The prototype of RNN for Intel platform, especially for Xeon and Xeon Phi.
- vRNN
- LSTM
- GRU
- peephole
- Stack-X
- Bidirection-X
This is a fast implemention for RNN with fused operations and better utilization of GEMM/Batch GEMM/Pack GEMM.
Before the following steps, please make sure the pytorch is installed.
First get the code:
git clone https://github.com/pengzhao-intel/RNN.git
cd RNN
create a build directory:
mkdir build
cd build
cmake ..
make
cd ..
The C library should now be built along with test executables.
Then we bind this C library to pytorch:
cd pytorch_binding
python setup.py install
cd ..
For a simple test to varify that all the interface are called correctly:
python test.py