Bump the all-julia-packages group across 7 directories with 12 updates

[dependabot]

Updates the requirements on OrderedCollections, PartitionedDistributions, StatsFuns, DynamicPPL, Bijectors, AbstractPPL, ADTypes, ForwardDiff, DifferentiationInterface, Enzyme, Distributions and LogDensityProblems to permit the latest version.
Updates OrderedCollections to 2.0.1

Release notes

Sourced from OrderedCollections's releases.

v2.0.1

OrderedCollections v2.0.1

Diff since v2.0.0

Merged pull requests:

• avoid rehashing in read-only operations (and fix some bugs) (#169) (@​KristofferC)
• Bump version from 2.0.0 to 2.0.1 (#170) (@​KristofferC)

Commits

• ee78af1 Bump version from 2.0.0 to 2.0.1 (#170)
• 304d5ed avoid rehashing in read-only operations (and fix some bugs) (#169)
• d5d76c7 Merge pull request #164 from JuliaCollections/ox/v2
• 68ab740 Remove depreacations for v2.0 release
• 7bb0847 Bump version from 1.8.1 to 1.8.2
• 6af78d8 OrderedDict: improve performance for non-concrete values and keys (#166)
• 1eaa980 Bump actions/checkout from 5 to 6 (#157)
• 7be2211 Bump actions/checkout from 4 to 5 (#156)
• a25bffa Add Integer type annotation to sizehint! overloads (#155)
• a64f186 issorted(ld; byvalue=true) for LittleDicts (#151)
• Additional commits viewable in compare view

Updates PartitionedDistributions to 0.1.0

Release notes

Sourced from PartitionedDistributions's releases.

v0.1.0

PartitionedDistributions v0.1.0

Breaking changes

• No breaking changes. Version number updated from 0.0.x series to 0.1.x series, which Julia semver interprets as breaking.

Closed issues:

• Renaming the package (#12)
• TagBot trigger issue (#13)
• Fallback for things like pointwise_conditional_logpdfs(::UnivariateDistribution, ::AbstractArray)? (#20)

Commits

• 3061e9f chore: Bump version from 0.0.2 to 0.1.0 (#23)
• 70da5c1 build(deps): Update LogExpFunctions requirement from 0.3.2 to 0.3.2, 1.0 (#21)
• 6cef6fd build(deps): Bump codecov/codecov-action from 6 to 7 (#22)
• 9c6310d build(deps): Bump julia-actions/setup-julia from 2 to 3 (#15)
• 8a29d95 chore: Tweak dependabot settings (#19)
• c0b0ba6 Clarify description of PartitionedDistributions (#14)
• See full diff in compare view

Updates OrderedCollections to 2.0.1

Release notes

Sourced from OrderedCollections's releases.

v2.0.1

OrderedCollections v2.0.1

Diff since v2.0.0

Merged pull requests:

• avoid rehashing in read-only operations (and fix some bugs) (#169) (@​KristofferC)
• Bump version from 2.0.0 to 2.0.1 (#170) (@​KristofferC)

Commits

• ee78af1 Bump version from 2.0.0 to 2.0.1 (#170)
• 304d5ed avoid rehashing in read-only operations (and fix some bugs) (#169)
• d5d76c7 Merge pull request #164 from JuliaCollections/ox/v2
• 68ab740 Remove depreacations for v2.0 release
• 7bb0847 Bump version from 1.8.1 to 1.8.2
• 6af78d8 OrderedDict: improve performance for non-concrete values and keys (#166)
• 1eaa980 Bump actions/checkout from 5 to 6 (#157)
• 7be2211 Bump actions/checkout from 4 to 5 (#156)
• a25bffa Add Integer type annotation to sizehint! overloads (#155)
• a64f186 issorted(ld; byvalue=true) for LittleDicts (#151)
• Additional commits viewable in compare view

Updates StatsFuns to 2.2.0

Release notes

Sourced from StatsFuns's releases.

v2.2.0

StatsFuns v2.2.0

Diff since v2.1.0

Merged pull requests:

