Skip to content

v0.7.16

Choose a tag to compare

@github-actions github-actions released this 06 Nov 20:02
· 10 commits to master since this release
9ff8e64

Zygote v0.7.16

Diff since v0.7.15

Merged pull requests:

Closed issues:

  • Wrong gradient results for multiple getindex chained by foldl (#1558)
  • Triage 15 June 2026 (#1631)
  • accumulate(...; init=x) silently drops the gradient of x (#1684)
  • First gradient of vcat(xs...)/hcat(xs...) with a vector splat takes seconds and grows with the number of arguments (#1687)
  • Ambiguity MethodError when broadcasting -, identity, tanh, conj, real, imag, abs2, xlogx, log1pexp over a Fill (#1688)
  • repeat(x; inner, outer) adjoint is a scalar loop: ~140x slower on CPU, errors on CuArray (#1689)
  • Broaden the precompile workload to cut time-to-first-gradient (#1690)
  • mapreduce/reduce with +/* over arrays has a quadratic gradient; add rules (#1691)
  • Broadcast x .^ p backward uses a runtime-exponent ^ and is not vectorised (#1692)
  • Broadcast adjoint eagerly computes cotangents for constant and data arguments (extra kernel, reduction and D2H sync) (#1694)
  • Broadcast Dual path: retained Dual array and extra kernels; cheaper alternatives and dedicated sqrt/inv/exp/log rules (#1696)
  • Pullbacks of functions with branches or loops are Pullback{F,Any} (type-unstable, allocating) (#1697)
  • Pullback generation: redundant IR transforms and compile time superlinear in basic blocks (#1698)
  • Linear-algebra adjoints: pinv builds m×m temporaries, Diagonal \ and / densify, \ refactorizes (#1699)
  • CUDA: host round trips in tr/Diagonal accumulation and in ProjectTo(::CuSparseMatrixCSC) (#1700)
  • First gradient of vcat(xs...) with a vector of scalars or a mixed splat takes seconds and grows with N (follow-up to #1687) (#1716)
  • MethodError accumulating the cotangent of a splatted vector in vcat(a, vs..., b) when vs is also used elsewhere (#1717)
  • Implicit Grads store raw thunks: pairs(gs) returns a thunk and gs[w] recomputes it (#1721)