diff --git a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/README.md b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/README.md
index ad6fa55819c430d0984848e39e52bea1a55ce529..9f487365fde4cce6996c68318d26a484a80e89a3 100644
--- a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/README.md
+++ b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/README.md
@@ -107,13 +107,20 @@ fairseq-preprocess \
   --workers 70
 
 ```
-
+# 性能测试
+```单卡训练流程：
+1.安装环境
+2.bash ./test/train_performance_1p.sh --data_path=/train_data/en_ro（数据集路径为当前磁盘路径，写到en_ro）
+```
+```多卡训练流程：
+1.安装环境
+2.bash ./test/train_performance_8p.sh --data_path=/train_data/en_ro（数据集路径为当前磁盘路径，写到en_ro）
+```
 # 在数据集上进行fine-tune
-
 ```bash
-1. 修改./test/train_full_8p.sh中PRETRAIN为模型的路径，DATA_PATH为数据集的路径（train_data/en_ro 或train_data/en_de）
+1. 修改./test/train_full_8p.sh中DATA_PATH为数据集的路径（train_data/en_ro 或train_data/en_de）,PRETRAIN为model.pt的路径，BPE_PATH为sentence.bpe.model路径；
  [若需要训练en_de数据集，则需要将train_full_8p.sh中dropout的参数设置为0.1,total-num-update与max-update设置为300000，target-lang设置为de_DE]
-2. 执行 bash ./test/train_full_8p.sh
+2. 执行 bash ./test/train_full_8p.sh --data_path=DATA_PATH
 ```
 # 在数据集上进行评估
 
@@ -129,11 +136,11 @@ pip3.7 install sacrebleu==1.5.1
 
 ```bash
 验证en_ro精度
-1. 修改generate_on_en_ro.sh中DATA_PATH为数据集的路径，BPE_PATH为sentence.bpe.model的路径，SCRIPTS为mosesdecoder/scripts的路径，WMT16_SCRIPTS为wmt16-scripts的路径
-2. 执行 bash ./test/generate_on_en_ro.sh  checkpoints/checkpoint_best.pt 验证en_ro的训练精度
+1. 修改 generate_on_en_ro.sh中DATA_PATH为数据集的路径，BPE_PATH为sentence.bpe.model的路径，SCRIPTS为mosesdecoder/scripts的路径，WMT16_SCRIPTS为wmt16-scripts的路径
+2. 执行 bash generate_on_en_ro.sh  checkpoints/checkpoint_best.pt 验证en_ro的训练精度
 验证en_de精度
-1.  修改generate_on_en_de.sh中DATA_PATH为数据集的路径，BPE_PATH为sentence.bpe.model的路径，DETOKENIZER为mosesdecoder/scripts/tokenizer/detokenizer.perl的路径
-2.  执行 bash ./test/generate_on_en_de.sh  checkpoints/checkpoint_best.pt 验证en_de的训练精度
+1.  修改 generate_on_en_de.sh中DATA_PATH为数据集的路径，BPE_PATH为sentence.bpe.model的路径，DETOKENIZER为mosesdecoder/scripts/tokenizer/detokenizer.perl的路径
+2.  执行 bash generate_on_en_de.sh  checkpoints/checkpoint_best.pt 验证en_de的训练精度
 ```
 
 # Docker容器训练
diff --git a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/generate_on_en_de.sh b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/generate_on_en_de.sh
similarity index 97%
rename from PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/generate_on_en_de.sh
rename to PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/generate_on_en_de.sh
index 7e13b087392c54d9996f155cc1f409c63298ba0d..6b4547e41a0f4c7c573d6956778a3177a8472ded 100644
--- a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/generate_on_en_de.sh
+++ b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/generate_on_en_de.sh
@@ -1,4 +1,4 @@
-source env.sh
+source ./test/env_npu.sh
 DATA_PATH=path_of_data                     # fix it to your own train data path
 BPE_PATH=/path/sentence.bpe.model         # fix it to your own sentence.bpe.model path
 langs=ar_AR,cs_CZ,de_DE,en_XX,es_XX,et_EE,fi_FI,fr_XX,gu_IN,hi_IN,it_IT,ja_XX,kk_KZ,ko_KR,lt_LT,lv_LV,my_MM,ne_NP,nl_XX,ro_RO,ru_RU,si_LK,tr_TR,vi_VN,zh_CN
diff --git a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/generate_on_en_ro.sh b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/generate_on_en_ro.sh
similarity index 94%
rename from PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/generate_on_en_ro.sh
rename to PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/generate_on_en_ro.sh
