Skip to content

TensorFlow Addons v0.12.0
Compare
Choose a tag to compare
@github-actions github-actions released this 23 Dec 19:21
· 6 commits to r0.12 since this release
v0.12.0
d26e2ed
This commit was created on github.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

Release Notes

Changelog

  • Add AVX2 support (#2299)
  • Drop TF2.2 compatibility (#2224)
  • Drop python3.5 support (#2204)
  • Expose tfa.types doc (#2162)
  • Rename "Arguments:" to "Args:" (#2267)
  • Add support for ARM architecture build from source (#2182)

tfa.activations

  • Add tf.nn.gelu alias for TF >= 2.4 (#2265)
  • Remove custom op activations (#2247)

tfa.image

  • Speedup gaussian kernel generation (#2149)
  • Support fill_mode for transform (#2153)
  • Use ImageProjectiveTransformV3 for TF >= 2.4.0 (#2293)
  • Support unknown rank image (#2300)
  • Fix sparse_image_warp partially unknown shape (#2308)
  • Make cutout compatible with keras layer (#2302)
  • Remove unsupported data_format (#2296)
  • Refactor sharpness (#2287)
  • Rodert fix image random cutout 2276 (#2285)
  • Remove tf.function decorator in tfa.image.equalize (#2264)
  • Support empty batches in ResamplerOp (#2219)
  • Make cutout op compatible with non eager mode (#2190)

tfa.layers

  • Add stochastic depth layer (#2154)

  • Add MaxUnpooling2D layer (#2272)

  • Add noisy dense layers. (#2099)

  • Add discriminative Layer Training (#969)

  • Make MultiHeadAttention agnostic to dtype (float32 vs. float16) (#2253)

  • Change CRF layer dtype (#2270)

  • Change GroupNormalization default groups to 32. (#2241)

tfa.optimizers

  • Standardized Testing Module (#2233)
  • Fix LazyAdam resource variable ops performance issue (#2274)
  • Add experimental_aggregate_gradients support (#2137)

tfa.rnn

  • Bug fix for conflict variable name in layernorm cells. (#2284)

tfa.seq2seq

  • Graduate _BaseAttentionMechanism to a public base class (#2209)
  • Add a doctest example for BasicDecoder (#2214)
  • Add a doctest example for AttentionWrapper (#2215)
  • Improve sampler documentation, use doctest (#2213)
  • Beam search decoding procedure added to seq2seq_nmt tutorial (#2140)

Thanks to our Contributors

@DanBmh, @DavidWAbrahams, @Harsh188, @JulianRodert, @LeonShams, @MHStadler, @MarkDaoust, @SamuelMarks, @WindQAQ, @aaronmondal, @abhishek-niranjan, @albertz, @bhack, @crccw, @edend10, @fsx950223, @gabrieldemarmiesse, @guillaumekln, @HMPH, @hp77-creator, @hwaxxer, @hyang0129, @kaixih, @lamberta, @marksandler2, @matwilso, @napsternxg, @nataliyah123, @perfinion, @qlzh727, @rmlarsen, @rushabh-v, @rybakov, @seanpmorgan, @stephengmatthews, @tgaddair and @thaink

Assets 2
Loading