Skip to content

Make SkipConnection have the same args as Parallel #1481

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
Qfl3x wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Make SkipConnection have the same args as Parallel #1481

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
44a1db4
Make SkipConnection have the same args as Parallel
Qfl3x Sep 15, 2025
0b839a7
Updated tests and docs
Qfl3x Sep 15, 2025
38350f9
Fixed Bilinear tests
Qfl3x Sep 15, 2025
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
4 changes: 2 additions & 2 deletions examples/CIFAR10/conv_mixer.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@ function ConvMixer(; dim, depth, kernel_size=5, patch_size=2)
[
Chain(
SkipConnection(
+,
Chain(
Conv(
(kernel_size, kernel_size),
Expand All @@ -24,8 +25,7 @@ function ConvMixer(; dim, depth, kernel_size=5, patch_size=2)
pad=SamePad(),
),
BatchNorm(dim),
),
+,
)
),
Conv((1, 1), dim => dim, relu),
BatchNorm(dim),
Expand Down
2 changes: 1 addition & 1 deletion ext/LuxFluxExt.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,7 +67,7 @@ function Lux.convert_flux_model(l::Flux.SkipConnection; kwargs...)
else
Lux.convert_flux_model(l.connection; kwargs...)
end
return Lux.SkipConnection(Lux.convert_flux_model(l.layers; kwargs...), connection)
return Lux.SkipConnection(connection, Lux.convert_flux_model(l.layers; kwargs...))
end

function Lux.convert_flux_model(l::Flux.Bilinear; preserve_ps_st::Bool=false, kwargs...)
Expand Down
6 changes: 3 additions & 3 deletions src/layers/containers.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,13 +1,13 @@
"""
SkipConnection(layers, connection; name=nothing)
SkipConnection(connection, layers; name=nothing)
SkipConnection(; layers, connection, name=nothing)

Create a skip connection which consists of a layer or [`Chain`](@ref) of consecutive layers
and a shortcut connection linking the block's input to the output through a user-supplied
2-argument callable. The first argument to the callable will be propagated through the given
`layer` while the second is the unchanged, "skipped" input.

The simplest "ResNet"-type connection is just `SkipConnection(layer, +)`.
The simplest "ResNet"-type connection is just `SkipConnection(+, layer)`.

## Arguments

Expand Down Expand Up @@ -51,7 +51,7 @@ end

PrettyPrinting.printable_children(l::SkipConnection) = (; l.connection, l.layers)

function SkipConnection(layers, connection; name::NAME_TYPE=nothing)
function SkipConnection(connection, layers; name::NAME_TYPE=nothing)
return SkipConnection(; layers, connection, name)
end

Expand Down
2 changes: 1 addition & 1 deletion test/downstream/Flux/flux_integrationtest.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ toluxforce = FromFluxAdaptor(; force_preserve=true, preserve_ps_st=true)
end

@testset "Skip Connection" begin
model = dev(Flux.SkipConnection(Flux.Dense(2 => 2), +))
model = dev(Flux.SkipConnection(+, Flux.Dense(2 => 2)))
x = aType(rand(Float32, 2, 1))

model_lux = toluxpsst(model)
Expand Down
2 changes: 1 addition & 1 deletion test/enzyme_tests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,7 +57,7 @@ const MODELS_LIST = Any[
),
(Maxout(() -> Dense(5 => 4, tanh), 3), randn(Float32, 5, 2)),
(Bilinear((2, 2) => 3), randn(Float32, 2, 3)),
(SkipConnection(Dense(2 => 2), vcat), randn(Float32, 2, 3)),
(SkipConnection(vcat, Dense(2 => 2)), randn(Float32, 2, 3)),
(ConvTranspose((3, 3), 3 => 2; stride=2), rand(Float32, 5, 5, 3, 1)),
(StatefulRecurrentCell(RNNCell(3 => 5)), rand(Float32, 3, 2)),
(StatefulRecurrentCell(RNNCell(3 => 5, gelu)), rand(Float32, 3, 2)),
Expand Down
6 changes: 3 additions & 3 deletions test/layers/basic_tests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -289,7 +289,7 @@ end
display(d)
b = Bilinear((2, 2) => 3)
display(b)
layer = SkipConnection(d, b)
layer = SkipConnection(b, d)
display(layer)
ps, st = dev(Lux.setup(rng, layer))
x = aType(randn(rng, Float32, 2, 1))
Expand All @@ -304,7 +304,7 @@ end
display(d)
b = Bilinear((2, 2) => 3; use_bias=false)
display(b)
layer = SkipConnection(d, b)
layer = SkipConnection(b, d)
display(layer)
ps, st = dev(Lux.setup(rng, layer))
x = aType(randn(rng, Float32, 2, 1))
Expand All @@ -319,7 +319,7 @@ end
display(d)
b = Bilinear((3, 2) => 5)
display(b)
layer = SkipConnection(d, b)
layer = SkipConnection(b, d)
display(layer)
ps, st = dev(Lux.setup(rng, layer))
x = aType(randn(rng, Float32, 2, 7, 11))
Expand Down
4 changes: 2 additions & 2 deletions test/layers/containers_tests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@

@testset "$mode" for (mode, aType, dev, ongpu) in MODES
@testset "zero sum" begin
layer = SkipConnection(WrappedFunction(Broadcast.BroadcastFunction(zero)), .+)
layer = SkipConnection(+, WrappedFunction(Broadcast.BroadcastFunction(zero)))
display(layer)
ps, st = dev(Lux.setup(rng, layer))
x = aType(randn(rng, Float32, 10, 10, 10, 10))
Expand All @@ -14,7 +14,7 @@
end

@testset "concat size" begin
layer = SkipConnection(Dense(10, 10), hcat)
layer = SkipConnection(hcat, Dense(10, 10))
display(layer)
ps, st = dev(Lux.setup(rng, layer))
x = aType(randn(rng, Float32, 10, 2))
Expand Down