $i -eq 0 ]; then
         model=$arg
         let i++
@@ -40,11 +41,12 @@ for arg in "$@"; do
 done
 echo "## Training model: ${model}"
 
+
 : ${BATCH_SIZE:=16}
 # TRAIN_EPOCHS=10
 # optional optimizer: momentum, rmsprop, momentum, sgd
 OPTIMIZER=momentum
-DATE=$(date +%Y%m%d%H%M%S)
+DATE=`date +%Y%m%d%H%M%S`
 
 LOG_DIR="logs/${model}_distributed"
 DATA_DIR=./imagenette
@@ -56,7 +58,8 @@ mkdir -p ${BASE_DIR}
 rm -rf ${TRAIN_DIR}
 
 EXIT_STATUS=0
-check_status() {
+check_status()
+{
     if ((${PIPESTATUS[0]} != 0)); then
         EXIT_STATUS=1
     fi
@@ -66,7 +69,7 @@ check_status() {
 # Prepare devices
 #################################################
 devices=$CUDA_VISIBLE_DEVICES
-if [ -n "$devices" ]; then
+if [ -n "$devices"  ]; then
     devices=(${devices//,/ })
     num_devices=${#devices[@]}
 else
@@ -86,7 +89,8 @@ fi
 #################################################
 worker_hosts=""
 i=0
-for device in "${devices[@]}"; do
+for device in "${devices[@]}";
+do
     if [ "$i" == "0" ]; then
         let i++
         continue
@@ -102,12 +106,12 @@ echo "worker_hosts: ${worker_hosts}"
 #################################################
 trap ctrl_c INT
 function ctrl_c() {
-    echo "*** Trapped CTRL-C, killing process running background"
-    for pid in "${pid_list[@]}"; do
-        echo "Killing pid ${pid}"
-        kill ${pid}
-    done
-    exit 0
+  echo "*** Trapped CTRL-C, killing process running background"
+  for pid in "${pid_list[@]}"; do
+    echo "Killing pid ${pid}"
+    kill ${pid}
+  done
+  exit 0
 }
 
 #################################################
@@ -115,9 +119,10 @@ function ctrl_c() {
 #################################################
 
 pid_list=()
-last_device=$(expr ${num_devices} - 1)
+last_device=`expr ${num_devices} - 1`
 i=0
-for device in "${devices[@]}"; do
+for device in "${devices[@]}";
+do
     job_name="worker"
     if [ "${i}" == "0" ]; then
         job_name="ps"
@@ -126,26 +131,30 @@ for device in "${devices[@]}"; do
     if [ ${i} -le 1 ]; then
         task_index=0
     else
-        task_index=$(expr ${i} - 1)
+        task_index=`expr ${i} - 1`
     fi
 
