FluxML/Flux.jl v0.16.10

on GitHub

Flux v0.16.10

Diff since v0.16.9

Merged pull requests:

• Bump julia-actions/cache from 2 to 3 (#2654) (@dependabot[bot])
• Add AGENTS.md with instructions for AI agents (#2655) (@CarloLucibello)
• Use ParallelTestRunner.jl for parallel test execution (#2656) (@CarloLucibello)
• Run Enzyme tests on julia 1.12 (#2659) (@CarloLucibello)
• Bump dorny/paths-filter from 3.0.2 to 4.0.1 (#2661) (@dependabot[bot])
• mark all documented methods as public (#2662) (@CarloLucibello)
• fix AutoZygote gradient (#2664) (@CarloLucibello)
• Use GPUArrays caching allocator in train! (#2665) (@CarloLucibello)
• Implement _show_leaflike(::Tuple{}) to handle parameters consisting of an empty Tuple (#2667) (@JoshuaBillson)
• Bump codecov/codecov-action from 5 to 6 (#2668) (@dependabot[bot])
• Fix number formatting for PairwiseFusion (#2669) (@hardik-xi11)
• CompatHelper: bump compat for cuDNN in [weakdeps] to 6, (keep existing compat) (#2671) (@github-actions[bot])
• CompatHelper: bump compat for CUDA in [weakdeps] to 6, (keep existing compat) (#2672) (@github-actions[bot])

Closed issues:

• Incorrect link on docs (#2380)
• bad formatting for PairwiseFusion docstring (#2406)
• Use Caching Allocator in train! (#2636)
• all documented functions should be marked as public (#2638)
• Can't install flux. (#2651)
• GSoC 2026 Interest: RL Environments Project (#2663)