unnonouno / cudnnenv Goto Github PK

When cudnnenv command is interrupted, a new directory is not deleted.

Currently cudnnenv only support released version of cuDNN. Sometimes I want to install pre-release version such as beta version and release candidate.
I want to support local tar.gz files like this:

$ cudnnenv install-file path.to.tar.gz version-name

There is no option to show version of cudnnenv.

Hi Guys,

When i am trying to install the fossology 3.2.0rc1 version using source code, getting below error so please help me to resolve this issue.

make[3]: Entering directory /root/fossology-3.2.0rc1/src/copyright/agent' sha256sum: /tmp/tmp.ykKjQ4crki: no properly formatted SHA256 checksum lines found make[3]: *** [json.hpp] Error 1 make[3]: Leaving directory /root/fossology-3.2.0rc1/src/copyright/agent'
make[2]: *** [all] Error 2
make[2]: Leaving directory /root/fossology-3.2.0rc1/src/copyright' make[1]: *** [build-copyright] Error 2 make[1]: Leaving directory /root/fossology-3.2.0rc1/src'
make: *** [build-src] Error 2

Currently cudnnenv always uses ~/.cudnn but sometimes I want to share it with other users.

I failed to install v6-cuda8 when I executed cudnnenv install v6-cuda8. The error indicated that the internal shell command returned non-zero exit status. It seems to be echo-ing checksum in the source (__init__.py: L49) needs exactly two spaces between words to be successfully checked. Indeed, I could install without problems after adding one more space.

This reference says the spacing is ignored in newer versions of sha256sum, but I weirdly encountered this problem while having version 8.4. I guess it's better to just add another space to make sure checksum verification won't fail.

Please see below:

$ sha256sum --version
sha256sum (GNU coreutils) 8.4
Copyright (C) 2010 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>.
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.

$ echo "9b09110af48c9a4d7b6344eb4b3e344daa84987ed6177d5c44319732f3bb7f9c cudnn-8.0-linux-x64-v6.0.tgz" | sha256sum -cw
sha256sum: standard input: 1: improperly formatted SHA256 checksum line
sha256sum: standard input: no properly formatted SHA256 checksum lines found

$ echo "9b09110af48c9a4d7b6344eb4b3e344daa84987ed6177d5c44319732f3bb7f9c  cudnn-8.0-linux-x64-v6.0.tgz" | sha256sum -cw
cudnn-8.0-linux-x64-v6.0.tgz: OK

Sometimes I want to switch cudnn version based on a working directory like pyenv. I want to support local settings.

cuDNN v7 is available.

I wrote configuration of environment variable in REAME, thought it is useful if the command can show the configuration.

http://developer.download.nvidia.com now redirects to https version. You need to either call curl with -L flag or fetch from https directly. Currently cudnnenv fails because of an empty download:

/usr/local/bin/cudnnenv install v7.6.5-cuda101
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0
sha256sum: WARNING: 1 computed checksum did NOT match
cudnn-10.1-linux-x64-v7.6.5.32.tgz: FAILED
Traceback (most recent call last):
  File "/usr/local/bin/cudnnenv", line 10, in <module>
    sys.exit(main())
  File "/usr/local/lib/python3.6/dist-packages/cudnnenv/__init__.py", line 976, in main
    args.func(args)
  File "/usr/local/lib/python3.6/dist-packages/cudnnenv/__init__.py", line 860, in install
    download_if_not_exist(args.version)
  File "/usr/local/lib/python3.6/dist-packages/cudnnenv/__init__.py", line 811, in download_if_not_exist
    download_cudnn(ver)
  File "/usr/local/lib/python3.6/dist-packages/cudnnenv/__init__.py", line 805, in download_cudnn
    subprocess.check_call(cmd, shell=True)
  File "/usr/lib/python3.6/subprocess.py", line 311, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '
curl -o cudnn-10.1-linux-x64-v7.6.5.32.tgz http://developer.download.nvidia.com/compute/redist/cudnn/v7.6.5/cudnn-10.1-linux-x64-v7.6.5.32.tgz &&
echo "7eaec8039a2c30ab0bc758d303588767693def6bf49b22485a2c00bf2e136cb3  cudnn-10.1-linux-x64-v7.6.5.32.tgz" | sha256sum -cw --quiet - &&
tar -xzf cudnn-10.1-linux-x64-v7.6.5.32.tgz -C /root/.cudnn/versions/v7.6.5-cuda101 &&
rm cudnn-10.1-linux-x64-v7.6.5.32.tgz
' returned non-zero exit status 1.

How about adding a feature to install and manage NCCL? It is also prevalent in the deep learning community and many users install it like cuDNN, so it would be very useful.