• add owens_t function (#214) (@​stevengj)

Closed issues:

• Feature request: Owen's t function (#99)

Commits

• 5568af3 add owens_t function (#214)
• f8d7f5b CompatHelper: bump compat for LogExpFunctions to 1, (keep existing compat) (#...
• 21903f7 Fix srdist by passing nranges=1 to Rmath.ptukey/qtukey (#211)
• 53dc188 Bump julia-actions/setup-julia from 2 to 3 (#210)
• 37d3c86 Bump julia-actions/cache from 2 to 3 (#208)
• dae083e Bump codecov/codecov-action from 5 to 6 (#207)
• e0a0cab Remove Rmath sampling methods (#188)
• 08d54f5 Support JET@0.11 in tests (#195)
• dee41f6 add nbinom pdfs and cdfs (#197)
• 63373be Bump actions/checkout from 5 to 6 (#192)
• Additional commits viewable in compare view

Updates OrderedCollections to 2.0.1

Release notes

Sourced from OrderedCollections's releases.

v2.0.1

OrderedCollections v2.0.1

Diff since v2.0.0

Merged pull requests:

• avoid rehashing in read-only operations (and fix some bugs) (#169) (@​KristofferC)
• Bump version from 2.0.0 to 2.0.1 (#170) (@​KristofferC)

Commits

• ee78af1 Bump version from 2.0.0 to 2.0.1 (#170)
• 304d5ed avoid rehashing in read-only operations (and fix some bugs) (#169)
• d5d76c7 Merge pull request #164 from JuliaCollections/ox/v2
• 68ab740 Remove depreacations for v2.0 release
• 7bb0847 Bump version from 1.8.1 to 1.8.2
• 6af78d8 OrderedDict: improve performance for non-concrete values and keys (#166)
• 1eaa980 Bump actions/checkout from 5 to 6 (#157)
• 7be2211 Bump actions/checkout from 4 to 5 (#156)
• a25bffa Add Integer type annotation to sizehint! overloads (#155)
• a64f186 issorted(ld; byvalue=true) for LittleDicts (#151)
• Additional commits viewable in compare view

Updates DynamicPPL to 0.42.1

Release notes

Sourced from DynamicPPL's releases.

v0.42.1

DynamicPPL v0.42.1

Diff since v0.42.0

Closed issues:

• Separate Mooncake tests into its own test suite (#1352)
• DynamicPPL 0.41.8 vs 0.42.0: posteriordb-bench Comparison (#1412)

Changelog

Sourced from DynamicPPL's changelog.

0.42.1

Fixed a type-inference failure that made nested submodels (a ~ to_submodel(...) statement inside a model that is itself evaluated as a submodel) very slow.

Previously, evaluating a submodel recursed through the shared _evaluate!!(::Model, ::AbstractVarInfo) method. Each level of nesting adds another context layer to the Model type, which tripped Julia's type-inference recursion limit: from the first level of nesting onwards the return type was inferred as Any and evaluation fell back to runtime dispatch, slowing down both primal and gradient evaluation (the primal slowdown was roughly 15x, growing with nesting depth). Submodel evaluation now calls the model function directly, keeping nested submodels type-stable. See Turing.jl#2844.

0.42.0

LogDensityFunction now performs AD preparation through AbstractPPL's prepare / value_and_gradient!! interface instead of calling DifferentiationInterface directly. Internally this removes the _use_closure heuristic and the explicit DI.Constant plumbing; the choice between closure and constants now lives in AbstractPPL.

logdensity_at has been renamed to logdensity_internal. The old name is kept as a const alias so external callers do not break.

LogDensityAt is now a deprecation shim that emits a warning and returns an AbstractPPL.Evaluators.VectorEvaluator whose call forwards to logdensity_internal. New code should call AbstractPPL.prepare(logdensity_internal, x; context=...) directly.

