idiap / pkwrap

A pytorch wrapper for LF-MMI training and parallel training in Kaldi
Other
73 stars 12 forks source link

fatal error: cublas_v2.h: No such file or directory #20

Closed iezhanqingran closed 3 years ago

iezhanqingran commented 3 years ago

Hi Srikanth,

When I complied the pkwrap I got the following error: error

I was using cuda11.2+torch1.9+kaldi (commit 7547a9008317ebb574df0520e2c3f99ac676f375)

Could you please figure out what's wrong with it?

Thanks, Qingran

mrsrikanth commented 3 years ago

Hi @iezhanqingran ,

you may want to try including the path to cuda libraries in your CXXFLAGS. Perhaps something like this might help to set before compiling:

export CXXFLAGS="$CXXFLAGS -I/path/to/cuda_headers"

The /path/to/cuda_headers is the folder that would contain cublas_v2.h

Srikanth

iezhanqingran commented 3 years ago

Hi,

I set the following environment: export CXXFLAGS="$CXXFLAGS -I/usr/local/cuda-11.1/targets/x86_64-linux/include" And compiled the Kaldi, I got some errors: /home/qzhan/kaldi/tools/openfst-1.7.2/include/fst/symbol-table.h:221: undefined reference to fst::internal::SymbolTableImpl::ReadText(std::istream&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, fst::SymbolTableTextOptions const&)' lattice-best-path.o: In function fst::internal::FstImpl<fst::ArcTpl<fst::LatticeWeightTpl<float> > >::WriteFstHeader(fst::Fst<fst::ArcTpl<fst::LatticeWeightTpl<float> > > const&, std::ostream&, fst::FstWriteOptions const&, int, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, unsigned long, fst::FstHeader*)': Do you have ideas on how to solve it? Thanks!

Qingran

mrsrikanth commented 3 years ago

I think you might have to recompile the tools folder in Kaldi with the new CXXFLAGS.

iezhanqingran commented 3 years ago

Hi I solved it by ln -s /usr/local/cuda-11.1/targets/x86_64-linux/include/* ~/kaldi/src

Thanks, Qingran