index 18e527b1086180eb5d05a78bcbc9353017d2a811..e9efd429e9dbcd63b935eb3ee64c9cf4067e99af 100644
--- a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/generate_on_en_ro.sh
+++ b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/generate_on_en_ro.sh
@@ -1,4 +1,4 @@
-source env.sh
+source ./test/env_npu.sh
 DATA_PATH=path_of_data                     # fix it to your own train data path
 BPE_PATH=/path/sentence.bpe.model         # fix it to your own sentence.bpe.model path
 SCRIPTS=mosesdecoder/scripts              # fix it to your own mosesdecoder path
@@ -34,8 +34,8 @@ for f in $HYP $REF
 	perl $REPLACE_UNICODE_PUNCT | \
 	perl $NORM_PUNC -l ro | \
 	perl $REM_NON_PRINT_CHAR | \
-	python3 $NORMALIZE_ROMANIAN | \
-	python3 $REMOVE_DIACRITICS | \
+	python3.7 $NORMALIZE_ROMANIAN | \
+	python3.7 $REMOVE_DIACRITICS | \
 	perl $TOKENIZER -no-escape -threads 16 -a -l ro >"en_ro."$f
 	done
 sacrebleu -tok 'none' -s 'none' en_ro.ref < en_ro.hyp
\ No newline at end of file
diff --git a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_full_8p.sh b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_full_8p.sh
index eb9b5ffa21a018997992a7bfdc9390cd16177e12..cc9262dcc63a704f2e48ac5a17d093c846bf68f1 100644
--- a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_full_8p.sh
+++ b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_full_8p.sh
@@ -11,6 +11,16 @@ export RANK_SIZE=8
 Network="mBART_for_PyTorch"
 #训练batch_size
 token_size=1024
+# 数据集路径,保持为空,不需要修改
+data_path=""
+
+# 参数校验，data_path为必传参数，其他参数的增删由模型自身决定；此处新增参数需在上面有定义并赋值
+for para in $*
+do
+    if [[ $para == --data_path* ]];then
+        data_path=`echo ${para#*=}`
+    fi
+done
 
 #校验是否传入data_path,不需要修改
 if [[ $data_path  == "" ]];then
@@ -18,7 +28,8 @@ if [[ $data_path  == "" ]];then
 	exit 1
 fi
 
-# cd到与test文件同层级目录下执行脚本，提高兼容性；test_path_dir为包含test文件夹的路径
+###############指定训练脚本执行路径###############
+# cd到与test文件夹同层级目录下执行脚本，提高兼容性；test_path_dir为包含test文件夹的路径
 cur_path=`pwd`
 cur_path_last_dirname=${cur_path##*/}
 if [ x"${cur_path_last_dirname}" == x"test" ]; then
@@ -51,7 +62,7 @@ if [ x"${etp_flag}" != x"true" ];then
 fi
 
 # 将对应的数据以及模型等放到对应路径 或 修改以下路径以适应本地训练
-DATA_PATH=train_data/en_ro
+DATA_PATH=train_data/en_ro            
 PRETRAIN=mbart.cc25/model.pt
 BPE_PATH=mbart.cc25/sentence.bpe.model
 model_dir=checkpoints/checkpoint_best.pt
@@ -84,7 +95,7 @@ do
 		then
 			let a=0+RANK_ID*24
 			let b=23+RANK_ID*24
-			taskset -c $a-$b fairseq-train $DATA_PATH --fp16 --distributed-world-size 8 --npu \
+			nohup taskset -c $a-$b fairseq-train $DATA_PATH --fp16 --distributed-world-size 8 --npu \
 							  --device-id $RANK_ID --distributed-rank $RANK_ID --distributed-no-spawn --max-update 40000 \
 							  --encoder-normalize-before --decoder-normalize-before \
 							  --arch mbart_large --layernorm-embedding \
@@ -100,9 +111,9 @@ do
 							  --restore-file $PRETRAIN \
 							  --reset-optimizer --reset-meters --reset-dataloader --reset-lr-scheduler \
 							  --langs $langs \
-							  --ddp-backend no_c10d > ${test_path_dir}/output/${RANK_ID}/train_${RANK_ID}.log 2>&1 &
+							  --ddp-backend no_c10d & > ${test_path_dir}/output/${ASCEND_DEVICE_ID}/train_${ASCEND_DEVICE_ID}.log 2>&1 &
 	else
-		fairseq-train $DATA_PATH --fp16 --distributed-world-size 8 --npu \
+		nohup fairseq-train $DATA_PATH --fp16 --distributed-world-size 8 --npu \
 							  --device-id $RANK_ID --distributed-rank $RANK_ID --distributed-no-spawn --max-update 40000 \
 							  --encoder-normalize-before --decoder-normalize-before \
 							  --arch mbart_large --layernorm-embedding \
@@ -118,7 +129,7 @@ do
 							  --restore-file $PRETRAIN \
