Release 1.11.1

Use accurate JPEG decoding.

New Features

N/A

Bug Fixes

1. "INTEGER_ACCURATE" JPEG decoding is used to make output PIL equivalent.

Breaking Changes

N/A

Release 1.11.0

Adds support for TensorFlow 2.16.

New Features

N/A

Breaking Changes

N/A

Release 1.10.0

Added support for Tensorflow 2.15 and Python 3.11

New Features

N/A

Bug fixes

1. Python 3.11 supports

Breaking Changes

Release 1.9.0

Added support for Tensorflow 2.11

New Features

N/A

Bug fixes

N/A

Breaking Changes

TensorFlow version < 2.11 not supported

Release 1.8.0

New features

1. Support more than one JSON file for a single dataset tfrecord conversion of text data.
2. Support text data as part of the labels columns family. This enhances the support for Question/Answer relates tasks and GLUE benchmarks.

Breaking changes

N/A

Bug fixes

N/A

Release 1.7.0

New features

1. Disabled default padding of elements within a batch.
   To enable padding of elements using datum interpreted shape use dataset_configs.use_dautm_padding=True.

Breaking Changes

N/A

Bug Fixes

N/A

Release 1.6.0

Datum now supports Tensorflow 2.9.1. This release updates the TensorFlow version.

Release 1.5.0

Datum 1.5.0 support tensorflow==2.8.1

Release 1.4.0

Set drop_remainder, num_parallel_calls and deterministic via configs. num_parallel_calls only available when using deterministic.

Release 1.3.0

Now datum supports tensorflow=2.7.0