- Environment
The code was tested on GHC machines. Run "source env.sh" to enable CUDA and TensorFlow environment.
- Performance
Modify the settings in makefile. Use "make demo" to run sequential code, TensorFlow, and CuLSTM. Use "make seqlstm/tflstm/cu" to each of them. The result is in "res.txt".
- CuLSTM Interface
In "culstm" directory, modify "test.py" to test different LSTM settings. Use "make" to generate and run CUDA code.
- Directories
- seq: Sequential code in python
- tf: TensorFlow code
- rnn: Plain CUDA code
- culstm: Auto-generator library