If you can repeat and copy both the input and output it would give a better idea of what’s going on.
I have also just built myself a new environment and created a batch file for it. It should be run from a conda prompt and I’m not sure whether atari_py & gym are still required (but they don’t block the installation)
You can also copy and paste individual commands… There are some things that are just for me in the final section, but I left them in so you can see they are compatible. There was an issue with one dependency that needed to be downgraded at some point, but I’ve forgotten which - I’m sure it’ll be fairly obvious if it happens to you.
Also: if you don’t have an nVidia GPU you don’t need the cudatoolkit, cudnn installations.
REM conda options -y = yes, i.e. no confirmation; -q = quiet, i.e. no progress bar
REM pip options -q = quiet
REM This environment created 20/07/2021
REM minimal set of conda installations especially cuda support for TensorFlow
REM conda -c and the channel name must be adjacent
echo Conda stuff - create and activate environment %envName% with cuda support for TF2.5
call conda create -y -q --name %envName% -c conda-forge python==3.8.10
call conda activate %envName%
call conda install -y -q -c conda-forge cudatoolkit=11.2.2
call conda install -y -q -c conda-forge cudnn=126.96.36.199
REM pip installations. NB tensorflow_decision_forests is linux only 30/06/2021 so can't install yet
echo Conda installations done. Proceeding with pip installations: pip, setuptools, TensorFlow, Perceptilabs, seaborn, etc.
pip install -q --upgrade pip setuptools
pip install -q --no-input tensorflow-gpu==2.5.0
echo Pip install Perceptilabs and atari-py dependencies
pip install perceptilabs
pip install -f https://github.com/Kojoley/atari-py/releases atari_py
pip install gym[atari]
REM Base PL installation done. Additional packages
pip install -q jupyterlab
pip install -q --no-input seaborn tabulate pandoc
pip install -q ipyfilechooser tqdm
Hope that helps