Breaking changes

DifferentiationInterface is no longer a hard dependency of DynamicPPL. With AbstractPPL 0.15.2, the following backends now have native AbstractPPL extensions and only need the concrete AD package loaded:

• AutoForwardDiff — load ForwardDiff
• AutoMooncake, AutoMooncakeForward — load Mooncake

For other DI-routed backends like AutoReverseDiff, users must load DifferentiationInterface together with the concrete AD package:

using DynamicPPL, ADTypes, DifferentiationInterface, ReverseDiff
ldf = LogDensityFunction(model; adtype=AutoReverseDiff())

For distributed sampling the same packages must be loaded on every worker.

Compatibility bounds bumped:

• AbstractPPL 0.14 → 0.15
• Bijectors 0.15.17 → 0.16

The integration test suites for MarginalLogDensities, ReverseDiff, and Enzyme now live in their own environments under test/ext/DynamicPPL*Ext/ and run as separate CI jobs.

0.41.8

Override MarginalLogDensities.optimize_marginal! for LogDensityFunctionWrapper so the underlying OptimizationProblem is rebuilt with the current non-marginalised parameters on each call, rather than reusing a stale problem.

0.41.7

Enable usage of ComponentVectors on the left-hand side of tilde-statements.

Accessing a nonexistent variable in a VarNamedTuple now throws a KeyError with the original VarName, instead of an opaque type NamedTuple has no field ... error.

0.41.6

Add a factorize::Bool keyword argument for pointwise_logdensities(model, values), which controls whether pointwise logdensities for factorisable distributions (e.g. MvNormal, product_distribution, etc.) are returned as a single log-density for the whole distribution, or as an array of log-densities for each factor.

... (truncated)

Commits

• 0a1499f Fix type-inference failure for nested submodels (#1427)
• b310eec skip codecov upload on Dependabot PRs (#1425)
• 173defb Bump codecov/codecov-action from 6 to 7 in the all-github-actions-packages gr...
• b887a35 Bump the all-github-actions-packages group with 2 updates (#1421)
• a81ff4d Remove CompatHelper and update Dependabot config (#1420)
• 9347d11 Update HISTORY.md for 0.42.0 (#1408)
• a115fa8 Use AbstractPPL AD interface (#1363)
• d2052a1 Benchmark improvements (#1396)
• 9e5ae37 Bump patch version to 0.41.8 (#1397)
• 90a74c3 Override optimize_marginal! for LogDensityFunctionWrapper (#1395)
• Additional commits viewable in compare view

Updates Bijectors to 0.16.1

Release notes

Sourced from Bijectors's releases.

v0.16.1

Bijectors v0.16.1

Diff since v0.16.0

Add Mooncake@v0.6 forward-mode support for find_alpha, and widen Mooncake compat to include 0.6.

Merged pull requests:

• Add CDFBijector and QuantileBijector (#273) (@​sethaxen)
• Remove CompatHelper and update Dependabot config (#477) (@​shravangoswami-bot)
• Update Buildkite pipeline for the new JuliaGPU cluster (#479) (@​maleadt)
• Support Mooncake 0.6 forward mode in find_alpha rules (#480) (@​yebai)

Changelog

Sourced from Bijectors's changelog.

0.16.1

Add Mooncake 0.6 forward-mode support for find_alpha, and widen Mooncake compat to include 0.6.

0.16.0

Breaking changes

ChainRulesCore and EnzymeCore have been moved from direct dependencies to weak dependencies. Their integration code, including the find_alpha Enzyme rule and the rrules previously defined in src/chainrules.jl, now lives in the new BijectorsChainRulesCoreExt and BijectorsEnzymeCoreExt package extensions. These extensions are only loaded when the user also loads ChainRulesCore or EnzymeCore themselves, so downstream packages that relied on Bijectors to pull in these packages transitively must now load them directly.

DifferentiationInterface has been removed from the package's direct dependencies. It is no longer used at runtime.

0.15.24

Export a few more functions from Bijectors.VectorBijectors, namely scalar_to_scalar_bijector, TypedIdentity, Log, and Untruncate.

0.15.23

Implement vector bijectors for Distributions.MvTDist.

0.15.22

Implement vector bijectors for Bijectors.TransformedDistribution.

