fix: remove Optimisers.jl patch (#1247)

avik-pal · web-flow · commit 2db68937fccf · 2025-04-01T20:34:13.000-04:00
* fix: remove Optimisers.jl patch

feat: support general optimisers

fix: incorrect field name

fix: setup under jit

chore: run formatter

ci: run only reactant for now (drop me)

fix: make_reactant_compatible after setup

fix: support all optimisers correctly

test: annotate

fix: offload all pathces to Optimisers

* ci: run all tests

* fix: remove branch

* fix: qa tests

* chore: update comment

* fix: module
diff --git a/Project.toml b/Project.toml
@@ -104,7 +104,7 @@ MacroTools = "0.5.13"
 Markdown = "1.10"
 NCCL = "0.1.1"
 NNlib = "0.9.26"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Preferences = "1.4.3"
 Random = "1.10"
 Reactant = "0.2.55"
diff --git a/docs/Project.toml b/docs/Project.toml
@@ -62,7 +62,7 @@ LuxLib = "1.3.4"
 LuxTestUtils = "1.5"
 MLDataDevices = "1.6.10"
 NNlib = "0.9.26"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Printf = "1.10"
 Random = "1.10"
 Reactant = "0.2.55"
diff --git a/examples/Basics/Project.toml b/examples/Basics/Project.toml
@@ -11,5 +11,5 @@ Zygote = "e88e6eb3-aa80-5325-afca-941959d7151f"
 ComponentArrays = "0.15.22"
 ForwardDiff = "0.10, 1"
 Lux = "1"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Zygote = "0.6.70, 0.7"
diff --git a/examples/CIFAR10/Project.toml b/examples/CIFAR10/Project.toml
@@ -35,7 +35,7 @@ LuxCUDA = "0.3.2"
 MLDatasets = "0.7.14"
 MLUtils = "0.4.4"
 OneHotArrays = "0.2.5"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Printf = "1.10"
 ProgressTables = "0.1"
 Random = "1.10"
diff --git a/examples/ConvolutionalVAE/Project.toml b/examples/ConvolutionalVAE/Project.toml
@@ -23,7 +23,7 @@ Lux = "1.4.1"
 MLDatasets = "0.7.18"
 MLUtils = "0.4.4"
 OneHotArrays = "0.2.6"
-Optimisers = "0.4"
+Optimisers = "0.4.6"
 Printf = "1.10"
 Random = "1.10"
 Reactant = "0.2.55"
diff --git a/examples/DDIM/Project.toml b/examples/DDIM/Project.toml
@@ -36,7 +36,7 @@ JLD2 = "0.4.48, 0.5"
 Lux = "1"
 LuxCUDA = "0.3"
 MLUtils = "0.4"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 ParameterSchedulers = "0.4.1"
 ProgressBars = "1"
 Random = "1.10"
diff --git a/examples/GCN_Cora/Project.toml b/examples/GCN_Cora/Project.toml
@@ -18,7 +18,7 @@ GNNGraphs = "1"
 Lux = "1.5"
 MLDatasets = "0.7.18"
 OneHotArrays = "0.2"
-Optimisers = "0.4.4"
+Optimisers = "0.4.6"
 Printf = "1.10"
 Random = "1.10"
 Reactant = "0.2.55"
diff --git a/examples/HyperNet/Project.toml b/examples/HyperNet/Project.toml
@@ -15,5 +15,5 @@ Lux = "1"
 MLDatasets = "0.7"
 MLUtils = "0.4"
 OneHotArrays = "0.2.5"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Reactant = "0.2.55"
diff --git a/examples/ImageNet/Project.toml b/examples/ImageNet/Project.toml
@@ -38,7 +38,7 @@ MLUtils = "0.4.4"
 MPI = "0.20.21"
 NCCL = "0.1.1"
 OneHotArrays = "0.2.5"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 ParameterSchedulers = "0.4.2"
 Random = "1.10"
 Setfield = "1.1.1"
diff --git a/examples/NeuralODE/Project.toml b/examples/NeuralODE/Project.toml
@@ -21,7 +21,7 @@ LuxCUDA = "0.3"
 MLDatasets = "0.7"
 MLUtils = "0.4"
 OneHotArrays = "0.2.5"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 OrdinaryDiffEqTsit5 = "1"
 SciMLSensitivity = "7.63"
 Statistics = "1"
diff --git a/examples/PINN2DPDE/Project.toml b/examples/PINN2DPDE/Project.toml
@@ -18,7 +18,7 @@ Enzyme = "0.13"
 Lux = "1"
 MLUtils = "0.4.4"
 OnlineStats = "1.7.1"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Printf = "1.10"
 Random = "1.10"
 Reactant = "0.2.55"
diff --git a/examples/PolynomialFitting/Project.toml b/examples/PolynomialFitting/Project.toml
@@ -12,7 +12,7 @@ Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 ADTypes = "1.10"
 CairoMakie = "0.12, 0.13"
 Lux = "1"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Printf = "1.10"
 Random = "1.10"
 Reactant = "0.2.55"
diff --git a/examples/RealNVP/Project.toml b/examples/RealNVP/Project.toml
@@ -16,7 +16,7 @@ ConcreteStructs = "0.2.3"
 Enzyme = "0.13.35"
 Lux = "1.5"
 MLUtils = "0.4.5"
-Optimisers = "0.4.4"
+Optimisers = "0.4.6"
 Printf = "1.10"
 Random = "1.10"
 Reactant = "0.2.55"
diff --git a/examples/SimpleChains/Project.toml b/examples/SimpleChains/Project.toml
@@ -17,7 +17,7 @@ Lux = "1"
 MLDatasets = "0.7.14"
 MLUtils = "0.4"
 OneHotArrays = "0.2.5"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Random = "1"
 Reactant = "0.2.55"
 SimpleChains = "0.4.6"
diff --git a/examples/SimpleRNN/Project.toml b/examples/SimpleRNN/Project.toml
@@ -13,5 +13,5 @@ ADTypes = "1.10"
 JLD2 = "0.5"
 Lux = "1"
 MLUtils = "0.4"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Reactant = "0.2.55"
diff --git a/ext/LuxReactantExt/LuxReactantExt.jl b/ext/LuxReactantExt/LuxReactantExt.jl
@@ -2,7 +2,8 @@ module LuxReactantExt
 
