This section contains additional collections of API reference pages for
projects and packages separate from the tensorflow
package, but that
do not have dedicated subsite pages.
The TensorFlow Models repository
The TensorFlow Models repository provides implementations of state-of-the-art (SOTA) models.
The official/projects directory contains a collection of SOTA models that use TensorFlow’s high-level API. They are intended to be well-maintained, tested, and kept up-to-date with the latest TensorFlow API.
The library code used to build and train these models is available as a pip package. You can install it using:
$ pip install tensorflow-models-official # For the latest release
$ #or
$ pip install tf-models-nightly # For the nightly build
To install the package from source, refer to these instructions.
The tensorflow-models-official
pip package contains two top-level modules:
tensorflow_models
and orbit
. You can import them with:
import tensorflow_models as tfm
import orbit
Tensorflow Models
The tensorflow_models
module handles building models and configuring training.
Application-specific functionality is available under tfm.vision
and tfm.nlp
.
Orbit
The orbit
module defines a flexible and lightweight library for writing
customized training loop code in TensorFlow. Orbit is flexible about the type of
models it works with. You can use Orbit to train Keras
Models (as an alternative to Keras' Model.fit
), but you don't have to use
Keras at all. Orbit integrates seamlessly with tf.distribute
and
supports running on different device types (CPU, GPU, and TPU).
TensorFlow Compression
The TensorFlow Compression repository implements learnable compression algorithms that can be used to efficiently compress your data or models.
On Linux and Mac OS, the package can be installed with pip:
pip install tensorflow_compression
To install from source refer to these instructions.