0.15.21

Add compatibility with Roots.jl v3.

0.15.20

Export Bijectors.VectorBijectors.has_constant_vec_bijector to allow downstream packages to overload it.

0.15.19

Optimise performance of vector bijectors for products of univariate distributions.

0.15.18

Implemented vector bijectors for Bijectors.OrderedDistribution.

0.15.17

Adds a new module, Bijectors.VectorBijectors. This module implements bijectors that always return vectors as the output of a linking operation. This provides additional guarantees over the existing Bijectors interface, which in general will return some collection of values that are in Euclidean space, but whose shape is not generally a vector (it could be a scalar or a multidimensional array). The intention is to make it easier to implement bijectors that are compatible with libraries that use vector inputs, such as LogDensityProblems.jl (which is in turn heavily used in the Turing ecosystem).

Please see the documentation for further information.

0.15.16

... (truncated)

Commits

• de1127e Support Mooncake 0.6 forward mode in find_alpha rules (#480)
• 529125a Update Buildkite pipeline for the new JuliaGPU cluster. (#479)
• 880764e Remove CompatHelper and update Dependabot config (#477)
• 201d057 Add CDFBijector and QuantileBijector (#273)
• 1b9e3eb Bump version to 0.16.0 (#474)
• d758493 Use AbstractPPL AD interface in tests. (#460)
• e2ed59a Package deps cleanup (#471)
• 7bb32c5 Move AD backends to standalone integration test suites (#470)
• ad20881 Bump julia-actions/setup-julia from 2 to 3 (#467)
• f99122b bump patch (#466)
• Additional commits viewable in compare view

Updates AbstractPPL to 0.15.3

Release notes

Sourced from AbstractPPL's releases.

v0.15.3

AbstractPPL v0.15.3

Diff since v0.15.2

Closed issues:

• Add ForwardDiff support (#165)

Changelog

Sourced from AbstractPPL's changelog.

0.15.3

Added the of type system: a self-contained, declarative way to specify the shape, element type, and support of model variables. Construct specifications with the exported of function or the exported @of macro:

using AbstractPPL
s = of(Real, 0, 1)                       # a real in [0, 1]
RegressionParams = @of(
    y = of(Array, Float64, 100), beta = of(Array, Float64, 3), sigma = of(Real, 0, nothing),
)

of-types support rand/zero (drawing or zeroing a value of the declared shape) and size/length (querying the declared shape), and can be flattened to / unflattened from a flat numeric vector. rand accepts an optional AbstractRNG for reproducible draws. flatten returns a vector whose element type is the promotion of the declared leaf types, and unflatten is automatic-differentiation transparent (float leaves take promote_type(declared, eltype(flat)), so ForwardDiff.Dual numbers flow through); both are type-stable. These are narrow methods dispatched on AbstractPPL-owned OfType subtypes. The supporting types (OfType and its subtypes) and the inspection/flatten/unflatten helpers are marked public for downstream use without being exported.

0.15.2

Added AbstractPPLForwardDiffExt, a direct ForwardDiff path for AutoForwardDiff (gradient, Jacobian, Hessian, context, chunk size, custom tag).

0.15.1

Added Hessian support to the AD interface. Pass order=2 to prepare(adtype, problem, x) to build a Hessian-capable evaluator. The new value_gradient_and_hessian!!(prepared, x) then returns (value, gradient, hessian) in a single call. Both the DifferentiationInterface and Mooncake extensions implement this.

order=2 accepts the same context::Tuple and check_dims options as order=1. NamedTuple inputs are not supported with order=2.

The AbstractPPLTestExt conformance harness now provides a :hessian group covering value, gradient and Hessian round-trips, along with an order-mismatch edge case.

