TensorFlow Addons v0.23.0

Release Notes

• Built against TensorFlow 2.15
• CUDA kernels are compiled with CUDA 12.2 and cuDNN 8.6
• Windows build support dropped pending upstream fix
• API docs found on the website

Thanks to our Contributors

@georgiyekkert @bhack and @seanpmorgan

TensorFlow Addons v0.22.0

Release Notes

• Built against TensorFlow 2.14
• CUDA kernels are compiled with CUDA11.8 and cuDNN 8.6
• API docs found on the website

Changelog

• Set permissions to GITHUB_TOKEN (#2848)
• Update compatibility matrix (#2843)
• Increase version to 0.22 (#2842)
• Update feature request template to reflect TFA wind down (#2818)

Thanks to our Contributors

@joycebrum, @seanpmorgan, @bhack, @fsx950223

TensorFlow Addons v0.21.0

TensorFlow Addons v0.21.0

Release Notes

• Built against TensorFlow 2.13
• CUDA kernels are compiled with CUDA11.8 and cuDNN 8.6
• API docs found on the website

Changelog

• Add EOL to README.md (#2836)
• Fix tf.functions with None inputs (#2837)
• Fix Keras imports. (#2829)

Thanks to our Contributors

@faizan-m, @reedwm @bhack and @seanpmorgan

TensorFlow Addons v0.20.0

Release Notes

• Built against TensorFlow 2.12
• CUDA kernels are compiled with CUDA11.8 and cuDNN 8.6
• API docs found on the website

Changelog

• Added End of Life Message for May 2024

tfa.layers

• Update embedding_bag.py for save_model in 'tf' mode (#2804)

tfa.optimizers

• Fix wrong lookahead hints (#2817)

tfa.image

• Fix Cutout memory leak (#2815)

Thanks to our Contributors

@372046933, @AlexanderLavelle, @drauh, @faizan-m, @fsx950223, @seanpmorgan and @xiruizhao @bhack

TensorFlow Addons v0.19.0

Release Notes

• Built against TensorFlow 2.11
• CUDA kernels are compiled with CUDA11.2 and cuDNN 8.1.0
• API docs found on the website

Changelog

• Replace tensorflow::Status::OK() with tensorflow::OkStatus(). (#2765)
• Fix local build with GPU support (#2764)

tfa.metrics

• Fix metrics testing failure due to optimizer change (#2784)

tfa.optimizers

• Fix optimizer failures for tf 2.11 (#2783)
• Get compatible with optimizer migration in TF 2.11 (#2766)

Thanks to our Contributors

@bhack, @chenmoneygithub, @manivaradarajan, @seanpmorgan and @shkarupa-alex

TensorFlow Addons v0.18.0

Release Notes

• Built against TensorFlow 2.10
• CUDA kernels are compiled with CUDA11.2 and cuDNN 8.1.0
• API docs found on the website

Changelog

• Built with C++17 compatibility

tfa.metrics

• Fix the Kendalls Tau metric when used in graph mode (#2739)
• Add Kendall's Tau-c, Spearman's Rank, Pearson's Correlation metrics (#2741)
• Correlation metrics optimization (#2747)

tfa.optimizers

• Make MultiOptimizer serializable (#2719)
• Point optimizer to tf.keras.optimizer.legacy.Optimizer (#2706)
• Use consistent markdown formatting for the AdamW paper (#2722)

Thanks to our Contributors

@JackWindows, @chenmoneygithub, @nicolaspi, @njzjz, @ronshapiro, @bhack, @lgeiger and @seanpmorgan

TensorFlow Addons v0.17.1

Release Notes

• Add Python 3.10 support for Apple Silicon. (#2718)

TensorFlow Addons v0.17.0

Release Notes

• Built against TensorFlow 2.9
• CUDA kernels are compiled with CUDA11.2 and cuDNN 8.1.0
• API docs found on the website

Changelog

• Manylinux2014 compatible
• Add T=bfloat16 to custom_ops registration (#2688)

tfa.metrics

• AUTOTUNE 'y_shape' for r_square (#2702)

tfa.layers

• fix compute_output_shape behavior in normalizations.py (#2678)

tfa.optimizers

• Failing to restore AdaBelief optimizer from checkpoint (#2705)
• Rephrase documentation of adabelief (#2704)
• Fix typo in average optimizers callback (#2692)
• Simplify exclude_from_weight_decay implementation (#2676)

Thanks to our Contributors

@Zhaopudark, @bhack, @denadai2, @fsx950223, @gautam1858, @jondo, @lgeiger, @markub3327, @seanpmorgan, @szutenberg, @trevor-m and @yuetan1988

TensorFlow Addons v0.16.1

Release Notes

• Patch release to use consistent build container

TensorFlow Addons v0.16.0

Release Notes

• Built against TensorFlow 2.8
• CUDA kernels are compiled with CUDA11.2 and cuDNN 8.1.0
• API docs found on the website

Changelog

• Add support for publishing macOS M1 ARM64 wheels for tfa-nightly (#2559)
• Utilize SIG Build Docker Images (#2515)
• Pin mac build environment's OS version (#2627)

Tutorials

tfa.activations

• Update hardshrink.py (#2649)

tfa.iamge

• Generate Assertion Ops for interpolate_bilinear (#2609)

tfa.layers

• Add MaxUnpooling2DV2 layer (#2594)
• Fix layers_normalizations.ipynb typos (#2606)

tfa.optimizers

• Save exclude_from_weight_decay in config for LAMB (#2619)
• Fix typo in average optimizers callback (#2621)
• In LAMB optimizer, rename weight_decay_rate to weight_decay (#2611)
• exclude_from_weight_decay for AdamW and SGDW (#2624)

tfa.text

• Fix kwargs in crf_decode_forward (#2642)

Thanks to our Contributors

@abattery, @bhack, @faizan-m, @gabrieldemarmiesse, @inyong37, @jvishnuvardhan, @ksachdeva, @leondgarse, @midsterx, @qlzh727, @reedwm, @seanpmorgan, @shkarupa-alex and @tgsmith61591