diff --git a/PaddleNLP/dialogue_model_toolkit/auto_dialogue_evaluation/.run_ce.sh b/PaddleNLP/dialogue_model_toolkit/auto_dialogue_evaluation/.run_ce.sh
index 4851202a6ad178d1fc74412ebaea3452f5c7ace9..fea71c7a316a0a85c231a7dbea5a11bf1ddcfe80 100644
--- a/PaddleNLP/dialogue_model_toolkit/auto_dialogue_evaluation/.run_ce.sh
+++ b/PaddleNLP/dialogue_model_toolkit/auto_dialogue_evaluation/.run_ce.sh
@@ -19,7 +19,7 @@ python -u main.py \
       --save_model_path="data/saved_models/matching_pretrained" \
       --save_param="params" \
       --training_file="data/input/data/unlabel_data/train.ids" \
-      --epoch=20 \
+      --epoch=3 \
       --print_step=1 \
       --save_step=400 \
       --batch_size=256 \
@@ -31,22 +31,17 @@ python -u main.py \
       --enable_ce="store_true" | python _ce.py
 
 
-save_model_path="data/saved_models/human_finetuned"
-if [ ! -d ${save_model_path} ]
-then
-    mkdir ${save_model_path}
-fi
+export CUDA_VISIBLE_DEVICES=0,1,2,3
 
 python -u main.py \
       --do_train=true \
       --use_cuda=true \
-      --loss_type="L2" \
+      --loss_type="CLS" \
       --max_seq_len=50 \
-      --init_from_pretrain_model="data/saved_models/matching_pretrained" \
-      --save_model_path="data/saved_models/human_finetuned" \
+      --save_model_path="data/saved_models/matching_pretrained" \
       --save_param="params" \
-      --training_file="data/input/data/label_data/human/train.ids" \
-      --epoch=50 \
+      --training_file="data/input/data/unlabel_data/train.ids" \
+      --epoch=3 \
       --print_step=1 \
       --save_step=400 \
       --batch_size=256 \
@@ -56,3 +51,4 @@ python -u main.py \
       --learning_rate=0.001 \
       --sample_pro=0.1 \
       --enable_ce="store_true" | python _ce.py
+