A fast parallel implementation of RNN Transducer (Graves 2013 joint network), on both CPU and GPU.
GPU implementation is now available for Graves2012 add network.
Benchmarked on a GeForce GTX 1080 Ti GPU.
| T=150, L=40, A=28 | warp-transducer |
|---|---|
| N=1 | 8.51 ms |
| N=16 | 11.43 ms |
| N=32 | 12.65 ms |
| N=64 | 14.75 ms |
| N=128 | 19.48 ms |
| T=150, L=20, A=5000 | warp-transducer |
|---|---|
| N=1 | 4.79 ms |
| N=16 | 24.44 ms |
| N=32 | 41.38 ms |
| N=64 | 80.44 ms |
| N=128 | 51.46 ms |
The interface is in include/rnnt.h. It supports CPU or GPU execution, and you can specify OpenMP parallelism
if running on the CPU, or the CUDA stream if running on the GPU. We took care to ensure that the library does not
preform memory allocation internally, in oder to avoid synchronizations and overheads caused by memory allocation.
warp-transducer has been tested on Ubuntu 16.04 and CentOS 7. Windows is not supported at this time.
First get the code:
git clone https://github.com/HawkAaron/warp-transducer
cd warp-transducercreate a build directory:
mkdir build
cd buildif you have a non standard CUDA install, add -DCUDA_TOOLKIT_ROOT_DIR=/path/to/cuda option to cmake so that CMake detects CUDA.
Run cmake and build:
cmake ..
makeThe C library should now be built along with test executables. If CUDA was detected, then test_gpu will be built;
test_cpu will always be built.
To run the tests, make sure the CUDA libraries are in LD_LIBRARY_PATH (DYLD_LIBRARY_PATH for OSX).
We welcome improvements from the community, please feel free to submit pull requests.