« Changelog History


frugally-deep v0.20.0 Release Notes

Release Date: 2026-05-03 // about 1 month ago

  • 🚀 Minor release adding broad new Keras 3 layer support (#465).

    Recurrent layers (restored after the Keras 3 weight-format change broke the previous implementation, plus newly added):

    • 🔀 LSTM, GRU, SimpleRNN, Bidirectional (all merge_mode values)
    • ConvLSTM1D, ConvLSTM2D, ConvLSTM3D
    • RNN wrapping LSTMCell / GRUCell / SimpleRNNCell / StackedRNNCells

    Attention / dense:

    • 💅 EinsumDense (generic einsum interpreter; covers transformer-style projections)
    • GroupedQueryAttention (with optional use_gate=True)

    Pooling / convolution:

    • AdaptiveAveragePooling1D/2D/3D, AdaptiveMaxPooling1D/2D/3D
    • DepthwiseConv1D

    Normalization:

    • RMSNormalization
    • GroupNormalization

    Preprocessing:

    • Discretization, IntegerLookup
    • Masking (passthrough)
    • 🌐 Many training-only image augmentation layers as identity passthroughs (RandomBrightness, RandomFlip, RandomCrop, RandomContrast, RandomRotation, RandomTranslation, RandomZoom, RandomHue, RandomSaturation, RandomSharpness, AugMix, CutMix, MixUp, RandAugment, Equalization, Solarization, Pipeline, MaxNumBoundingBoxes, etc.)

    🔧 Numerical correctness verified end-to-end against Keras 3.14 across multiple distinct configurations of every new layer.


Previous changes from v0.19.5

  • 🚀 Patch release: bump tested TensorFlow / Keras versions to the latest.

    • Bump CI pin and the README's documented tested versions to TensorFlow 2.21.0 / Keras 3.14.0 (was 2.18.0 / 3.8.0).
    • Accept the renamed Threshold layer config keys Keras 3.14 emits (threshold and default_value), with fallback to the old keys (threshold_value and value) for models exported by older Keras versions.
    • ⚡️ Update the perf-test Dockerfile to match: bazel-7.7.0 (required by TF 2.21's .bazelversion), xxd tooling, and an explicit keras==3.14.0 pin.

    🏗 Verified by running the test suite on the new versions and a full end-to-end docker build of the perf-test Dockerfile.