     if [ "${i}" == "${last_device}" ]; then
-        CUDA_VISIBLE_DEVICES=${device} UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py --data_name=imagenette --data_dir=${DATA_DIR} \
-            --data_format=NCHW \
-            --optimizer=${OPTIMIZER} --datasets_use_prefetch=False --local_parameter_device=gpu --num_gpus=${num_devices} \
-            --batch_size=${BATCH_SIZE} --model=${model} \
-            --variable_update=distributed_replicated \
-            --job_name=${job_name} --ps_hosts=127.0.0.1:50000 --worker_hosts="${worker_hosts}" \
-            --train_dir=${TRAIN_DIR} --task_index=${task_index} --print_training_accuracy=True "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log
-        [[ ${PIPESTATUS[0]} == 0 ]] || exit
+        CUDA_VISIBLE_DEVICES=${device} UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py\
+         --data_name=imagenette --data_dir=${DATA_DIR}\
+         --data_format=NCHW \
+         --optimizer=${OPTIMIZER} --datasets_use_prefetch=False\
+         --local_parameter_device=gpu --num_gpus=${num_devices}\
+         --batch_size=${BATCH_SIZE} --model=${model} \
+         --variable_update=distributed_replicated \
+         --job_name=${job_name} --ps_hosts=127.0.0.1:50000 --worker_hosts="${worker_hosts}"\
+         --train_dir=${TRAIN_DIR} --task_index=${task_index} --print_training_accuracy=True "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log; [[ ${PIPESTATUS[0]} == 0 ]] || exit
         echo "Distributed training PID ($!) on device ${device} where job name = ${job_name}"
     else
-        CUDA_VISIBLE_DEVICES=${device} UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py --data_name=imagenette --data_dir=${DATA_DIR} \
-            --data_format=NCHW \
-            --optimizer=${OPTIMIZER} --datasets_use_prefetch=False --local_parameter_device=gpu --num_gpus=${num_devices} \
-            --batch_size=${BATCH_SIZE} --model=${model} \
-            --variable_update=distributed_replicated --job_name=${job_name} --ps_hosts=127.0.0.1:50000 --worker_hosts="${worker_hosts}" \
-            --train_dir=${TRAIN_DIR} --task_index=${task_index} --print_training_accuracy=True "${new_args[@]}" &
+        CUDA_VISIBLE_DEVICES=${device} UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py\
+         --data_name=imagenette --data_dir=${DATA_DIR}\
+         --data_format=NCHW \
+         --optimizer=${OPTIMIZER} --datasets_use_prefetch=False\
+         --local_parameter_device=gpu --num_gpus=${num_devices}\
+         --batch_size=${BATCH_SIZE} --model=${model}\
+         --variable_update=distributed_replicated\
+         --job_name=${job_name} --ps_hosts=127.0.0.1:50000 --worker_hosts="${worker_hosts}"\
+         --train_dir=${TRAIN_DIR} --task_index=${task_index} --print_training_accuracy=True "${new_args[@]}" &
         echo "Distributed training PID ($!) on device ${device} where job name = ${job_name} and task_index = ${task_index}"
     fi
     let i++
diff --git a/cv/classification/vgg/tensorflow/run_train_vgg16_imagenette.sh b/cv/classification/vgg/tensorflow/run_train_vgg16_imagenette.sh
index aeea001bcade9882bd62a66aea792200898842e2..343dbe372b175d10e7a6c6855e3348125e0d117d 100644
--- a/cv/classification/vgg/tensorflow/run_train_vgg16_imagenette.sh
+++ b/cv/classification/vgg/tensorflow/run_train_vgg16_imagenette.sh
@@ -23,7 +23,7 @@ export TF_CPP_MIN_LOG_LEVEL=1
 #TRAIN_EPOCHS=10
 # optional optimizer: adam, rmsprop, momentum, sgd
 OPTIMIZER=momentum
-DATE=$(date +%Y%m%d%H%M%S)
+DATE=`date +%Y%m%d%H%M%S`
 
 LOG_DIR="logs/vgg16"
 DATA_DIR=./imagenette
@@ -35,14 +35,16 @@ mkdir -p ${BASE_DIR}
 rm -rf ${TRAIN_DIR}
 
 EXIT_STATUS=0
-check_status() {
+check_status()
+{
     if ((${PIPESTATUS[0]} != 0)); then
         EXIT_STATUS=1
     fi
 }
 
 i=0
-for arg in "$@"; do
+for arg in "$@"
+do
     if [[ $arg =~ "--epoch" ]]; then
         new_args[$i]="--num_epochs"
     else
@@ -51,13 +53,15 @@ for arg in "$@"; do
     let i++
 done
 
-python3 -u tf_cnn_benchmarks.py --data_name=imagenette --data_dir=${DATA_DIR} \
-    --data_format=NCHW --batch_size=${BATCH_SIZE} \
-    --model=vgg16 --optimizer=${OPTIMIZER} --num_gpus=1 \
-    --weight_decay=1e-4 --train_dir=${TRAIN_DIR} \
-    --eval_during_training_every_n_epochs=2 --num_eval_epochs=1 \
-    --datasets_use_caching --stop_at_top_1_accuracy=0.9 --num_intra_threads=1 \
-    --num_inter_threads=1 "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log
-[[ ${PIPESTATUS[0]} == 0 ]] || exit
+python3 -u tf_cnn_benchmarks.py\
+ --data_name=imagenette --data_dir=${DATA_DIR}\
+ --data_format=NCHW --batch_size=${BATCH_SIZE}\
+ --model=vgg16 --optimizer=${OPTIMIZER} --num_gpus=1\
+ --weight_decay=1e-4 --train_dir=${TRAIN_DIR}\
+ --eval_during_training_every_n_epochs=2\
+ --num_eval_epochs=1 --datasets_use_caching\
+ --stop_at_top_1_accuracy=0.9\
+ --num_intra_threads=1 --num_inter_threads=1 "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log; [[ ${PIPESTATUS[0]} == 0 ]] || exit
+
 
