Getting started with Keras

Learning resources

Are you a machine learning engineer looking for a Keras introduction one-pager? Read our guide Introduction to Keras for engineers.

Want to learn more about Keras 3 and its capabilities? See the Keras 3 launch announcement.

Are you looking for detailed guides covering in-depth usage of different parts of the Keras API? Read our Keras developer guides.

Are you looking for tutorials showing Keras in action across a wide range of use cases? See the Keras code examples: over 150 well-explained notebooks demonstrating Keras best practices in computer vision, natural language processing, and generative AI.

Installing Keras 3

You can install Keras from PyPI via:

pip install --upgrade keras

You can check your local Keras version number via:

import keras
print(keras.__version__)

To use Keras 3, you will also need to install a backend framework – either JAX, TensorFlow, or PyTorch:

• Installing JAX
• Installing TensorFlow
• Installing PyTorch

If you install TensorFlow 2.15, you should reinstall Keras 3 afterwards. The cause is that tensorflow==2.15 will overwrite your Keras installation with keras==2.15. This step is not necessary for TensorFlow versions 2.16 onwards as starting in TensorFlow 2.16, it will install Keras 3 by default.

Installing KerasCV and KerasHub

KerasCV and KerasHub can be installed via pip:

pip install --upgrade keras-cv
pip install --upgrade keras-hub
pip install --upgrade keras

Configuring your backend

You can export the environment variable KERAS_BACKEND or you can edit your local config file at ~/.keras/keras.json to configure your backend. Available backend options are: "jax", "tensorflow", "torch". Example:

export KERAS_BACKEND="jax"

In Colab, you can do:

import os
os.environ["KERAS_BACKEND"] = "jax"
import keras

Note: The backend must be configured before importing Keras, and the backend cannot be changed after the package has been imported.

GPU dependencies

Colab or Kaggle

If you are running on Colab or Kaggle, the GPU should already be configured, with the correct CUDA version. Installing a newer version of CUDA on Colab or Kaggle is typically not possible. Even though pip installers exist, they rely on a pre-installed NVIDIA driver and there is no way to update the driver on Colab or Kaggle.

Universal GPU environment

If you want to attempt to create a "universal environment" where any backend can use the GPU, we recommend following the dependency versions used by Colab (which seeks to solve this exact problem). You can install the CUDA driver from here, then pip install backends by following their respective CUDA installation instructions: Installing JAX, Installing TensorFlow, Installing PyTorch

Most stable GPU environment

This setup is recommended if you are a Keras contributor and are running Keras tests. It installs all backends but only gives GPU access to one backend at a time, avoiding potentially conflicting dependency requirements between backends. You can use the following backend-specific requirements files:

• requirements-jax-cuda.txt
• requirements-tensorflow-cuda.txt
• requirements-torch-cuda.txt

These install all CUDA-enabled dependencies via pip. They expect a NVIDIA driver to be preinstalled. We recommend a clean python environment for each backend to avoid CUDA version mismatches. As an example, here is how to create a JAX GPU environment with Conda:

conda create -y -n keras-jax python=3.10
conda activate keras-jax
pip install -r requirements-jax-cuda.txt
pip install --upgrade keras

TensorFlow + Keras 2 backwards compatibility

From TensorFlow 2.0 to TensorFlow 2.15 (included), doing pip install tensorflow will also install the corresponding version of Keras 2 – for instance, pip install tensorflow==2.14.0 will install keras==2.14.0. That version of Keras is then available via both import keras and from tensorflow import keras (the tf.keras namespace).

Starting with TensorFlow 2.16, doing pip install tensorflow will install Keras 3. When you have TensorFlow >= 2.16 and Keras 3, then by default from tensorflow import keras (tf.keras) will be Keras 3.

Meanwhile, the legacy Keras 2 package is still being released regularly and is available on PyPI as tf_keras (or equivalently tf-keras – note that - and _ are equivalent in PyPI package names). To use it, you can install it via pip install tf_keras then import it via import tf_keras as keras.

Should you want tf.keras to stay on Keras 2 after upgrading to TensorFlow 2.16+, you can configure your TensorFlow installation so that tf.keras points to tf_keras. To achieve this:

1. Make sure to install tf_keras. Note that TensorFlow does not install it by default.
2. Export the environment variable TF_USE_LEGACY_KERAS=1.

There are several ways to export the environment variable:

1. You can simply run the shell command export TF_USE_LEGACY_KERAS=1 before launching the Python interpreter.
2. You can add export TF_USE_LEGACY_KERAS=1 to your .bashrc file. That way the variable will still be exported when you restart your shell.
3. You can start your Python script with:

import os
os.environ["TF_USE_LEGACY_KERAS"] = "1"

These lines would need to be before any import tensorflow statement.

Compatibility matrix

JAX compatibility

The following Keras + JAX versions are compatible with each other:

• jax==0.4.20 & keras~=3.0

TensorFlow compatibility

The following Keras + TensorFlow versions are compatible with each other:

To use Keras 2:

• tensorflow~=2.13.0 & keras~=2.13.0
• tensorflow~=2.14.0 & keras~=2.14.0
• tensorflow~=2.15.0 & keras~=2.15.0

To use Keras 3:

• tensorflow~=2.16.1 & keras~=3.0

PyTorch compatibility

The following Keras + PyTorch versions are compatible with each other:

• torch~=2.1.0 & keras~=3.0