Skip to content

Add support for recursing Chains in ReactantCompatibleOptimisers #1622

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ispielma wants to merge 3 commits into
base: main
Choose a base branch
from
+16 −0
Open

Add support for recursing Chains in ReactantCompatibleOptimisers #1622

Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
76f09d0
Add support for recursing Chains in ReactantCompatibleOptimisers
Jan 8, 2026
ed66c0b
Add missing :
Jan 8, 2026
dfa0891
Merge branch 'LuxDL:main' into optimizers
ispielma Jan 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions src/helpers/optimizers.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,22 @@ function Optimisers.init(
return zero(x), zero(x), Utils.convert_eltype.((T,), opt.opt.beta), zero(x)
end

# Recurse through chains
function Optimisers.adjust(
chain::ReactantOptimiser{<:Optimisers.OptimiserChain},
eta::Real
)
results = Optimisers.OptimiserChain([Optimisers.adjust(opt, eta) for opt in chain.opt.opts]...)
return ReactantOptimiser(results)
end
function Optimisers.adjust(
chain::ReactantOptimiser{<:Optimisers.OptimiserChain};
kw...
)
Comment on lines +51 to +61
Copy link
Member

@avik-pal avik-pal Jan 9, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

will overloading _adjust work instead?

Copy link
Contributor Author

@ispielma ispielma Jan 12, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It could be done that way, but there would be more duplicated code since the implementation would end up replicating the recursion code already present in Optimisers.adjust.

results = Optimisers.OptimiserChain([Optimisers.adjust(opt; kw...) for opt in chain.opt.opts]...)
return ReactantOptimiser(results)
end

function Optimisers._adjust(opt::ReactantOptimiser, nt::NamedTuple)
dev = with_track_numbers(get_device(opt), AbstractFloat)
return ReactantOptimiser(Optimisers._adjust(opt.opt, dev(nt)))
Expand Down