TensorFlow Model Optimization 0.4.0

TFMOT 0.4.0 is the last release to support Python 2. Python 2 support officially ended on January 1, 2020 and TF 2.1.0 was the last release to support Python 2.

Keras clustering API:

• New API for weight clustering
• Major Features
  • Support for clustering convolutional (except DepthwiseConv), Dense and other commonly used standard Keras layers
  • Support for different initialization methods for the cluster centroids: density-based, linear, random
  • Fine-tuning of cluster centroids during training
• Tested against TensorFlow 1.14.0, 2.0.0, and nightly, and Python 3.

Keras quantization API:

• Bug Fixes and Other Changes
  • Fixed Sequential model support for BatchNorm layers that follow Conv/DepthwiseConv (issue).
  • Improved error message for not using quantize_scope with custom Keras layers and objects.
• Tested against TensorFlow nightly, and Python 2/3.

Keras pruning API:

• Tested against TensorFlow 1.14.0, 2.0.0, and nightly, and Python2/3.

TensorFlow Model Optimization 0.3.0

This is the final release of TensorFlow Model Optimization 0.3.0.

Keras quantization API:

• This includes the initial release of the Keras quantization API.
• Tested against TensorFlow nightly

Keras pruning API:

• Features:
  • Added support for TensorFlowOpLayer
• Bugs Fixes:
  • Fixed edge case of propagating training=True to BatchNormalization layer when the arg is passed to call
    function instead of relying on tf.keras.backend.learning_phase().
  • Removed usage of _log_metrics API, which is no longer available in TF 2.2.0+.
• Tested against TensorFlow nightly, 1.14.0 and 2.0.0.

TensorFlow Model Optimization 0.2.1.

This is the final release of TensorFlow Model Optimization 0.2.1.

This includes a bug fix with pruning summaries when using TF 1.X.

It is tested against TensorFlow 1.14.0 and 2.0.0.

TensorFlow Model Optimization 0.2.0

This is the final release of TensorFlow Model Optimization 0.2.0.

This includes:

• Pruning TF 2.X support (#172)
• Usability improvements in code and documentation

It is tested against TensorFlow 1.14.0 and 2.0.0.

TensorFlow Model Optimization 0.1.3

This is the final release of TensorFlow Model Optimization 0.1.3.

This includes the necessary changes for the package to run fine against tf-nightly, after it switches to 2.XX by default.

It is tested against TensorFlow 1.14.0, tf-nightly, and tf-nightly-2.0-preview (using tf.compat.v1).

TensorFlow Model Optimization 0.1.2

This is the final release of TensorFlow Model Optimization 0.1.2.

This includes the necessary changes for TensorFlow Federated to start depending on TensorFlow Model Optimization.

It is tested against TensorFlow 1.14.0.

TensorFlow Model Optimization 0.1.1

This is the final release of TensorFlow Model Optimization 0.1.1.

This includes a fix for Python 3 compatibility.

It is tested against TensorFlow nightly.

TensorFlow Model Optimization 0.1.1-rc0

This is the RC0 release candidate of TensorFlow Model Optimization 0.1.1.

It is tested against TensorFlow nightly.

TensorFlow Model Optimization 0.1.0

This is the final release of TensorFlow Model Optimization 0.1.0.

This includes the first release of the model pruning library.

It is tested against TensorFlow nightly.

TensorFlow Model Optimization 0.1.0-rc1

This is the RC1 release candidate of TensorFlow Model Optimization 0.1.0

It is tested against TensorFlow nightly.