0.15.0

Added a new evaluator-preparation interface for AD backends. prepare binds a callable to a sample input, either a vector or a NamedTuple, and returns a Prepared wrapper. value_and_gradient!! and value_and_jacobian!! then compute value-and-derivative pairs from that wrapper. The !! suffix means the returned derivative may alias backend cache storage, so copy it if you need to keep it across calls.

using ADTypes, Mooncake  # or DifferentiationInterface + ForwardDiff
using AbstractPPL: prepare, value_and_gradient!!
prepared = prepare(AutoMooncake(), x -> -0.5 * sum(abs2, x), zeros(3))
val, grad = value_and_gradient!!(prepared, [1.0, 2.0, 3.0])
# val == -7.0; grad == [-1.0, -2.0, -3.0]

Added two AD-backend extensions:

• AbstractPPLDifferentiationInterfaceExt, supporting DifferentiationInterface backends.
• AbstractPPLMooncakeExt, supporting AutoMooncake and AutoMooncakeForward.

The new evaluator path supports scalar gradients, vector-output Jacobians, NamedTuple inputs, reusable AD caches, input-shape checks, and constant context arguments via context::Tuple.

AbstractPPLTestExt now provides a small AD conformance harness through generate_testcases and run_testcases, with reserved groups :vector, :namedtuple, :edge, and :cache_reuse.

See https://github.com/TuringLang/AbstractPPL.jl/blob/main/docs/src/evaluators.md for the full interface, the check_dims and context::Tuple options, the NamedTuple input path, and extension-author guidance.

0.14.2

... (truncated)

Commits

• 0c49dba Add the of type from JuliaBUGS.jl (#168)
• 831c9c9 Bump codecov/codecov-action in the all-github-actions-packages group (#175)
• a6da56d skip codecov upload on Dependabot PRs (#174)
• 123cdd4 Bump the all-github-actions-packages group with 3 updates (#172)
• 5c1a26c Replace CompatHelper with Dependabot (#171)
• 9ebf18f Add direct ForwardDiff extension (#166)
• 2d35e51 Update HISTORY.md for 0.15.1 (#164)
• 4390c43 Add order=2 Hessian preparation via value_gradient_and_hessian!! (#163)
• 1599516 Revise 0.15.0 HISTORY.md (#162)
• 451bfa6 Add Mooncake extension; extend AD conformance suite with NamedTuple and cache...
• Additional commits viewable in compare view

Updates ADTypes to 1.22.1

Release notes

Sourced from ADTypes's releases.

v1.22.1

Release v1.22.1 (tag created to match the registered version).

Commits

• 6cbd3d8 Release v1.22.1 (#160)
• 839efbb Use SciMLTesting v1.2 folder-based run_tests (#159)
• b690050 Split tests into Core/QA groups; CI.yml -> grouped-tests.yml caller (#157)
• d5b0a58 Canonical CI cleanup: TagBot thin-caller + downgrade-caller cleanup (#156)
• 676f171 Drop allow-reresolve input from downgrade workflow(s) (#155)
• 5ff806c ci: add standard centralized workflows (auto-format, dependabot-automerge, do...
• 410f861 Merge pull request #153 from ChrisRackauckas-Claude/downgrade-ci-reenable
• 4c94a00 Downgrade: drop Julia 'pre' from matrix (Resolver.jl --min fails on 1.13-rc1:...
• ce2d236 Re-enable downgrade CI with genuine test-at-floor (allow_reresolve: false)
• 61ada37 Merge pull request #152 from ChrisRackauckas-Claude/centralize-ci
• Additional commits viewable in compare view

Updates ForwardDiff to 1.4.1

Release notes

Sourced from ForwardDiff's releases.

v1.4.1

ForwardDiff v1.4.1

Diff since v1.4.0

Merged pull requests:

• Remove allocations of nested Jacobians of StaticArrays (#810) (@​albert-de-montserrat)
• Bump codecov/codecov-action from 6 to 7 (#814) (@​dependabot[bot])