 							  --reset-optimizer --reset-meters --reset-dataloader --reset-lr-scheduler \
 							  --langs $langs \
-							  --ddp-backend no_c10d > ${test_path_dir}/output/${RANK_ID}/train_${RANK_ID}.log 2>&1 &
+							  --ddp-backend no_c10d  > ${test_path_dir}/output/${ASCEND_DEVICE_ID}/train_${ASCEND_DEVICE_ID}.log 2>&1 &
 	fi
 done
 wait
@@ -127,7 +138,7 @@ end_time=$(date +%s)
 e2e_time=$(( $end_time - $start_time ))
 
 
-fairseq-generate $DATA_PATH \
+nohup fairseq-generate $DATA_PATH \
   --fp16 --path $model_dir --max-tokens 4096 \
   --task translation_from_pretrained_bart \
   --gen-subset test \
@@ -146,15 +157,15 @@ for f in $HYP $REF
 	perl $REPLACE_UNICODE_PUNCT | \
 	perl $NORM_PUNC -l ro | \
 	perl $REM_NON_PRINT_CHAR | \
-	python3 $NORMALIZE_ROMANIAN | \
-	python3 $REMOVE_DIACRITICS | \
+	python3.7 $NORMALIZE_ROMANIAN | \
+	python3.7 $REMOVE_DIACRITICS | \
 	perl $TOKENIZER -no-escape -threads 16 -a -l ro >"en_ro."$f
 	done
 sacrebleu -tok 'none' -s 'none' en_ro.ref < en_ro.hyp > res.log
 wait
 ASCEND_DEVICE_ID=0
 
-
+cp ${cur_path}/nohup.out ${test_path_dir}/output/${ASCEND_DEVICE_ID}/train_${ASCEND_DEVICE_ID}.log
 #结果打印，不需要修改
 echo "------------------ Final result ------------------"
 #输出性能FPS，需要模型审视修改
diff --git a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_performance_1p.sh b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_performance_1p.sh
index 70974d82c25d5b4a2613bee7bcd4534f9c90ac5f..d9b7154ca9241f900e56eb02fb9f838b4ef7ad8c 100644
--- a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_performance_1p.sh
+++ b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_performance_1p.sh
@@ -94,7 +94,7 @@ wait
 
 pip3.7 install --editable ./ 
 start=$(date +%s)
-nohup python3.7 ${cur_path}/train.py $data_path/en_ro/ \
+nohup python3.7 ${cur_path}/train.py $data_path/ \
   --distributed-world-size 1 --npu --npu-id $ASCEND_DEVICE_ID --fp16 --encoder-normalize-before --decoder-normalize-before \
   --arch mbart_large --layernorm-embedding \
   --task translation_from_pretrained_bart \
diff --git a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_performance_8p.sh b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_performance_8p.sh
index c4bdd78c247b5c7d5902d1b8e67cb5b44d7d4c89..b9de8aada6219a23c48b1c25937c22afc159dc78 100644
--- a/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_performance_8p.sh
+++ b/PyTorch/built-in/nlp/mBART_ID2372_for_PyTorch/test/train_performance_8p.sh
@@ -97,7 +97,7 @@ do
 		then
 			let a=0+RANK_ID*24
 			let b=23+RANK_ID*24
-			taskset -c $a-$b nohup python3.7 ${cur_path}/train.py $data_path/en_ro/ --fp16 --distributed-world-size 8 --npu \
+			taskset -c $a-$b nohup python3.7 ${cur_path}/train.py $data_path/ --fp16 --distributed-world-size 8 --npu \
 							  --device-id $RANK_ID --distributed-rank $RANK_ID --distributed-no-spawn --max-update 50 \
 							  --encoder-normalize-before --decoder-normalize-before \
 							  --arch mbart_large --layernorm-embedding \
@@ -116,7 +116,7 @@ do
 							  --langs $langs \
 							  --ddp-backend no_c10d > ${test_path_dir}/output/${RANK_ID}/train_${RANK_ID}.log 2>&1 &
 	else
-		nohup python3.7 ${cur_path}/train.py $data_path/en_ro/ --fp16 --distributed-world-size 8 --npu \
+		nohup python3.7 ${cur_path}/train.py $data_path/ --fp16 --distributed-world-size 8 --npu \
 							  --device-id $RANK_ID --distributed-rank $RANK_ID --distributed-no-spawn --max-update 50 \
 							  --encoder-normalize-before --decoder-normalize-before \
 							  --arch mbart_large --layernorm-embedding \