English | 简体中文
Please refer to the following steps to compile, install and verify the custom device implementaion for Cambricon MLU.
| Module | Version |
|---|---|
| cntoolkit | 3.1.2-1 |
| cnnl | 1.13.2-1 |
| cncl | 1.4.1-1 |
| mluops | 0.3.0-1 |
# clone source
git clone --recursive https://github.com/PaddlePaddle/PaddleCustomDevice
cd PaddleCustomDevice
# get the latest submodule source code
git submodule sync
git submodule update --remote --init --recursive# navigate to implementaion for Cambricon MLU
cd backends/mlu
# before compiling, ensure that Paddle is installed, you can run the following command
pip install paddlepaddle==0.0.0 -f https://www.paddlepaddle.org.cn/whl/linux/cpu-mkl/develop.html
# create the build directory and navigate in
mkdir build && cd build
# compile in X86_64 environment
cmake ..
make -j8
# using pip to install the output
pip install dist/paddle_custom_mlu*.whl# list available hardware backends
python -c "import paddle; print(paddle.device.get_all_custom_device_type())"
# expected output
['CustomMLU']
# run a simple model
python ../tests/test_MNIST_model.py
# expected similar output
... ...
Epoch 0 step 0, Loss = [2.3313463], Accuracy = 0.046875
Epoch 0 step 100, Loss = [1.9624571], Accuracy = 0.484375
Epoch 0 step 200, Loss = [2.002725], Accuracy = 0.453125
Epoch 0 step 300, Loss = [1.912869], Accuracy = 0.546875
Epoch 0 step 400, Loss = [1.9169667], Accuracy = 0.5625
Epoch 0 step 500, Loss = [1.9007692], Accuracy = 0.5625
Epoch 0 step 600, Loss = [1.8512673], Accuracy = 0.625
Epoch 0 step 700, Loss = [1.8759218], Accuracy = 0.59375
Epoch 0 step 800, Loss = [1.8942316], Accuracy = 0.5625
Epoch 0 step 900, Loss = [1.8966292], Accuracy = 0.5625