1、创建pytorch环境
conda create -n pytorch python=3.6
2、安装pytorch0.4
conda install pytorch=0.4.0 cuda80 -c pytorch
3、编译ctc
git clone https://github.com/SeanNaren/warp-ctc.git cd warp-ctc mkdir build; cd build cmake .. make
编译过程中会一般不会出现问题,
cd pytorch_binding python setup.py install
这个过程会出现问题:
/home/sobhe/repo/warp-ctc/pytorch_binding/src/binding.cpp:105:61: error: too few arguments to function ‘cudaError_t THCudaMalloc(THCState*, void**, size_t)
修改文件 "pytorch_binding/src/binding.cpp"
1 at 92 line
int probs_size = THCudaTensor_size(state, probs, 2);
2 at l05 lines
void* gpu_workspace;
THCudaMalloc(state, &gpu_workspace, gpu_size_bytes);
再次运行:
cd pytorch_binding
python setup.py install
恭喜你,成功了