 using Enzyme: Enzyme, Const
 using Optimisers: Optimisers
-using Reactant: Reactant, @compile, @code_hlo, AnyTracedRArray, TracedRArray, TracedRNumber
+using Reactant:
+    Reactant, @compile, @code_hlo, @jit, AnyTracedRArray, TracedRArray, TracedRNumber
 using ReactantCore: ReactantCore, @trace
 using Setfield: @set!
 using Static: True, False
diff --git a/ext/LuxReactantExt/patches.jl b/ext/LuxReactantExt/patches.jl
@@ -2,3 +2,8 @@ Utils.vec(x::AnyTracedRArray) = ReactantCore.materialize_traced_array(vec(x))
 
 # XXX: Use PoolDims once EnzymeJAX supports stablehlo.reduce_window adjoint
 Lux.calculate_pool_dims(g::Lux.GlobalPoolMode, ::TracedRArray) = g
+
+# Optimisers setup
+function Lux.ReactantCompatibleOptimisers.optimisers_setup_with_jit(opt, ps)
+    return @jit Optimisers.setup(opt, ps)
+end
diff --git a/src/helpers/optimizers.jl b/src/helpers/optimizers.jl
@@ -1,163 +1,23 @@
-# This is mostly an internal implementation detail that users shouldn't need to worry about.
-# We can remove this once https://github.com/FluxML/Optimisers.jl/issues/205 is resolved.
+# We use this module mostly as a placeholder for patches that should be merged into
+# Optimisers.jl for Reactant compatibility.
 module ReactantCompatibleOptimisers
 
-using ConcreteStructs: @concrete
-using Optimisers: Optimisers, AbstractRule
-using Setfield: Setfield, @set!
+using Optimisers: Optimisers
 
 using ..Lux: Lux, Utils
 
-abstract type ReactantCompatibleOptimisersRule <: AbstractRule end
-
-function make_reactant_compatible(opt::AbstractRule)
-    @warn "`make_reactant_compatible` is not defined for $(opt). Returning the original \
-           optimizer. This means adjusting learning rate and other parameters won't \
-           reflect in the generated MLIR." maxlog = 1
-    return opt
-end
-make_reactant_compatible(opt::ReactantCompatibleOptimisersRule) = opt
-
-function setfield_if_present(opt, field::Symbol, nt::NamedTuple)
-    if hasfield(typeof(nt), field)
-        return Setfield.set(
-            opt,
-            Setfield.PropertyLens{field}(),
-            convert(typeof(getproperty(opt, field)), getproperty(nt, field)),
-        )
-    end
-    return opt
-end
-
-function Optimisers._adjust(opt::ReactantCompatibleOptimisersRule, nt::NamedTuple)
-    for field in fieldnames(typeof(opt))
-        opt = setfield_if_present(opt, field, nt)
-    end
-    return opt
-end
-
-# OptimiserChain
 function make_reactant_compatible(opt::Optimisers.OptimiserChain)
-    return Optimisers.OptimiserChain(make_reactant_compatible.(opt.opts))
-end
-
-# Descent
-@concrete struct ReactantDescent <: ReactantCompatibleOptimisersRule
-    eta
-end
-
-function make_reactant_compatible(opt::Optimisers.Descent)
-    return ReactantDescent(Utils.to_rarray(opt.eta; track_numbers=true))
-end
-
-Optimisers.init(::ReactantDescent, ::AbstractArray) = nothing
-
-function Optimisers.apply!(opt::ReactantDescent, state, x::AbstractArray{T}, dx) where {T}
-    η = T(opt.eta)
-    return state, @. dx * η
-end
-
-# Momentum
-@concrete struct ReactantMomentum <: ReactantCompatibleOptimisersRule
-    eta
-    rho
-end
-
-function make_reactant_compatible(opt::Optimisers.Momentum)
-    return ReactantMomentum(
-        Utils.to_rarray(opt.eta; track_numbers=true),
-        Utils.to_rarray(opt.rho; track_numbers=true),
-    )
-end
-
-function Optimisers.init(::ReactantMomentum, x::AbstractArray)
-    return Optimisers.init(Optimisers.Momentum(0.0, 0.0), x)
-end
-
-function Optimisers.apply!(opt::ReactantMomentum, mvel, ::AbstractArray{T}, dx) where {T}
-    η, ρ = T(opt.eta), T(opt.rho)
-    @. mvel = ρ * mvel + η * dx
-    return mvel, mvel
-end
-
-# Adam
-@concrete struct ReactantAdam <: ReactantCompatibleOptimisersRule
-    eta
-    beta
-    epsilon
+    return Optimisers.OptimiserChain(make_reactant_compatible.(opt.opts)...)
 end
 
