How to Upgrade or Downgrade Python in Conda on nvidia/cuda Docker?

If you need to run a specific version of Python, you can Upgrade or Downgrade Python Conda on nvidia/cuda Docker as shown in this tutorial.

1. Attach a terminal to the running docker.

fyicenter# docker exec -it my_cuda /bin/bash 

(base) root@e4395c83ac54:/# 

2. Install "pip" in the default Conda environment "base".

(base) root@e4395c83ac54:/# conda install -y -c conda-forge pip 

  Downloading and Extracting Packages
  certifi-2021.5.30    | 141 KB    |
  python_abi-3.9       | 4 KB      |
  openssl-1.1.1k       | 2.1 MB    |
  wheel-0.37.0         | 31 KB     |
  conda-4.10.3         | 3.1 MB    |
  pip-21.2.4           | 1.1 MB    |
  ca-certificates-2021 | 136 KB    |

3. Install Python 3.7.

(base) root@e4395c83ac54:/# conda install -y -c conda-forge python=3.7

Downloading and Extracting Packages
conda-package-handli | 921 KB    | 
pycosat-0.6.3        | 107 KB    | 
ruamel_yaml-0.15.80  | 270 KB    | 
chardet-4.0.0        | 204 KB    | 
cffi-1.14.6          | 225 KB    | 
setuptools-49.6.0    | 947 KB    | 
conda-4.10.3         | 3.1 MB    | 
cryptography-3.4.7   | 1.1 MB    | 
brotlipy-0.7.0       | 341 KB    | 
certifi-2021.5.30    | 141 KB    | 
python_abi-3.7       | 4 KB      | 
pysocks-1.7.1        | 27 KB     | 
python-3.7.10        | 57.3 MB   |

4. Verify Python version inside Conda.

(base) root@e4395c83ac54:/# python --version 

Python 3.7.10

5. Verify Python version outside Conda.

(base) root@e4395c83ac54:/# conda deactivate 

root@e4395c83ac54:/# python --version 
  bash: python: command not found

6. Create a new Conda environment with Python 3.9.

root@e4395c83ac54:/# conda create -n py3.9 
  environment location: /opt/conda/envs/py3.9

root@e4395c83ac54:/# conda activate py3.9
(py3.9) root@e4395c83ac54:/# 

7. Upgrade the new Conda environment to Python 3.9.

(py3.9) root@e4395c83ac54:/# conda install -y -c conda-forge python=3.9

Downloading and Extracting Packages
readline-8.1         | 295 KB    |
_libgcc_mutex-0.1    | 3 KB      |
sqlite-3.36.0        | 1.4 MB    |
libgcc-ng-11.1.0     | 908 KB    |
tk-8.6.10            | 3.2 MB    |
ncurses-6.2          | 985 KB    |
python-3.9.6         | 27.5 MB   |
xz-5.2.5             | 343 KB    |
zlib-1.2.11          | 106 KB    |
tzdata-2021a         | 121 KB    |
_openmp_mutex-4.5    | 22 KB     |
ld_impl_linux-64-2.3 | 667 KB    |
libffi-3.3           | 51 KB     |
setuptools-49.6.0    | 943 KB    |
libgomp-11.1.0       | 427 KB    |
libstdcxx-ng-11.1.0  | 4.2 MB    |

8. Verify Python version in both Conda environmwnts:

(py3.9) root@e4395c83ac54:/# python --version 
  Python 3.9.6

(py3.9) root@e4395c83ac54:/# conda activate base 

(base) root@e4395c83ac54:/# python --version
  Python 3.7.10

⇒ Install CUDA Toolkit in Conda on nvidia/cuda Docker

⇐ Install Miniconda3 on nvidia/cuda Docker

⇑ NVIDIA GPU Server Docker Image

⇑⇑ Docker Container Platform - Tutorials