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. Install PaddlePaddle XPU2 KP package. Currently, only python3.7 is supported.

pip install https://paddle-wheel.bj.bcebos.com/2.3.0/xpu2/kp/paddlepaddle_xpu-2.3.0-cp37-cp37m-linux_x86_64.whl

    1. Download model library and install

git clone -b develop https://github.com/PaddlePaddle/PaddleClas.git
cd PaddleClas
python -m pip install -r requirements.txt

    1. Downloading datasets 

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

    1. Model operation

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 &

    1. Creenshot is as follows:
      Model

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