cuDNN install Error

I got cuDNN v4(cudnn-7.0-linux-x64-v4.0-rc.tgz).

And I copied the *.h files to /usr/local/cuda-7.5/include/
and the .so files to /usr/local/cuda-7.5/lib64/ .

I set the environment variables LD_LIBRARY_PATH, LIBRARY_PATH and CPATH to the above directories.

Then, in order to confirm CUDA and cuDNN enable,

import cupy
was no problem.

import cupy.cudnn
I got the below Error Message,

import cupy.cudnn
Traceback (most recent call last):
File “”, line 1, in
File “/usr/local/lib/python2.7/dist-packages/cupy/”, line 7, in
from cupy.cuda import cudnn
ImportError: cannot import name cudnn

Please tell me how to treat this problem.
(My enviroment is Ubuntu14.04)

What’s the framework using in your tests? Did you configure/build it with enabled cuDNN support?

I use CUDA Toolkit 7.5 “”.

Which Deep Learning framework are you using?

cuDNN accelerates widely-used deep learning frameworks, including Caffe, Theano and Torch as well as many other software projects. Please see the documentation for each framework/application on enabling cuDNN acceleration

I use mainly Chainer, but sometimes use Caffe and Theano.

I think my situation dosen’t depend on Deep Learning frameworks.

Simply, I could not execute “>>>import cupy.cudnn” at Python source.

It seems your PATH were not set up correctly when you installed Chainer.

First untar the cudnn-7.0-linux-x64-v4.0-rc.tgz archive file in the right directory (no need to copy), with:
$ sudo tar xvfz cudnn-7.0-linux-x64-v4.0-rc.tgz -C /usr/local

Set up your paths:
$ export CPATH=/usr/local/cuda/include:CPATH export LIBRARY_PATH=/usr/local/cuda/lib:LIBRARY_PATH export LD_LIBRARY_PATH=/usr/local/cuda/lib:$LD_LIBRARY_PATH
(to have that persistent, do that in your .bashrc file and source it). Note that this is also needed for other frameworks.

Then install chainer with:
$ pip install --user chainer

And chainer will have cudnn support, you’ll be able to have:

import cupy.cudnn

I really appreciate it.

I change PATH and installed chainer agein following your advise,
I succeeded “>>>import cupy.cudnn”!!

Thanks for your kindful treatment.