Set Up TensorFlow with Docker and Jupyter Notebook

Before the first line of code, you need an environment that won’t fight you.

If you’ve ever tried setting up a deep learning environment locally, you know how quickly things can break.

This TensorFlow Docker setup guide provides a complete TensorFlow Jupyter Docker environment that solves dependency conflicts and creates a reproducible machine learning environment. Whether you’re dealing with CUDA compatibility issues or Python version mismatches, this Docker TensorFlow environment setup eliminates those headaches.

Before we start coding, we need a place to work that’s clean, predictable, and easy to rebuild. Machine learning libraries depend on specific versions of Python and GPU drivers, and setting them up manually can quickly turn into a frustrating loop of errors. Believe me, I learned this the hard way.

To avoid that, we’ll run everything inside a contained environment, one that isolates our setup from the rest of the system and gives us a fresh start every time.

What We’ll Use and Why

By combining these tools, you’ll create a reproducible machine learning environment that works identically across all systems—eliminating the classic ‘it works on my machine’ problem that plagues traditional TensorFlow installations.

Docker

Docker is a container system that lets you run isolated environments. It guarantees that everyone following my AI tutorials, including you a year from now, will have the same working setup. No conflicting dependencies, no reinstalling Python packages, and no “works on my machine” moments.

Docker’s reproducibility becomes even more valuable in modern development workflows where AI-assisted coding and vibe coding practices are becoming standard—you need environments that work consistently regardless of who (or what) generated the code.

Jupyter Notebook

Jupyter is an interactive coding environment that runs in your browser. It lets you mix text, code, and visualizations in one place, making it ideal for tutorials and experiments.

TensorFlow

TensorFlow is the deep learning framework used to build our neural networks. It’s widely supported, integrates smoothly with Docker, and includes high-level APIs like Keras that make experimentation simple. There’s another excellent option called PyTorch, which you’ll often see in research settings. PyTorch is known for flexibility and debugging convenience, while TensorFlow shines when it comes to deployment, performance, and ecosystem tooling. For a beginner-friendly introduction to machine learning, TensorFlow is the more stable and forgiving choice.

TensorFlow Docker Setup: Installing and Pulling the Image

If you don’t have Docker yet, download Docker Desktop for Windows from docker.com and follow the installer’s steps. If you are on macOS or Linux, download the respective installer. At the moment of writing, I’m using version 4.49.0 on macOS.

Once it’s running, open it. You’ll see on the left tabs for Containers, Images, Volumes, etc. We’ll use the Images tab first. In the middle, there’s the Search images to run button.

Click on it to open the search field and type in it:

tensorflow/tensorflow:latest-gpu-jupyter

A dropdown with several options will appear. The first one should be tensorflow/tensorflow. From the Tag dropdown on the right, choose latest-gpu-jupyter and click Pull.

This process downloads the official TensorFlow image that includes everything we need: Python, TensorFlow, CUDA, and Jupyter Notebook. It’s maintained by the TensorFlow team, so you can rely on it being compatible and up to date. The download might take a few minutes the first time.

TensorFlow Container Configuration: Running Your Environment

When the image finishes downloading, it’ll appear in your list of available images. Click Run to start a new container. A configuration window will open. Expand the Optional settings and fill in the following fields:

In the Ports field, enter 8888. That’s the port Jupyter uses, and it’s how we’ll access it from the browser. Next, scroll to Volumes. This is where we connect a folder on your computer to the container so that anything you save inside Jupyter persists outside Docker too.

On Windows, I usually use a path like C:Users<your-name>Python-Projects. Set that as the Host path, and enter /tf/notebooks as the Container path. That’s where Jupyter stores notebooks by default.

Click Run. Docker starts the container and automatically launches Jupyter Notebook inside it.

Creating a Container for macOS Users

Note for macOS users on Apple Silicon processors: The setup above won’t work due to incompatibilities with hardware. After installing and opening Docker Desktop, via Terminal, run this command:

docker run --rm -it -p 8888:8888 --shm-size=1g -e JUPYTER_TOKEN=dev -v /path/to/your/notebooks:/home/jovyan/work -w /home/jovyan/work quay.io/jupyter/tensorflow-notebook:53b1d144db49

Replace the /path/to/your/notebooks path with where you want your notebooks to be. This image contains Jupyter Lab instead of regular Jupyter. It’s basically the same but the UI is slightly different.

Note: the images can be a few GB in size. Give it enough time to download everything and do not be surprised if it feels slow to setup.

Opening Jupyter Notebook

The setup process should already take you to the container’s logs. If it does not, go to the Containers tab. You’ll see your running container there. Click it, then open the Logs view.

Scroll through the logs and look for a long line starting with: http://127.0.0.1:8888/tree?token=

That’s your unique Jupyter link. Click on it in your browser. You’ll see Jupyter’s familiar interface appear. Navigate to the /notebooks directory. That’s the folder you connected earlier. Anything you save here will stay on your computer, even after you stop or delete the container.

Security note: This token grants full access to your notebooks. Don’t share this URL publicly or commit it to version control.

Managing Your Container

To stop the container: In Docker Desktop, go to Containers and click the Stop button on your running container.

To restart next time: Click the Run button on the same container. Your notebooks in the mounted folder will still be there.

Pro tip. Your Jupyter token will be different each time you restart. Check the logs for the new URL.

Verifying TensorFlow

Before we move on, it’s worth confirming that everything works as expected. Create a new Python 3 notebook in Jupyter and paste the following snippet in the empty cell:

import sys
import tensorflow as tf

print(sys.version())
print("TensorFlow version:", tf.__version__)
print("Available GPUs:", tf.config.list_physical_devices('GPU'))

With the focus on the cell, click Shift + Enter to run it. This will print on screen the version of Python and TensorFlow used, and the GPUs detected. If a GPU appears in the list, great — TensorFlow is using your NVIDIA hardware. If not, don’t worry; it’s running on the CPU instead.

Note: When you run this code for the first time you might get many warnings in red. You can safely ignore them. Run the same cell again to get rid of them.

Let’s create your first tensor as an additional test. Go to the next cell in your notebook and paste this code:

# Create a simple tensor
hello = tf.constant('Hello, TensorFlow!')
print(hello)

# Do a basic calculation
a = tf.constant(2)
b = tf.constant(3)
print(f"2 + 3 = {a + b}")

If you see the output without errors, congratulations—your environment is fully functional!

It’s Time to Code

You now have a self-contained TensorFlow lab running on your machine. Everything you build inside Jupyter — from notebooks to trained models — stays organized and reproducible. The next time you need to revisit or share your setup, just start the same container, and you’ll be back where you left off.

This environment isn’t just for deep learning experiments. It’s perfect for anyone exploring AI tools and machine learning workflows, whether you’re training CNNs or experimenting with AI-powered features for your projects.

What about you, have you run into Docker issues or TensorFlow setup quirks on your machine? Share your experience in the comments below.