Release 0.4

0.4.3

Fix the requirement deps issues.

0.4.2

Fix the version of tflite-support and flatbuffers to be consistent with other packages.

0.4.1

Fix the version of scann to 1.2.6 to solve the import failure.

0.4.0

Add one new task: Searcher that support Image / Text as the inputs.

Release 0.3

0.3.4

Minor fix: remove tensorflow-estimator version requirements to avoid the incompatible version error message during pip installation.

0.3.3

Minor fix: update tensorflow==2.6 and fix the failure of evaluate_tflite for the object detection task.

0.3.2

Minor fix: Move librosa to an optional library for audio task.

0.3.1

Polish document showing in https://www.tensorflow.org/lite/api_docs/python/tflite_model_maker.
Add some 0.2.x APIs for backward compatibility.

0.3.0

Export TFLite Model Maker public APIs, and adapt colabs as demo code.
Add three new tasks: Object Detection, Audio Classification and Recommendation.

Release 0.2

0.2.5

Refactor the code to be more flexible and extentable without changing API.
Add executable test script for pip package.
Update tensorflow-hub version in requirements.txt to fix incompatible error when installing pip package.

0.2.4

Update stable tensorflow version in requirements.txt.
Support exporting to TFJS models.
Refine Code: refactor the gen_dataset internal API in DataLoader and remove redundant methods.
Split requirements to stable version and nightly version.

0.2.3

Fix a bug for missing default_batch_size.
Add a demo of customized task.
Refine Code: Refactor model_spec to split different model specifications into image/text and add ClassificationDataLoader.

0.2.2

Update tf-nightly version in requirements.txt.
Refine Code: Change pre-defined model specification to function and extract Keras callback logic into a reusable function.

0.2.1

Set the required package version in requirements.txt.
Refine Code: Move load_from_tfds into ImageDataLoader, refactor the DataLoader code, unify the predict method and provide a default implementation for TFLite expo.

0.2.0

Setup virtualenv and test pip package.

Release 0.1

Initial version for TFLite Model Maker: a model customization library for on-device applications.

Support tasks:

0.1.2

Add metadata_writer for nlp tasks and set with_metadata=True by default when exporting to TFLite.
Update model_info in metadata for image classifier.

0.1.1

Update README.
Add mobilebert_qa_squad model_spec, a mobilebert model pre-trained with Squad1.1 dataset.

0.1.0

Initial version.