Commits

• 777420f Remove allocations of nested Jacobians of StaticArrays (#810)
• 5bf530a Bump codecov/codecov-action from 6 to 7 (#814)
• 0a3d297 CompatHelper: bump compat for LogExpFunctions to 1, (keep existing compat) (#...
• 39916da CompatHelper: bump compat for JET in [extras] to 0.11, (keep existing compat)...
• 7e52ffa Bump julia-actions/setup-julia from 2 to 3 (#802)
• 39a27fe Bump codecov/codecov-action from 5 to 6 (#799)
• 1295777 Avoid reshape allocation in extract_jacobian! for Matrix results (#797)
• ff0d903 Bump julia-actions/cache from 2 to 3 (#796)
• 7262054 Remove explicit != methods for Dual (#793)
• 71258ec Update documentation formatting and infrastructure (#792)
• Additional commits viewable in compare view

Updates DynamicPPL to 0.42.1

Release notes

Sourced from DynamicPPL's releases.

v0.42.1

DynamicPPL v0.42.1

Diff since v0.42.0

Closed issues:

• Separate Mooncake tests into its own test suite (#1352)
• DynamicPPL 0.41.8 vs 0.42.0: posteriordb-bench Comparison (#1412)

Changelog

Sourced from DynamicPPL's changelog.

0.42.1

Fixed a type-inference failure that made nested submodels (a ~ to_submodel(...) statement inside a model that is itself evaluated as a submodel) very slow.

Previously, evaluating a submodel recursed through the shared _evaluate!!(::Model, ::AbstractVarInfo) method. Each level of nesting adds another context layer to the Model type, which tripped Julia's type-inference recursion limit: from the first level of nesting onwards the return type was inferred as Any and evaluation fell back to runtime dispatch, slowing down both primal and gradient evaluation (the primal slowdown was roughly 15x, growing with nesting depth). Submodel evaluation now calls the model function directly, keeping nested submodels type-stable. See Turing.jl#2844.

0.42.0

LogDensityFunction now performs AD preparation through AbstractPPL's prepare / value_and_gradient!! interface instead of calling DifferentiationInterface directly. Internally this removes the _use_closure heuristic and the explicit DI.Constant plumbing; the choice between closure and constants now lives in AbstractPPL.

logdensity_at has been renamed to logdensity_internal. The old name is kept as a const alias so external callers do not break.

LogDensityAt is now a deprecation shim that emits a warning and returns an AbstractPPL.Evaluators.VectorEvaluator whose call forwards to logdensity_internal. New code should call AbstractPPL.prepare(logdensity_internal, x; context=...) directly.

Breaking changes

DifferentiationInterface is no longer a hard dependency of DynamicPPL. With AbstractPPL 0.15.2, the following backends now have native AbstractPPL extensions and only need the concrete AD package loaded:

• AutoForwardDiff — load ForwardDiff
• AutoMooncake, AutoMooncakeForward — load Mooncake

For other DI-routed backends like AutoReverseDiff, users must load DifferentiationInterface together with the concrete AD package:

using DynamicPPL, ADTypes, DifferentiationInterface, ReverseDiff
ldf = LogDensityFunction(model; adtype=AutoReverseDiff())

For distributed sampling the same packages must be loaded on every worker.

Compatibility bounds bumped:

• AbstractPPL 0.14 → 0.15
• Bijectors 0.15.17 → 0.16

The integration test suites for MarginalLogDensities, ReverseDiff, and Enzyme now live in their own environments under test/ext/DynamicPPL*Ext/ and run as separate CI jobs.

0.41.8

Override MarginalLogDensities.optimize_marginal! for LogDensityFunctionWrapper so the underlying OptimizationProblem is rebuilt with the current non-marginalised parameters on each call, rather than reusing a stale problem.

0.41.7

Enable usage of ComponentVectors on the left-hand side of tilde-statements.

Accessing a nonexistent variable in a VarNamedTuple now throws a KeyError with the original VarName, instead of an opaque type NamedTuple has no field ... error.

0.41.6

Add a factorize::Bool keyword argument for pointwise_logdensities(model, values), which controls whether pointwise logdensities for factorisable distributions (e.g. MvNormal, product_distribution, etc.) are returned as a single log-density for the whole distribution, or as an array of log-densities for each factor.

