How to downgrade python and torch on Colab?

I'm trying to reproduce the study from https://github.com/sanghyun-son/EDSR-PyTorch on colab. To do that, it is necessary to use Python 3.6 and torch 1.2.0

I've successfully downgraded Python to 3.6.

!python --version
# first install python 3.6
!sudo apt-get update -y
!sudo apt-get install python3.6
# change alternatives
!sudo update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.6 1
# select python version
!sudo update-alternatives --config python3
# check python version
!python --version
# install pip for new python
!sudo apt-get install python3.6-distutils
!wget https://bootstrap.pypa.io/get-pip.py
!python get-pip.py
# upgrade pip
!sudo apt install python3-pip
!python -m pip install --upgrade pip
!python --version
!pip --version

I've uninstall pytorch and download and install the 1.2 torch version

!wget https://download.pytorch.org/whl/cpu/torch-1.2.0%2Bcpu-cp36-cp36m-manylinux1_x86_64.whl
!pip uninstall torch -y
!pip install torch-1.2.0+cpu-cp36-cp36m-manylinux1_x86_64.whl

However, when I try to import torch, the "No module named 'torch'" error occurred

import torch
print(torch.__version__)

Any idea why this is happening?

TLDR: it is not possible, but there is a workaround.

It is not possible to downgrade/upgrade the colab Python:
Even though you installed a different python on the VM, the process running this colab is still handled by the original Python. Restarting the backend will make the VM to use the new python, but it will miss all the dependencies necessary to run the colab backend. You might have some luck installing them by hands, but it's rather impractical and is likely to have lasting side effects on the runtime.

Workaround: Colab can connect your local runtime. Create a virtual environment, run a local jupyter backend from it, connect it with colab. This way, you have full control over the environment plus all the perks of a Colab (except free GPU/TPU).