Add KP model list

docs/dev_guides/kernel_primitive_api/example_cn.rst

@@ -3,10 +3,12 @@ API 示例
 
 - `ElementwiseAdd <./add_example_cn.html>`_ : 加法操作，输入和输出具有相同 Shape。
 - `Reduce <./reduce_example_cn.html>`_ : 针对最高维进行规约操作。
+- `Model <./model_example_cn.html>`_ : Resnet50执行流程。
 
 ..  toctree::
     :hidden:
 
 
     add_example_cn.md
     reduce_example_cn.md
+    model_example_cn.md

docs/dev_guides/kernel_primitive_api/example_en.rst

@@ -3,10 +3,11 @@ API Examples
 
 - `ElementwiseAdd <./add_example_en.html>`_ : Addition operation, the input and output shapes are the same.
 - `Reduce <./reduce_example_en.html>`_ : Only the highest dimension is involved in reduce.
-
+- `Model <./model_example_en.html>`_ : Execution process of resnet50.
 ..  toctree::
     :hidden:
 
 
     add_example_en.md
     reduce_example_en.md
+    model_example_en.md

docs/dev_guides/kernel_primitive_api/model_example_cn.md

@@ -0,0 +1,57 @@
+# 示例 - Model
+## 模型运行说明
++ 在 GPU 平台上默认使用 Kernel Primitive API 编写的算子。
++ 在 XPU 平台上使用 Kernel Primitive API 编写的算子需要开启FLAGS\_run\_kp\_kernel环境变量。
+
+### XPU Kernel Primitive API Paddle模型运行
+
+以 resnet50 为例展示 XPU2 KP 模型运行的基本流程
+
++ 1. 下载 Paddle XPU2 KP 并安装，当前仅支持 python3.7</br>
+```
+pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/v2.3.0/xpu2_kp_py37/latest/paddlepaddle_xpu-2.3.0-cp37-cp37m-linux_x86_64.whl
+```
++ 2. 下载模型库并安装</br>
+```
+git clone -b develop https://github.com/PaddlePaddle/PaddleClas.git
+cd PaddleClas
+python -m pip install -r requirements.txt
+```
++ 3. 下载数据集<br>
+```
+cd dataset
+rm -rf ILSVRC2012
+wget -nc https://paddle-imagenet-models-name.bj.bcebos.com/data/whole_chain/whole_chain_CIFAR100.tar
+tar xf whole_chain_CIFAR100.tar
+ln -s whole_chain_CIFAR100 ILSVRC2012
+cd ILSVRC2012
+mv train.txt train_list.txt
+mv test.txt val_list.txt
+```
++ 4. 模型运行</br>
+```
+cd ../..
+export FLAGS_selected_xpus=0
+export FLAGS_run_kp_kernel=1
+export XPUSIM_DEVICE_MODEL=KUNLUN2
+nohup python tools/train.py \
+-c ppcls/configs/ImageNet/ResNet/ResNet50.yaml \
+-o Global.device=xpu > ResNet50_xpu2.log &
+```
++ 5. 成功截图如下：</br>
+![Model](./images/example_model.png)
+
+### XPU2 Kernel Primitive API 模型列表
+
+序号 | 模型名称 | 类别
+-- | -- | --
+1 | resnet50 | 图像分类
+2 | deepfm | 推荐网络
+3 | wide&deep | 推荐网络
+4 | yolov3-darknet53 | 目标检测
+5 | ssd-resnet34 | 目标检测
+6 | orc-db | 文本检测
+7 | bert-base | 自然语言
+8 | transformer | 自然语言
+9 | gpt-2 | 自然语言
+10 | unet | 图像分割

docs/dev_guides/kernel_primitive_api/model_example_en.md

@@ -0,0 +1,60 @@
+# Model
+## Description
++ On the GPU platform, the operator written by the kernel primitive API is used by default.
++ Operators written using the kernel primitive API on the XPU platform need to enable FLAGS\_run\_kp\_kernel environment variables.
+
+### XPU kernel primitive API model example
+
+Take resnet50 as an example to show the basic process of XPU2 KP model operation
+
++ 1. Download and install paddle XPU2 KP. Currently, only python3.7 is supported</br>
+```
+pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/v2.3.0/xpu2_kp_py37/latest/paddlepaddle_xpu-2.3.0-cp37-cp37m-linux_x86_64.whl
+```
+
++ 2. Download model library and install</br>
+```
+git clone -b develop https://github.com/PaddlePaddle/PaddleClas.git
+cd PaddleClas
+python -m pip install -r requirements.txt
+```
+
++ 3. Downloading datasets <br>
+
+```
+cd dataset
+rm -rf ILSVRC2012
+wget -nc https://paddle-imagenet-models-name.bj.bcebos.com/data/whole_chain/whole_chain_CIFAR100.tar
+tar xf whole_chain_CIFAR100.tar
+ln -s whole_chain_CIFAR100 ILSVRC2012
+cd ILSVRC2012
+mv train.txt train_list.txt
+mv test.txt val_list.txt
+```
+
++ 4. Model operation</br>
+```
+cd ../..
+export FLAGS_selected_xpus=0
+export FLAGS_run_kp_kernel=1
+export XPUSIM_DEVICE_MODEL=KUNLUN2
+nohup python tools/train.py \
+-c ppcls/configs/ImageNet/ResNet/ResNet50.yaml \
+-o Global.device=xpu > ResNet50_xpu2.log &
+```
++ 5. Creenshot is as follows: </br>
+![Model](./images/example_model.png)
+
+### XPU2 Kernel Primitive API Model List
+Number | Model Name | Category
+-- | -- | --
+1 | resnet50 | Image Classification
+2 | deepfm | Recommendation Network
+3 | wide&deep | Recommendation Network
+4 | yolov3-darknet53 | Object Detection
+5 | ssd-resnet34 | Object Detection
+6 | orc-db |  Text Detection
+7 | bert-base | Natural Language Processing
+8 | transformer | Natural Language Processing
+9 | gpt-2 | Natural Language Processing
+10 | unet | Image Segmentation