espnet: Make install fails to find installed cupy

Cuda version: 10.0.130

Runing the make command make KALDI=../../kaldi/ I get

. venv/bin/activate; python check_install.py
/bin/sh: 4: /opt/anaconda3/etc/conda/deactivate.d/cupy_deactivate.sh: [[: not found
INFO: python version = 3.7.3 (default, Mar 27 2019, 22:11:17)
[GCC 7.3.0]
INFO: pytorch version = 1.0.1.post2
INFO: library availableness check start.
INFO: # libraries to be checked = 8
WARNING: --> cupy is not installed.
INFO: --> warpctc_pytorch is installed.
INFO: --> espnet is installed.
INFO: --> kaldiio is installed.
INFO: --> matplotlib is installed.
INFO: --> chainer is installed.
INFO: --> chainer_ctc is installed.
INFO: --> warprnnt_pytorch is installed.
INFO: library availableness check done.
INFO: 7 / 8 libraries are correctly installed.
WARNING: please try to setup again and then re-run this script.
Makefile:108: recipe for target 'check_install' failed
make: *** [check_install] Error 1

Cupy was installed with pip install -U setuptools pip and pip install cupy-cuda100 and completed successfully with messages

Installing collected packages: cupy-cuda100
Successfully installed cupy-cuda100-7.5.0

but the error persists. Thank you for your help.

About this issue

  • Original URL
  • State: closed
  • Created 4 years ago
  • Comments: 25 (1 by maintainers)

Most upvoted comments

Thanks, so your problem is clear.

original error: libcuda.so.1: cannot open shared object file: No such file or directory

You don’t have it simply.

Thank you very much. I found out that this is due to the docker build I used is not an nvidia-runtime. Once switch the build runtime to nvidia, I think that this is resolved. I have learned a lot. Thank you again.

Sorry, but I want to understand what happened for the other people. Could you pull espnet repo and make again? I updated repository to show error.

Of course. I will try this right now.

Okay thanks. Makefile installed cupy correctly. There are no problems, but you failed to import cupy due to some error, so please do:

% /usr/bin/python3
>>> import cupy

Probably, you meet some error.

Thank you. I am leaving my office. I will check this tonight… Once I got the result for the test, I put the feedback here. Thank you again for the help and your valuable time.

That is fixed in #2091. To fix quickly, we can run make cupy.done and then check via make check_install. ESPnet creates portable conda env, so you need to install within the conda env.