 exit ${EXIT_STATUS}
diff --git a/cv/classification/vgg/tensorflow/run_train_vgg16_multigpu_imagenette.sh b/cv/classification/vgg/tensorflow/run_train_vgg16_multigpu_imagenette.sh
index 4de70ff725f1a22d56f1e8e80f9f3ca73d1b3fff..d4c2bcd2ab01f05ee348fbf32c330596dd1b4d88 100644
--- a/cv/classification/vgg/tensorflow/run_train_vgg16_multigpu_imagenette.sh
+++ b/cv/classification/vgg/tensorflow/run_train_vgg16_multigpu_imagenette.sh
@@ -23,7 +23,7 @@ export TF_CPP_MIN_LOG_LEVEL=1
 #TRAIN_EPOCHS=10
 # optional optimizer: adam, rmsprop, momentum, sgd
 OPTIMIZER=momentum
-DATE=$(date +%Y%m%d%H%M%S)
+DATE=`date +%Y%m%d%H%M%S`
 
 LOG_DIR="logs/vgg16_multigpu"
 DATA_DIR=./imagenette
@@ -35,14 +35,16 @@ mkdir -p ${BASE_DIR}
 rm -rf ${TRAIN_DIR}
 
 EXIT_STATUS=0
-check_status() {
+check_status()
+{
     if ((${PIPESTATUS[0]} != 0)); then
         EXIT_STATUS=1
     fi
 }
 
 i=0
-for arg in "$@"; do
+for arg in "$@"
+do
     if [[ $arg =~ "--epoch" ]]; then
         new_args[$i]="--num_epochs"
     else
@@ -53,13 +55,15 @@ done
 
 source ./get_num_devices.sh
 
-UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py --data_name=imagenette --data_dir=${DATA_DIR} \
-    --data_format=NCHW --batch_size=${BATCH_SIZE} \
-    --model=vgg16 --optimizer=${OPTIMIZER} --num_gpus=${IX_NUM_CUDA_VISIBLE_DEVICES} \
-    --weight_decay=1e-4 --train_dir=${TRAIN_DIR} \
-    --eval_during_training_every_n_epochs=2 --num_eval_epochs=1 --datasets_use_caching \
-    --stop_at_top_1_accuracy=0.9 --all_reduce_spec=pscpu --num_intra_threads=1 \
-    --num_inter_threads=1 "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log
-[[ ${PIPESTATUS[0]} == 0 ]] || exit
+UMD_WAITAFTERLAUNCH=1 python3 -u tf_cnn_benchmarks.py\
+ --data_name=imagenette --data_dir=${DATA_DIR}\
+ --data_format=NCHW --batch_size=${BATCH_SIZE}\
+ --model=vgg16 --optimizer=${OPTIMIZER} --num_gpus=${IX_NUM_CUDA_VISIBLE_DEVICES}\
+ --weight_decay=1e-4 --train_dir=${TRAIN_DIR}\
+ --eval_during_training_every_n_epochs=2\
+ --num_eval_epochs=1 --datasets_use_caching\
+ --stop_at_top_1_accuracy=0.9 --all_reduce_spec=pscpu\
+ --num_intra_threads=1 --num_inter_threads=1 "${new_args[@]}" 2>&1 | tee ${LOG_DIR}/${DATE}_${TRAIN_EPOCHS}_${BATCH_SIZE}_${OPTIMIZER}.log; [[ ${PIPESTATUS[0]} == 0 ]] || exit
+
 
