v0.4.183

Mooncake v0.4.183

Diff since v0.4.182

Merged pull requests:

• fix DynamicPPL integration tests (#874) (@penelopeysm)
• Improve compile times -- Don't use _methods_by_ftype for primitive detection (#877) (@MasonProtter)
• Some docs improvements (#880) (@penelopeysm)

Closed issues:

• Type Checking in Forward Mode (#672)
• DynamicPPL test errors (#873)

v0.4.182

Mooncake v0.4.182

Diff since v0.4.181

Merged pull requests:

• Add frule and rrule for jl_has_free_typevars (#872) (@sunxd3)

Closed issues:

• missing rrule for foreigncall :jl_has_free_typevars (#871)

v0.4.181

Mooncake v0.4.181

Diff since v0.4.180

Merged pull requests:

• Add Type Checking in Forward Mode (#808) (@sunxd3)
• Document limitations of differentiating CUDA kernels (#866) (@lassepe)
• Bump actions/checkout from 5 to 6 (#869) (@dependabot[bot])

Closed issues:

• Document that support for custom kernels is not planned (#861)
• Seemingly non-ending compilation time in Ferrite.jl Finite Element code (#867)

v0.4.180

Mooncake v0.4.180

Diff since v0.4.179

Merged pull requests:

• Fix inlining avoidance (#759) (@willtebbutt)

Closed issues:

• rrule!! not picked up: Mooncake's is_primative is too rigid sometimes leading to false negatives (#558)

v0.4.179

Mooncake v0.4.179

Diff since v0.4.178

Merged pull requests:

• feat: import rules for modulo from chainrules.jl (#863) (@jmurphy6895)

v0.4.178

Mooncake v0.4.178

Diff since v0.4.177

Merged pull requests:

• _copy_to_output!! error handling for mismatched types (#858) (@AstitvaAggarwal)
• Improve error message for replicating rule compilation errors (#859) (@penelopeysm)

Closed issues:

• _copy_to_output!! not defined for non-differentiable types (#855)

v0.4.177

Mooncake v0.4.177

Diff since v0.4.176

Merged pull requests:

• Support complex numbers in BLAS level 2 and 3 rules (#822) (@Technici4n)

Closed issues:

• Utility for stripping tangents (#472)

v0.4.176

Mooncake v0.4.176

Diff since v0.4.175

Merged pull requests:

• Make tangent of CuArray have the same type (#811) (@kshyatt)
• Bump version after #811 (#852) (@sunxd3)

v0.4.175

Mooncake v0.4.175

Diff since v0.4.174

Merged pull requests:

• More xrefs and ticks for tangents documentation (#849) (@kshyatt)
• Allow selective zeroing of cotangents, take 2 (#851) (@gdalle)

Closed issues:

• Integration with a few more interesting libraries (#378)
• Mooncake less efficient than Zygote or Enzyme on Flux layers (#466)
• Differentiating CUDA kernels (#648)
• Unable to Precompile Mooncake on v1.12 (#803)
• easy_rule like interface for writting rules (#842)

v0.4.174

Mooncake v0.4.174

Diff since v0.4.173

Merged pull requests:

• Add Julia v1.12 compatibility (#714) (@Copilot)
• Try adding fwd/rvs rules for matrix inv and det (#812) (@kshyatt)
• Add some @zero_derivatives for logging (#815) (@mhauru)
• Actually run matrix det derived rule test (#816) (@Technici4n)
• feat: LibuvStream compatibility (#820) (@MilesCranmer)
• Support differentiable data in SimpleVector (#827) (@Technici4n)
• Bump peter-evans/create-or-update-comment from 4 to 5 (#829) (@dependabot[bot])
• Bump peter-evans/find-comment from 3 to 4 (#830) (@dependabot[bot])
• Bump actions/upload-artifact from 4 to 5 (#831) (@dependabot[bot])
• Set the IR's valid worlds to only the interpreter world (#832) (@Technici4n)
• Remove duplicate test_rule call (#838) (@Technici4n)
• Add rules for LAPACK.lacpy! and BLAS.trsv! (#839) (@Technici4n)
• Update JET version to include 0.11 (#841) (@yebai)
• fix allocation counts for primals (plus _new_ frule) (#843) (@penelopeysm)
• Skip certain stability and performance tests for 1.12 (#844) (@yebai)
• Fix for TwicePrecision allocs (#845) (@penelopeysm)

Closed issues:

• Rules for LinearSolve.jl and NonlinearSolve.jl (#235)
• Rules for Integrals.jl (#495)
• Use Mooncake to derive ForwardDiff and ReverseDiff rules (#518)
• Docs have Turing.jl header (#646)
• Docs for OC / MC captures (#679)
• Set up Copilot instructions (#720)
• Import ChainRules for broadcast(f, As...) for Array{<:IEEEFloat} (#737)
• Nested forward errors on jl_get_world_counter foreigncall (#777)
• Support for Base.PipeEndpoint (#800)
• MooncakeRuleCompilationError when Training a CNN with Lux.jl (#804)
• Failing to precompile on Julia 1.12.1 (#821)
• Error when differentiating over KernelAbstractions.jl (#835)
• No method for prepare_gradient when the first argument is Constant or Cache (#836)