... (truncated)

Commits

• 0a1499f Fix type-inference failure for nested submodels (#1427)
• b310eec skip codecov upload on Dependabot PRs (#1425)
• 173defb Bump codecov/codecov-action from 6 to 7 in the all-github-actions-packages gr...
• b887a35 Bump the all-github-actions-packages group with 2 updates (#1421)
• a81ff4d Remove CompatHelper and update Dependabot config (#1420)
• 9347d11 Update HISTORY.md for 0.42.0 (#1408)
• a115fa8 Use AbstractPPL AD interface (#1363)
• d2052a1 Benchmark improvements (#1396)
• 9e5ae37 Bump patch version to 0.41.8 (#1397)
• 90a74c3 Override optimize_marginal! for LogDensityFunctionWrapper (#1395)
• Additional commits viewable in compare view

Updates DifferentiationInterface to 0.7.18

Release notes

Sourced from DifferentiationInterface's releases.

DifferentiationInterface-v0.7.18

DifferentiationInterface DifferentiationInterface-v0.7.18

Diff since DifferentiationInterface-v0.7.17

Merged pull requests:

• feat: add support for HyperHessian backend (#940) (@​KristofferC)
• fix: proper eltype in wrong-mode pushforward/pullback (#1003) (@​gdalle)

Closed issues:

• FiniteDiff with :central includes junk (#983)
• Tracking removal from Turing-verse (#992)

Commits

• f477fa3 fix: proper eltype in wrong-mode pushforward/pullback (#1003)
• cb6801e feat: add support for HyperHessian backend (#940)
• d38d905 fix: make DI compatible with latest Mooncake friendly tangents (#1001)
• 1b5d91c chore: clean up backend testing CI (#1002)
• 5a3ea38 test: activate Enzyme tests on Julia 1.12 (#966)
• ee29327 chore(deps): bump julia-actions/setup-julia from 2 to 3 (#994)
• 508572e chore: upper-bound Mooncake version (coherence with General) (#995)
• a5ecbe0 test: shorten Mooncake test suite (#985)
• 9780927 doc: add docstrings to the result anlysis methods for sparse matrix preparati...
• 1a1ff88 chore(deps): bump codecov/codecov-action from 5 to 6 (#982)
• Additional commits viewable in compare view

Updates Enzyme to 0.13.176

Release notes

Sourced from Enzyme's releases.

v0.13.176

Enzyme v0.13.176

Diff since v0.13.175

Commits

• e619385 Bump version to 0.13.176 and update dependencies
• 40071d5 Fix loosen requirements for absint of newstruct (#3286)
• 33378d2 Fix specfunc calling conv (#3285)
• 39b2721 Fix false_from_args in tape type (#3282)
• ae75a2b bump jll version (#3283)
• cab0404 Guard uses of GPUCompiler.llvm_targetinfo (#3281)
• 37b0ab3 Bump version from 0.13.173 to 0.13.174
• 4bc580c Add link_split_existing! to fix spurious "symbol multiply defined" link error...
• 03439e8 CI + target transformation info for Metal support (#3271)
• 1bfec1f Merge pull request #3274 from EnzymeAD/ksh/typo
• Additional commits viewable in compare view

Updates AbstractPPL to 0.15.3

Release notes

Sourced from AbstractPPL's releases.

v0.15.3

AbstractPPL v0.15.3

Diff since v0.15.2

Closed issues:

• Add ForwardDiff support (#165)

Changelog

Sourced from AbstractPPL's changelog.

0.15.3

Added the of type system: a self-contained, declarative way to specify the shape, element type, and support of model variables. Construct specifications with the exported of function or the exported @of macro:

using AbstractPPL
s = of(Real, 0, 1)                       # a real in [0, 1]
RegressionParams = @of(
    y = of(Array, Float64, 100), beta = of(Array, Float64, 3), sigma = of(Real, 0, nothing),
)

of-types support rand/zero (drawing or zeroing a value of the declared shape) and size/length (querying the declared shape), and can be flattened to / un...

Description has been truncated

[github-actions]

DynamicPPL.jl documentation for PR #1433 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1433/

[github-actions]

Benchmarks @ 770da26

Performance Ratio: gradient time divided by log-density time.