 exit ${EXIT_STATUS}
diff --git a/cv/face/facenet/tensorflow/init.sh b/cv/face/facenet/tensorflow/init.sh
index 7631e4b8060a61c36698b8d052972bc857d99fd5..13864c5f493c362b62659b35435b2893baeed2be 100644
--- a/cv/face/facenet/tensorflow/init.sh
+++ b/cv/face/facenet/tensorflow/init.sh
@@ -13,12 +13,20 @@
 #    WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
 #    License for the specific language governing permissions and limitations
 #    under the License.
-
-PY_VERSION=$(python3 -V 2>&1 | awk '{print $2}' | awk -F '.' '{print $2}')
-if [ "$PY_VERSION" == "10" ] || [ "$PY_VERSION" == "8" ] || [ "$PY_VERSION" == "9" ]; then
+PY_VERSION=$(python3 -V 2>&1|awk '{print $2}'|awk -F '.' '{print $2}')
+if [ "$PY_VERSION" == "10" ] || [ "$PY_VERSION" == "8" ] || [ "$PY_VERSION" == "9" ];
+then
    pip3 install -r requirements.txt
-   pip3 install scipy==1.7.2
+   pip3 install scipy==1.7.2 
+   pip3 install numpy==1.23.5
 else
    pip3 install -r requirements.txt
    pip3 install scipy
 fi
+
+cd data
+wget -q http://10.150.9.95/swapp/datasets/cv/face/lfw_data.tar.gz
+wget -q http://10.150.9.95/swapp/datasets/cv/face/webface_182_44.tar
+tar -zxf lfw_data.tar.gz
+tar -xf webface_182_44.tar
+cd -
diff --git a/cv/face/facenet/tensorflow/train_facenet_ddp.sh b/cv/face/facenet/tensorflow/train_facenet_ddp.sh
index ce406a10de98f86e927fa63554f7a2286203b103..948172bb8fd9f0d8415fc46364702ed5fc016df7 100644
--- a/cv/face/facenet/tensorflow/train_facenet_ddp.sh
+++ b/cv/face/facenet/tensorflow/train_facenet_ddp.sh
@@ -13,41 +13,40 @@
 #    WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
 #    License for the specific language governing permissions and limitations
 #    under the License.
-
 EXIT_STATUS=0
-check_status() {
+check_status()
+{
     if ((${PIPESTATUS[0]} != 0)); then
         EXIT_STATUS=1
     fi
 }
 
 horovodrun -np 16 --gloo python3 src/train_softmax_ddp.py \
-    --logs_base_dir ./logs/facenet/ \
-    --models_base_dir ./src/models/ \
-    --data_dir ./data/webface_182_44 \
-    --image_size 160 \
-    --model_def models.inception_resnet_v1 \
-    --lfw_dir ./data/lfw_data/lfw_160/ \
-    --learning_rate -1 \
-    --batch_size 128 \
-    --optimizer ADAM \
-    --max_nrof_epochs 500 \
-    --keep_probability 0.8 \
-    --random_flip \
-    --random_crop \
-    --use_fixed_image_standardization \
-    --learning_rate_schedule_file ./data/learning_rate_schedule_classifier_casia_ddp.txt \
-    --weight_decay 5e-4 \
-    --embedding_size 512 \
-    --lfw_distance_metric 1 \
-    --lfw_use_flipped_images \
-    --lfw_subtract_mean \
-    --validation_set_split_ratio 0.01 \
-    --validate_every_n_epochs 5 \
-    --prelogits_norm_loss_factor 5e-4 \
-    --gpu_memory_fraction 0.9 \
-    --seed 43 \
-    --epoch_size 200 "$@"
-check_status
+        --logs_base_dir ./logs/facenet/ \
+        --models_base_dir ./src/models/ \
+        --data_dir ./data/webface_182_44 \
+        --image_size 160 \
+        --model_def models.inception_resnet_v1 \
+        --lfw_dir ./data/lfw_data/lfw_160/ \
+        --learning_rate -1 \
+        --batch_size 128 \
+        --optimizer ADAM \
+        --max_nrof_epochs 500 \
+        --keep_probability 0.8 \
+        --random_flip \
+        --random_crop \
+        --use_fixed_image_standardization \
+        --learning_rate_schedule_file ./data/learning_rate_schedule_classifier_casia_ddp.txt \
+        --weight_decay 5e-4 \
+        --embedding_size 512 \
+        --lfw_distance_metric 1 \
+        --lfw_use_flipped_images \
+        --lfw_subtract_mean \
+        --validation_set_split_ratio 0.01 \
+        --validate_every_n_epochs 5 \
+        --prelogits_norm_loss_factor 5e-4 \
+        --gpu_memory_fraction 0.9 \
+        --seed 43 \
+        --epoch_size 200 "$@";  check_status
 
 exit ${EXIT_STATUS}