diff --git a/fluid/PaddleCV/deeplabv3+/.run_ce.sh b/fluid/PaddleCV/deeplabv3+/.run_ce.sh
index 540fb964ba94fd29dc28bb51342cdba839d433e7..c4e6055e1d9d3ad9a9b039d5973c100e76a8aadf 100755
--- a/fluid/PaddleCV/deeplabv3+/.run_ce.sh
+++ b/fluid/PaddleCV/deeplabv3+/.run_ce.sh
@@ -20,7 +20,7 @@ cudaid=${deeplabv3plus_m:=0,1,2,3} # use 0,1,2,3 card as default
 export CUDA_VISIBLE_DEVICES=$cudaid
 
 FLAGS_benchmark=true  python train.py \
---batch_size=2 \
+--batch_size=8 \
 --train_crop_size=769 \
 --total_step=50 \
 --save_weights_path=output4 \
diff --git a/fluid/PaddleCV/gan/cycle_gan/trainer.py b/fluid/PaddleCV/gan/cycle_gan/trainer.py
index 84d4c87a53d1b1d3f9b1ef794f373f50dd6da175..07e8d22f4610eb57cf4fb7d17d2d8bf6cd05ac7c 100644
--- a/fluid/PaddleCV/gan/cycle_gan/trainer.py
+++ b/fluid/PaddleCV/gan/cycle_gan/trainer.py
@@ -13,6 +13,8 @@ class GATrainer():
         self.program = fluid.default_main_program().clone()
         with fluid.program_guard(self.program):
             self.fake_B = build_generator_resnet_9blocks(input_A, name="g_A")
+            #FIXME set persistable explicitly to pass CE
+            self.fake_B.persistable = True
             self.fake_A = build_generator_resnet_9blocks(input_B, name="g_B")
             self.cyc_A = build_generator_resnet_9blocks(self.fake_B, "g_B")
             self.cyc_B = build_generator_resnet_9blocks(self.fake_A, "g_A")
@@ -58,6 +60,8 @@ class GBTrainer():
         with fluid.program_guard(self.program):
             self.fake_B = build_generator_resnet_9blocks(input_A, name="g_A")
             self.fake_A = build_generator_resnet_9blocks(input_B, name="g_B")
+            #FIXME set persistable explicitly to pass CE
+            self.fake_A.persistable = True
             self.cyc_A = build_generator_resnet_9blocks(self.fake_B, "g_B")
             self.cyc_B = build_generator_resnet_9blocks(self.fake_A, "g_A")
             self.infer_program = self.program.clone()
diff --git a/fluid/PaddleNLP/machine_reading_comprehension/run.py b/fluid/PaddleNLP/machine_reading_comprehension/run.py
index e9ba1d0b14023f75d7551728dd22571cf8b72fa4..54f9c9eda9f77c62022b6020e5f9a73ad473708a 100644
--- a/fluid/PaddleNLP/machine_reading_comprehension/run.py
+++ b/fluid/PaddleNLP/machine_reading_comprehension/run.py
@@ -207,10 +207,14 @@ def validation(inference_program, avg_cost, s_probs, e_probs, match, feed_order,
     """
         
     """
+    build_strategy = fluid.BuildStrategy()
+    build_strategy.enable_inplace = False
+    build_strategy.memory_optimize = False
     parallel_executor = fluid.ParallelExecutor(
         main_program=inference_program,
         use_cuda=bool(args.use_gpu),
-        loss_name=avg_cost.name)
+        loss_name=avg_cost.name,
+        build_strategy=build_strategy)
     print_para(inference_program, parallel_executor, logger, args)
 
     # Use test set as validation each pass
@@ -523,7 +527,7 @@ def evaluate(logger, args):
 
             inference_program = main_program.clone(for_test=True)
             eval_loss, bleu_rouge = validation(
-                inference_program, avg_cost, s_probs, e_probs, match, 
+                inference_program, avg_cost, s_probs, e_probs, match,
                 feed_order, place, dev_count, vocab, brc_data, logger, args)
             logger.info('Dev eval loss {}'.format(eval_loss))
             logger.info('Dev eval result: {}'.format(bleu_rouge))
diff --git a/fluid/PaddleRec/tagspace/README.md b/fluid/PaddleRec/tagspace/README.md
index 055099dbd29be6655472b8832ea08b608a810f1c..4263065bee2c5492684147f532e92c7c8083e16f 100644
--- a/fluid/PaddleRec/tagspace/README.md
+++ b/fluid/PaddleRec/tagspace/README.md
@@ -29,7 +29,9 @@ Tagspace模型学习文本及标签的embedding表示，应用于工业级的标
 
 ## 数据下载及预处理
 
-[ag news dataset](https://github.com/mhjabreel/CharCNN/tree/master/data/ag_news_csv)
+数据地址： [ag news dataset](https://github.com/mhjabreel/CharCNN/tree/master/data/)
+
+备份数据地址：[ag news dataset](https://paddle-tagspace.bj.bcebos.com/data.tar)
 
 数据格式如下
 
@@ -37,7 +39,7 @@ Tagspace模型学习文本及标签的embedding表示，应用于工业级的标
 "3","Wall St. Bears Claw Back Into the Black (Reuters)","Reuters - Short-sellers, Wall Street's dwindling\band of ultra-cynics, are seeing green again."
 ```
 
-将文本数据转为paddle数据，先将数据放到训练数据目录和测试数据目录
+备份数据解压后，将文本数据转为paddle数据，先将数据放到训练数据目录和测试数据目录
 ```
 mv train.csv raw_big_train_data
 mv test.csv raw_big_test_data
@@ -59,7 +61,7 @@ CUDA_VISIBLE_DEVICES=0 python train.py  --use_cuda 1
 ```
 CPU 环境
 ```
-python train.py 
+python train.py
 ```
 
 全量数据单机单卡训练