diff --git a/demo/auto_compression/detection/README.md b/demo/auto_compression/detection/README.md
index 4c7dcc2a4672883c9a58b733dbdbb19e178404a4..eb75ae20c7f98d9f3cf6a307c36e0250f1a64224 100644
--- a/demo/auto_compression/detection/README.md
+++ b/demo/auto_compression/detection/README.md
@@ -45,7 +45,8 @@ pip install paddlepaddle-gpu
 
 安装paddleslim：
 ```shell
-pip install paddleslim
+https://github.com/PaddlePaddle/PaddleSlim.git
+python setup.py install
 ```
 
 安装paddledet：
diff --git a/demo/auto_compression/image_classification/README.md b/demo/auto_compression/image_classification/README.md
index 1393e23b40fd06dd831d4ea5d2d54aca6907c4a5..f0b2224e900595514a65070cf7756bd57cf38ec0 100644
--- a/demo/auto_compression/image_classification/README.md
+++ b/demo/auto_compression/image_classification/README.md
@@ -43,7 +43,8 @@ pip install paddlepaddle-gpu
 
 安装paddleslim：
 ```shell
-pip install paddleslim
+https://github.com/PaddlePaddle/PaddleSlim.git
+python setup.py install
 ```
 
 #### 3.2 准备数据集
diff --git a/demo/auto_compression/nlp/README.md b/demo/auto_compression/nlp/README.md
index 06587000732e26d62613c6e1ed7b3ddaff3d3bb2..3500cbf9d114ed865113d657d8e0e2cf9f8e0dd3 100644
--- a/demo/auto_compression/nlp/README.md
+++ b/demo/auto_compression/nlp/README.md
@@ -56,7 +56,8 @@ pip install paddlepaddle-gpu
 
 安装paddleslim：
 ```shell
-pip install paddleslim
+https://github.com/PaddlePaddle/PaddleSlim.git
+python setup.py install
 ```
 
 安装paddlenlp：
diff --git a/demo/auto_compression/semantic_segmentation/README.md b/demo/auto_compression/semantic_segmentation/README.md
index a8d592994119af0ef215657739e53b3105b555cb..5385dd5a7fdac69386f2e83284510be08f1da9cd 100644
--- a/demo/auto_compression/semantic_segmentation/README.md
+++ b/demo/auto_compression/semantic_segmentation/README.md
@@ -48,7 +48,8 @@ pip install paddlepaddle-gpu
 
 安装paddleslim：
 ```shell
-pip install paddleslim
+https://github.com/PaddlePaddle/PaddleSlim.git
+python setup.py install
 ```
 
 安装paddleseg
diff --git a/paddleslim/auto_compression/utils/predict.py b/paddleslim/auto_compression/utils/predict.py
index d72369cbe8a1e251ca311097770d55b9a6f56e7b..0e268b846fa5a5cc6d56cd4383aad81842c81f66 100644
--- a/paddleslim/auto_compression/utils/predict.py
+++ b/paddleslim/auto_compression/utils/predict.py
@@ -27,9 +27,9 @@ def predict_compressed_model(model_dir,
         latency_dict(dict): The latency latency of the model under various compression strategies.
     """
     local_rank = paddle.distributed.get_rank()
-    quant_model_path = f'quant_model/rank_{local_rank}'
-    prune_model_path = f'prune_model/rank_{local_rank}'
-    sparse_model_path = f'sparse_model/rank_{local_rank}'
+    quant_model_path = f'quant_model_rank_{local_rank}_tmp'
+    prune_model_path = f'prune_model_rank_{local_rank}_tmp'
+    sparse_model_path = f'sparse_model_rank_{local_rank}_tmp'
 
     latency_dict = {}
 
@@ -116,7 +116,7 @@ def predict_compressed_model(model_dir,
             model_dir=sparse_model_path,
             model_filename=model_filename,
             params_filename=params_filename,
-            save_model_path='quant_model',
+            save_model_path=quant_model_path,
             quantizable_op_type=["conv2d", "depthwise_conv2d", "mul"],
             is_full_quantize=False,
             activation_bits=8,
@@ -131,10 +131,10 @@ def predict_compressed_model(model_dir,
         latency_dict.update({f'sparse_{sparse_ratio}_int8': latency})
 
     # NOTE: Delete temporary model files
-    if os.path.exists('quant_model'):
-        shutil.rmtree('quant_model', ignore_errors=True)
-    if os.path.exists('prune_model'):
-        shutil.rmtree('prune_model', ignore_errors=True)
-    if os.path.exists('sparse_model'):
-        shutil.rmtree('sparse_model', ignore_errors=True)
+    if os.path.exists(quant_model_path):
+        shutil.rmtree(quant_model_path, ignore_errors=True)
+    if os.path.exists(prune_model_path):
+        shutil.rmtree(prune_model_path, ignore_errors=True)
+    if os.path.exists(sparse_model_path):
+        shutil.rmtree(sparse_model_path, ignore_errors=True)
     return latency_dict
diff --git a/paddleslim/auto_compression/utils/prune_model.py b/paddleslim/auto_compression/utils/prune_model.py
index b9e27fa0e104e57e7cb1199739645820e7f21f74..5152d06acc778ddaa9c23a83025b6fefe71f175d 100644
--- a/paddleslim/auto_compression/utils/prune_model.py
+++ b/paddleslim/auto_compression/utils/prune_model.py
@@ -122,7 +122,7 @@ def get_prune_model(model_file, param_file, ratio, save_path):
     main_prog = static.Program()
     startup_prog = static.Program()
     place = paddle.CPUPlace()
-    exe = paddle.static.Executor()
+    exe = paddle.static.Executor(place)
     scope = static.global_scope()
     exe.run(startup_prog)