How to run TensorFlow GPU version on Google Colab with Python 2.7?

Question

I have a code written using the TensorFlow version 1.15 for an image to image translation task and I want to run it on Google Colab environment which currently has Python 3.10.12 installed by default.

Due to my source code's dependencies, I have to use TensorFlow version 1.15 and that's why I have used other dependencies that match with this version of TensorFlow.

First, I install Python version 2.7 on Google Colab using the command below:

!apt-get install python2

After that, I check the installed Python's version:

!python2 --version

And I get: Python 2.7.18

Next, I connect to Google Drive and refer to the path where my source is located (main.py, util.py, model.py and ops.py):

from google.colab import drive
drive.mount('/content/drive')
%cd /content/drive/MyDrive/TFCode/

Then to install the necessary packages, first I install pip using the commands below:

!curl https://bootstrap.pypa.io/pip/2.7/get-pip.py -o get-pip.py
!python2 get-pip.py

And then:

!python2 -m pip install six
!python2 -m pip install scipy
!python2 -m pip install imageio==2.4.1
!apt-get install build-essential
!python2 -m pip install grpcio==1.26.0
!python2 -m pip install tensorflow-gpu==1.15
!python2 -m pip install opencv-python==3.4.8.29
!python2 -m pip install scikit-image==0.14.5

Finally, I run my source using !python2 main.py and despite the fact that the code runs successfully, it takes a long time to finish the task becuase my code is not being run on GPU and I know this because of the following code block inside my main.py file:

if tf.test.gpu_device_name():
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
else:
    print("Please install GPU version of TF")

Inside my results, I see the "Please install GPU version of TF" string output which means the GPU cannot be detected but when I run this code like below directly in Colab:

import tensorflow as tf
if tf.test.gpu_device_name():
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
else:
    print("Please install GPU version of TF")

I get this: Default GPU Device: /device:GPU:0 which means it finds the GPU but as this code is being run directly in Colab environment, I know that it uses Python 3.10.12 and TensorFlow version 2.

How can I run my own source code with GPU?

Using version 2.7 of Python is not a must but version 1.15 of TensorFlow must be used and other dependencies must match with this version of TensorFlow.

Answer 1

Note: there is a TensorFlow GPU 1.15 is compatible with Python 3.9-3.11. TensorFlow 1.15 (Jan. 2020) is no longer listed.

Using version 2.7 of Python is not a must

Since TensorFlow 1.15 is compatible with Python 3.7, you try and create a virtual environment with Python 3.7 instead, avoiding any compatibility issue with Python 2:

!sudo apt-get install python3.7
!python3.7 -m venv tf15env

Activate the virtual environment:

!source /content/tf15env/bin/activate

And install TensorFlow 1.15 and other dependencies within the virtual environment:

!pip install tensorflow-gpu==1.15
!pip install six scipy imageio==2.4.1 grpcio==1.26.0 opencv-python==3.4.8.29 scikit-image==0.14.5

Run this script to make sure TensorFlow is using the GPU:

import tensorflow as tf
if tf.test.gpu_device_name():
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
else:
    print("Please install GPU version of TF")

After setting up the environment, you can run your code (main.py, util.py, model.py, ops.py) in this virtual environment.

---

See also "Changing the Python version on Google Colab" from Siladittya Manna for an alternative way to change the Python version on Google Colab.

Answer 2

I followed the guide "Changing the Python version on Google Colab" from Siladittya Manna stated by VonC and everything seems to work. I used Colab with an T4 GPU instance.

!apt-get update -y
!apt-get install python3.7 python3.7-distutils
!update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.7 1
!update-alternatives --config python3
!apt-get install python3-pip
!python3 -m pip install --upgrade pip --user

!python3 --version # Python 3.7.17

With the install instruction already mentioned by VonC:

!pip install tensorflow-gpu==1.15
!pip install six scipy imageio==2.4.1 grpcio==1.26.0 opencv-python==3.4.8.29 scikit-image==0.14.5

Including your version dependency for protobuf:

!pip install protobuf==3.20.0

Running

import tensorflow as tf
if tf.test.gpu_device_name():
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
else:
    print("Please install GPU version of TF")

I get Default GPU Device: /device:GPU:0.