kora.install.kaldi doesn't support Cuda 8.0 and above (hence can't run GPU on colab)

[Adithya-Jayan]

Issue

The current version of kaldi Included in the library does not support the latest Cuda versions.

Details

To test if kaldi has been configured for GPU properly, I run ./src/nnet3bin/cuda-gpu-available

Running it produces the following output.

LOG ([5.5.0~1-40c7]:main():cuda-gpu-available.cc:61) 
### IS CUDA GPU AVAILABLE? 'f346372dbbde' ###
### CUDA WAS NOT COMPILED IN! ###
To support CUDA, you must run 'configure' on a machine that has the CUDA compiler 'nvcc' available.

Running .configure produces the following.

Configuring KALDI to use MKL.
Backing up kaldi.mk to kaldi.mk.bak ...
Checking compiler g++ ...
Checking OpenFst library in /opt/kaldi/tools/openfst-1.6.7 ...
Checking cub library in /opt/kaldi/tools/cub-1.8.0 ...
Performing OS specific configuration ...
On Linux: Checking for linear algebra header files ...
Configuring MKL library directory: Found: /opt/intel/mkl/lib/intel64
MKL configured libs: -L/opt/intel/mkl/lib/intel64 -Wl,-rpath=/opt/intel/mkl/lib/intel64 -lmkl_intel_lp64 -lmkl_core -lmkl_sequential -ldl -lpthread -lm
MKL include directory: -I/opt/intel/mkl/include
Using Intel MKL as the linear algebra library.
Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications
Successfully configured for Linux with MKL libs from /opt/intel/mkl
Unsupported CUDA_VERSION (CUDA_VERSION=8_0), please report it to Kaldi mailing list, together with 'nvcc -h' or 'ptxas -h' which lists allowed -gencode values... 

Since Kaldi has already seemed to have release a fix to support the newer version, I believe this can be fixed my creating a newer release with using a more recent version of kaldi.

How to reproduce:

• Open a new colab notebook
• Run the following code to install kaldi

# !pip install kora -q
# import kora.install.kaldi

• Run the test script with

%cd ../opt/kaldi/src
!../tools/extras/install_mkl.sh