can it work with out GPU?

Can it work without GPU? If yes, what shall be the step for

export CUDA_HOME="/usr/local/cuda"
cd pytorch_binding
python setup.py instal

[vermasrijan]

@ryanleary , thanks for sharing these. I tried to work around the resources that you just shared. I was getting the following error on the ‘make’ process: error: unsupported option ‘-fopenmp’ , but solved this by making some changes in the CMakeList.txt. Following are the changes that i made (referred from-https://github.com/SeanNaren/warp-ctc/issues/25): option(DWITH_OMP “compile warp-ctc with openmp.” OFF)

add_definitions(-DCTC_DISABLE_OMP)

#if(NOT WITH_OMP) #add_definitions(-DCTC_DISABLE_OMP) #endif() #if (WITH_OMP) #set(CMAKE_CXX_FLAGS “${CMAKE_CXX_FLAGS} -fopenmp”) #set(CUDA_NVCC_FLAGS “${CUDA_NVCC_FLAGS} -Xcompiler -fopenmp”) #endif()

after this, when i ran the setup.py file (in the pytorch_binding directory), this is what i got:

And when i tried to import warpctc_pytorch, i got the same error as : No module named ‘warpctc_pytorch._warp_ctc.__warp_ctc’

Maybe I am missing out on something, cant figure it out though.

Thanks!

[ryanleary]

Have you tried it? The warp-ctc library has a CPU only implementation and the pytorch bindings also support cpu only. If it isn’t working, then there’s some small bug in the build process.

See here for the warp-ctc library https://github.com/SeanNaren/warp-ctc/blob/pytorch_bindings/CMakeLists.txt#L100 “Building shared library with no GPU support” and here for the PyTorch binding: https://github.com/SeanNaren/warp-ctc/blob/pytorch_bindings/pytorch_binding/setup.py#L45 where it’s clear that having GPU support is optional.