[dependabot]

Updates the requirements on AbstractNeuralNetworks to permit the latest version.

Release notes

Sourced from AbstractNeuralNetworks's releases.

v0.9.0

AbstractNeuralNetworks v0.9.0

Diff since v0.8.0

Methods that GeometricMachineLearning (GML) defined on this package's types move here, so GML no longer pirates them. The full notes are in the 0.9.0 section of CHANGELOG.md.

Breaking Changes

• A GML that still defines the 3-tensor methods on Dense and Linear fails to precompile against this release ("Method overwriting is not permitted during Module precompilation"). GML deletes them in the same change that raises its AbstractNeuralNetworks bound.
• The backend keyword of load(NeuralNetwork, …) is removed. The network loads on the CPU; changebackend(backend, nn) moves it to a device.

New Features

• Dense, Linear and Affine accept inputs of rank 3 or more, through reshape and one matrix product, on any array type.
• A weak extension HDF5Ext provides save and load for NeuralNetwork.

Bug Fixes

• A top-level Zygote.pullback(applychain, layers, x, ps) gives the storage gradient of each leaf and keeps the leaf's type; a parameter cotangent of nothing no longer raises a MethodError.
• parameterlength(::Dense) reads the bias flag as a Bool.

Dependencies

• Requires NeuralNetworkParameters 0.4.2 (and so GeometricBase 0.15) and Julia 1.11.

Merged pull requests:

• Take in Dense/Linear/Affine 3-tensor and NeuralNetwork HDF5 methods from GML (#40) (@​michakraus)
• Require NeuralNetworkParameters 0.4 (#41) (@​michakraus)
• Give the storage gradient from applychain's pullback (#42) (@​michakraus)
• Move the test suite to the common layout (#44) (@​michakraus)
• Update DocumenterCitations requirement to 1.5.0 in /docs (#46) (@​dependabot[bot])
• Update Documenter requirement to 1.19.0 in /docs (#47) (@​dependabot[bot])
• Update SafeTestsets requirement to 0.1.0 in /test (#49) (@​dependabot[bot])
• Update Documenter requirement to 1.19.0 in /test (#50) (@​dependabot[bot])
• Read the Dense bias flag as a Bool (#51) (@​michakraus)
• Remove the shared compat entries from test/Project.toml (#52) (@​michakraus)
• Raise the NeuralNetworkParameters floor to 0.4.2 (#53) (@​michakraus)

Changelog

Sourced from AbstractNeuralNetworks's changelog.

[0.9.0] — 2026-10-02

Methods that GeometricMachineLearning (GML) defined on this package's types move here. A method on Dense, Linear or NeuralNetwork is type piracy wherever it is written outside this package. GML defined three such methods, the 3-tensor methods on Dense and Linear, and the save/load methods for NeuralNetwork in its HDF5 extension.

Breaking Changes

• A GML that still defines the three 3-tensor methods fails to precompile against this release, with "Method overwriting is not permitted during Module precompilation". GML has to delete them in the same change that raises its AbstractNeuralNetworks bound.

• The backend keyword of load(NeuralNetwork, …) is removed; the network loads on the CPU and changebackend(backend, nn) moves it to a device with parameters included.

New Features

• Dense, Linear and Affine accept inputs of rank 3 or more. Each layer applies its weight matrix to the first dimension, preserving the remaining dimensions. The implementation uses only reshape and matrix multiplication, so it runs on any array type that supports both — including device-backed arrays.

• A new weak extension, HDF5Ext, provides save and load for NeuralNetwork. The extension defines methods for HDF5 stores; filename forms are NeuralNetworkParameters' own methods. The extension loads automatically when HDF5 is available.

Dependencies and tests

• Requires NeuralNetworkParameters 0.4 (compat only; no change to this package's own behaviour). Under Zygote, NNP 0.4 calls the loss with the NetworkParameters itself rather than the wrapped NamedTuple, and converts the cotangent of each leaf with storage_gradient to the gradient with respect to that leaf's storage. A Chain still accepts a whole set as a bare NamedTuple as well: SymbolicNeuralNetworks builds symbolic networks whose parameters are one.

• Requires NeuralNetworkParameters 0.4.2 (compat only; no change to this package's own behaviour). NNP 0.4.2 requires GeometricBase 0.15, which declares its stubs public and requires Julia 1.11; this package's Julia floor is 1.11 already.

• The test suite follows the common layout (no change to the package's behaviour). The test dependencies move from [extras]/[targets] to test/Project.toml; Aqua and Documenter are new test dependencies. runtests.jl runs the groups core and slow, and each test file sits at the path of the source file it tests. New: test/quality/aqua.jl, with the check for method ambiguities marked broken for the two ambiguities of apply (#43), and test/quality/doctests.jl in slow. LinearAlgebra and Random get the compat bound "1", which Aqua requires. The empty test/architecture_tests.jl, which held no test, is removed.

• test/Project.toml no longer repeats the root's bounds (test environment only; no change to the package). Its [compat] table loses the entries for HDF5, KernelAbstractions, LinearAlgebra, NeuralNetworkParameters, Random and StaticArrays, which are dependencies

... (truncated)

Commits

• e013d83 Release 0.9.0
• 43c1244 Raise the NeuralNetworkParameters floor to 0.4.2 (#53)
• 67dd00a Sync the shared git hooks
• f376ab2 Update the shared pre-push hook: test the suite's exit status directly
• 346cfac Read the Dense bias flag as a Bool (#51)
• cccf34d CI.yml, dependabot.yml: comment fixes (canonical Environment 1f3b574)
• 95e736d dependabot.yml: test/ and docs/ carry no bound for a shared dependency (canon...
• b915508 Remove the shared compat entries from test/Project.toml (#52)
• a1a7119 Dependabot: update the root Project.toml only
• d859a29 Update Documenter requirement to 1.19.0 in /docs (#47)
• Additional commits viewable in compare view

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 49.83%. Comparing base (e111300) to head (be6f954).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #55   +/-   ##
=======================================
  Coverage   49.83%   49.83%           
=======================================
  Files          13       13           
  Lines         616      616           
=======================================
  Hits          307      307           
  Misses        309      309           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dependabot]

Superseded by #57.