diff --git a/PaddleCV/image_classification/scripts/train/SE_ResNet50_vd_fp16.sh b/PaddleCV/image_classification/scripts/train/SE_ResNet50_vd_fp16.sh
index a6eb162e56250e1d28caa8fcdecdea2a86f43ca6..4f80603eff0a76c03e9dd27cc6faabed0a291c10 100644
--- a/PaddleCV/image_classification/scripts/train/SE_ResNet50_vd_fp16.sh
+++ b/PaddleCV/image_classification/scripts/train/SE_ResNet50_vd_fp16.sh
@@ -1,6 +1,6 @@
 #SE_ResNet50_vd
 
-export CUDA_VISIBLE_DEVICES=4
+export CUDA_VISIBLE_DEVICES=0
 
 export FLAGS_conv_workspace_size_limit=4000 #MB
 export FLAGS_cudnn_exhaustive_search=1
@@ -9,7 +9,9 @@ export FLAGS_cudnn_batchnorm_spatial_persistent=1
 DATA_DIR="Your image dataset path, e.g. /work/datasets/ILSVRC2012/"
 
 DATA_FORMAT="NHWC"
-USE_FP16=true #whether to use float16
+USE_AMP=true #whether to use amp
+USE_PURE_FP16=false
+MULTI_PRECISION=${USE_PURE_FP16}
 USE_DALI=true
 USE_ADDTO=true
 
@@ -26,7 +28,9 @@ python train.py \
        --data_dir=${DATA_DIR} \
        --batch_size=128 \
        --lr_strategy=cosine_decay \
-       --use_fp16=${USE_FP16} \
+       --use_amp=${USE_AMP} \
+       --use_pure_fp16=${USE_PURE_FP16} \
+       --multi_precision=${MULTI_PRECISION} \
        --data_format=${DATA_FORMAT} \
        --lr=0.1 \
        --num_epochs=200 \