TensorFlow from anaconda channel downgrades gpu to mkl version

Product Help
1

I have created an environment and installed TensorFlow with the command:

conda install -c anaconda tensorflow=*=gpu_py310h5ade2b3_0 cudatoolkit cudnn

And it installs the latest TensorFlow version available on the anaconda channel.
The output of conda info below:


     active environment : tensorflow
    active env location : C:\ProgramData\scoop\apps\miniconda3\current\envs\tensorflow
            shell level : 1
       user config file : C:\Users\Otávio Augusto Silva\.condarc
 populated config files : C:\Users\Otávio Augusto Silva\.condarc
          conda version : 22.9.0
    conda-build version : not installed
         python version : 3.9.13.final.0
       virtual packages : __cuda=11.7=0
                          __win=0=0
                          __archspec=1=x86_64
       base environment : C:\ProgramData\scoop\apps\miniconda3\current  (writable)
      conda av data dir : C:\ProgramData\scoop\apps\miniconda3\current\etc\conda
  conda av metadata url : None
           channel URLs : https://repo.anaconda.com/pkgs/main/win-64
                          https://repo.anaconda.com/pkgs/main/noarch
                          https://repo.anaconda.com/pkgs/r/win-64
                          https://repo.anaconda.com/pkgs/r/noarch
                          https://repo.anaconda.com/pkgs/msys2/win-64
                          https://repo.anaconda.com/pkgs/msys2/noarch
          package cache : C:\ProgramData\scoop\apps\miniconda3\current\pkgs
                          C:\Users\Otávio Augusto Silva\.conda\pkgs
                          C:\Users\Otávio Augusto Silva\AppData\Local\conda\conda\pkgs
       envs directories : C:\ProgramData\scoop\apps\miniconda3\current\envs
                          C:\Users\Otávio Augusto Silva\.conda\envs
                          C:\Users\Otávio Augusto Silva\AppData\Local\conda\conda\envs
               platform : win-64
             user-agent : conda/22.9.0 requests/2.28.1 CPython/3.9.13 Windows/10 Windows/10.0.22000
          administrator : True
             netrc file : None
           offline mode : False

But when I try to update everything by using the command conda update --all the following happens:

The following packages will be DOWNGRADED:

  tensorflow                      2.9.1-gpu_py310h5ade2b3_0 --> 2.9.1-mkl_py310h0b323c9_0 None
  tensorflow-base                 2.9.1-gpu_py310h9761872_0 --> 2.9.1-mkl_py310h6a7f48e_0 None

The question is, how do I prevent conda update from downgrading TensorFlow and keep the GPU version?

3

dear tavioasilva.

Thank you for your contribution to the anaconda community.

The following is written in both Lithuanian and English.

＞ The question is, how do I prevent conda update from downgrading TensorFlow and keep the GPU version?

English:
The solution to this problem is simple.

netrc file : None
offline mode : False

You can change this part.

Lietuviškai:
Šios problemos sprendimas yra paprastas.

netrc file : None
offline mode : False

Galite pakeisti šią dalį.

Best regards.

4

I’m sorry, but I don’t understand what exactly you mean by that. Could you elaborate further?

5

dear otavioasilva.

thank you for your reply.

It seems that you live in the Portuguese zone, so the following is explained in Portuguese.

netrc : False

Quando isso é exibido nas configurações do Tensorflow, o Python que baixa o Tensorflow e o curl que é a base dele tentará fazer o download com uma configuração chamada Anonymous.

Contudo,

conda install -c anaconda tensorflow-mkl

tensorflow-mkl, instalado com o comando, requer um login anaconda (mesmo para usuários profissionais).

Portanto, consulte o site a seguir e defina .netrc ou _netrc para que você possa efetuar login no servidor de distribuição anacoda corretamente.

Atenciosamente.

6

URL：.netrc - Everything curl

7

Hello, sorry for the delay. The problem is not the language here, I just don’t understand how that can help me. I found out a workaround by doing conda update --all tensorflow==2.9.1=gpu_py310* and it’s been working so far. I just need the Anaconda devs to be aware of this and find a more permanent solution, since the tensorflow-gpu package is very outdated.