For very small models these ratios are noisy across runs and machines; raw primal and gradient timings are more reliable. The benchmarks are aimed at DynamicPPL developers and mainly catch obvious allocation or type-stability regressions. See benchmark notes for details.

===================================================================================================
                                               eval                       gradient                 
                                            ----------  -------------------------------------------
Model                        dim    linked      primal     FwdDiff    RvsDiff    Mooncake    Enzyme
---------------------------------------------------------------------------------------------------
Simple assume observe*         1     false      4.2 ns       14.25    1624.81       43.72     18.96
Simple assume observe*         1      true     4.19 ns       14.37    1755.02       46.99     19.06
Smorgasbord                  201     false     5.71 μs       78.30     136.71        6.89      7.65
Smorgasbord                  201      true     7.45 μs       77.52     140.53        6.13      5.70
Loop univariate 1k          1000     false     16.2 μs     1185.21     331.94        9.84      7.94
Loop univariate 1k          1000      true     19.8 μs     1561.73     273.12        7.98      6.68
Multivariate 1k             1000     false     23.8 μs      329.05      63.16        7.86      2.10
Multivariate 1k             1000      true     23.6 μs      302.89      69.63        8.93      2.07
Loop univariate 10k        10000     false    157.0 μs    16477.48     372.67       10.30      8.05
Loop univariate 10k        10000      true    193.0 μs    14512.88     304.94        8.29      6.57
Multivariate 10k           10000     false    215.0 μs     4801.33      78.80       10.14      1.93
Multivariate 10k           10000      true    215.0 μs     5596.76      79.66       10.10      1.94
Dynamic                       15     false     1.44 μs         err      46.53       14.47     11.88
Dynamic                       10      true     1.95 μs        2.06      56.53       11.49     18.73
Submodel*                      1     false      4.2 ns       14.36    1853.60       44.13     18.95
Submodel*                      1      true     4.21 ns       14.42    2103.33       44.74     18.70
LDA                           12      true     24.9 μs        0.70       1.99       30.08       err
===================================================================================================

Main @ d7e84ce

===================================================================================================
                                               eval                       gradient                 
                                            ----------  -------------------------------------------
Model                        dim    linked      primal     FwdDiff    RvsDiff    Mooncake    Enzyme
---------------------------------------------------------------------------------------------------
Simple assume observe*         1     false     3.26 ns       15.32    1847.04       44.41     16.73
Simple assume observe*         1      true     3.26 ns       16.07    2440.59       48.64     15.57
Smorgasbord                  201     false     5.16 μs       85.09     137.41        7.87      8.46
Smorgasbord                  201      true     6.93 μs       83.16     141.16        7.04      6.63
Loop univariate 1k          1000     false     13.3 μs     1141.57     330.62       11.34      9.87
Loop univariate 1k          1000      true     17.5 μs     1476.96     248.18        8.04      7.36
Multivariate 1k             1000     false     25.1 μs      336.41      59.38        7.59      2.52
Multivariate 1k             1000      true     24.3 μs      329.05      54.87        9.10      2.77
Loop univariate 10k        10000     false    122.0 μs    21791.31     411.41       11.49      9.99
Loop univariate 10k        10000      true    172.0 μs    15896.31     305.64        8.07      7.17
Multivariate 10k           10000     false    181.0 μs     7965.18      87.61       11.87      2.18
Multivariate 10k           10000      true    180.0 μs     7627.84      84.94       10.75      2.36
Dynamic                       15     false     1.49 μs         err      60.48       20.93     13.39
Dynamic                       10      true     2.01 μs        2.47      62.77       23.62     23.78
Submodel*                      1     false     3.26 ns       15.67    2388.48       50.27     14.19
Submodel*                      1      true     3.28 ns       17.40    3086.94       50.14     17.60
LDA                           12      true     21.1 μs        0.78       1.56       29.08       err
===================================================================================================

Environment

Julia Version 1.11.9
Commit 53a02c0720c (2026-02-06 00:27 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 9V74 80-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver4)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)