Skip to content

Latest commit

 

History

History
54 lines (44 loc) · 2.21 KB

RELEASE.md

File metadata and controls

54 lines (44 loc) · 2.21 KB
Raw

Current version (0.5.0-dev)

  • Under development.

Release 0.4.0

  • hub.KerasLayer (for Hub/TF2 modules) can be used in graph mode and can be converted to/from a Keras config.
  • In TF1.x, hub.Module can be used inside a defun (helps with TPU).
  • References to TensorFlow Estimator correctly pick up its v1 API.

Release 0.3.0

  • Initial support for Tensorflow 2.0
  • Tensorflow Hub Library API for Tensorflow 2.0
  • Tensorflow Hub Keras API for Tensorflow 2.0
  • Enable using Tensorflow Hub library against Tensorflow 2.0 run-time.

Release 0.2.0

  • Add support for caching modules on GCS.
  • Add ModuleSpec.export(export_path, checkpoint_path) helper for module creation from a separately trained checkpoint.
  • Modules now have a key-value store of attached protocol messages for domain- or library-specific auxiliary data.
  • Image modules can have input shapes with variable height and width; hub.get_expected_image_size() comes from an attached ImageModuleInfo message.
  • Add with hub.eval_function_for_module("...") as f: out = f(in).
  • New experimental function hub.create_module_spec_from_saved_model().
  • Added property hub.Module.variables.
  • Bazel workspace dependency on protobuf library updated to version 3.6.0.
  • Added progress bar for module downloads when interactive.
  • Numerous minor fixes and updates from the master branch.

Release 0.1.1

  • Removed TensorFlow version checking.

Release 0.1.0

  • Initial TensorFlow Hub release.

Requirements

  • tf-nightly>=1.7.0.dev20180308 || tensorflow>=1.7.0rc0