diff --git a/deploy/cpp_infer/docs/windows_vs2019_build.md b/deploy/cpp_infer/docs/windows_vs2019_build.md
index 1f97f350a36fde3eaa6d0c29d8566033f45cb071..634aff172c5e7cf9dd3fd5f50b461be02840b283 100755
--- a/deploy/cpp_infer/docs/windows_vs2019_build.md
+++ b/deploy/cpp_infer/docs/windows_vs2019_build.md
@@ -12,13 +12,13 @@ PaddleClas在Windows 平台下基于`Visual Studio 2019 Community` 进行了测
 
 **下面所有示例以工作目录为 `D:\projects`演示**。
 
-### Step1: 下载PaddlePaddle C++ 预测库 fluid_inference
+### Step1: 下载PaddlePaddle C++ 预测库 paddle_inference_install_dir
 
-PaddlePaddle C++ 预测库针对不同的`CPU`和`CUDA`版本提供了不同的预编译版本，请根据实际情况下载:  [C++预测库下载列表](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/advanced_guide/inference_deployment/inference/windows_cpp_inference.html)。
+PaddlePaddle C++ 预测库针对不同的`CPU`和`CUDA`版本提供了不同的预编译版本，请根据实际情况下载:  [C++预测库下载列表](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc1/guides/05_inference_deployment/inference/windows_cpp_inference.html)。
 
-解压后`D:\projects\fluid_inference`目录包含内容为：
+解压后`D:\projects\paddle_inference_install_dir`目录包含内容为：
 ```
-fluid_inference
+paddle_inference_install_dir
 ├── paddle # paddle核心库和头文件
 |
 ├── third_party # 第三方依赖库和头文件
@@ -62,7 +62,7 @@ fluid_inference
 | CMAKE_BUILD_TYPE              | RelWithDebInfo     | [√]         |
 | CUDA_LIB                      | CUDA的库路径       | [√]         |
 | CUDNN_LIB                     | CUDNN的库路径      | [√]         |
-| OPENCV_DIR                    | OpenCV的安装路径   | [√]         |
+| OpenCV_DIR                    | OpenCV的安装路径   | [√]         |
 | PADDLE_LIB                    | Paddle预测库的路径 | [√]         |
 | WITH_GPU                      | [√]                | [√]         |
 | WITH_MKL                      | [√]                | [√]         |
diff --git a/ppcls/utils/model_zoo.py b/ppcls/utils/model_zoo.py
index 44a11c56d63cb58d31f41dfb4d39709e16930a06..d837aff2cec2121ddc2277387cda30e83815348f 100644
--- a/ppcls/utils/model_zoo.py
+++ b/ppcls/utils/model_zoo.py
@@ -58,8 +58,8 @@ class RetryError(Exception):
         super(RetryError, self).__init__(message)
 
 
-def _get_url(architecture, postfix="tar"):
-    prefix = "https://paddle-imagenet-models-name.bj.bcebos.com/"
+def _get_url(architecture, postfix="pdparams"):
+    prefix = "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/"
     fname = architecture + "_pretrained." + postfix
     return prefix + fname
 
@@ -201,7 +201,7 @@ def list_models():
     return
 
 
-def get(architecture, path, decompress=True, postfix="tar"):
+def get(architecture, path, decompress=False, postfix="pdparams"):
     """
     Get the pretrained model.
     """
diff --git a/tools/download.py b/tools/download.py
index 1523a7818a5a8c6887bedc3feb6f0a0eebf24424..7053634c54e875529fa7a1a01d325641405fa764 100644
--- a/tools/download.py
+++ b/tools/download.py
@@ -29,8 +29,8 @@ def parse_args():
     parser = argparse.ArgumentParser()
     parser.add_argument('-a', '--architecture', type=str, default='ResNet50')
     parser.add_argument('-p', '--path', type=str, default='./pretrained/')
-    parser.add_argument('--postfix', type=str, default="tar")
-    parser.add_argument('-d', '--decompress', type=str2bool, default=True)
+    parser.add_argument('--postfix', type=str, default="pdparams")
+    parser.add_argument('-d', '--decompress', type=str2bool, default=False)
     parser.add_argument('-l', '--list', type=str2bool, default=False)
 
     args = parser.parse_args()