-function make_reactant_compatible(opt::Optimisers.Adam)
-    return ReactantAdam(
-        Utils.to_rarray(opt.eta; track_numbers=true),
-        Utils.to_rarray(opt.beta; track_numbers=true),
-        Utils.to_rarray(opt.epsilon; track_numbers=true),
-    )
+function make_reactant_compatible(opt::Optimisers.AbstractRule)
+    return Utils.to_rarray(opt; track_numbers=AbstractFloat)
 end
 
-function Optimisers.init(opt::ReactantAdam, x::AbstractArray{T}) where {T}
-    return (
-        zero(x),
-        zero(x),
-        (Utils.promote_to(T, opt.beta[1]), Utils.promote_to(T, opt.beta[2])),
-    )
+function make_reactant_compatible(opt::Optimisers.AccumGrad)
+    return Utils.to_rarray(opt; track_numbers=Integer)
 end
 
-function Optimisers.apply!(o::ReactantAdam, state, ::AbstractArray{T}, dx) where {T}
-    η, β, ϵ = T(o.eta), T.(o.beta), T(o.epsilon) # XXX: See Optimisers._eps
-    mt, vt, βt = state
-
-    mt = @. β[1] * mt + (1 - β[1]) * dx
-    vt = @. β[2] * vt + (1 - β[2]) * abs2(dx)
-    dx′ = @. mt / (1 - βt[1]) / (sqrt(vt / (1 - βt[2])) + ϵ) * η
-
-    return (mt, vt, βt .* β), dx′
-end
-
-# AdamW
-@concrete struct ReactantAdamW <: ReactantCompatibleOptimisersRule
-    eta
-    beta
-    lambda
-    epsilon
-    couple::Bool
-end
-
-function make_reactant_compatible(opt::Optimisers.AdamW)
-    return ReactantAdamW(
-        Utils.to_rarray(opt.eta; track_numbers=true),
-        Utils.to_rarray(opt.beta; track_numbers=true),
-        Utils.to_rarray(opt.lambda; track_numbers=true),
-        Utils.to_rarray(opt.epsilon; track_numbers=true),
-        opt.couple,
-    )
-end
-
-function Optimisers.init(opt::ReactantAdamW, x::AbstractArray{T}) where {T}
-    return (
-        zero(x),
-        zero(x),
-        (Utils.promote_to(T, opt.beta[1]), Utils.promote_to(T, opt.beta[2])),
-    )
-end
-
-function Optimisers.apply!(o::ReactantAdamW, state, x::AbstractArray{T}, dx) where {T}
-    η, β, ϵ, λ = T(o.eta), T.(o.beta), T(o.epsilon), T(o.lambda) # XXX: See Optimisers._eps
-    mt, vt, βt = state
-
-    # standard Adam update with learning rate eta=1
-    mt = @. β[1] * mt + (1 - β[1]) * dx
-    vt = @. β[2] * vt + (1 - β[2]) * abs2(dx)
-    dx′ = @. mt / (1 - βt[1]) / (sqrt(vt / (1 - βt[2])) + ϵ)
-
-    # apply learning rate and weight decay
-    if o.couple
-        dx′′ = @. η * (dx′ + λ * x)
-    else
-        dx′′ = @. η * dx′ + λ * x
-    end
-
-    return (mt, vt, βt .* β), dx′′
-end
+function optimisers_setup_with_jit end
 
 end
diff --git a/src/helpers/training.jl b/src/helpers/training.jl
@@ -63,10 +63,11 @@ Constructor for [`TrainState`](@ref).
 [`TrainState`](@ref) object.
 """
 function TrainState(model::AbstractLuxLayer, ps, st, optimizer::Optimisers.AbstractRule)
-    st_opt = if get_device_type(ps) <: ReactantDevice
-        Optimisers.setup(ReactantCompatibleOptimisers.make_reactant_compatible(optimizer), ps)
+    if get_device_type(ps) <: ReactantDevice
+        optimizer = ReactantCompatibleOptimisers.make_reactant_compatible(optimizer)
+        st_opt = ReactantCompatibleOptimisers.optimisers_setup_with_jit(optimizer, ps)
     else
-        Optimisers.setup(optimizer, ps)
+        st_opt = Optimisers.setup(optimizer, ps)
     end
     return TrainState(nothing, nothing, model, ps, st, optimizer, st_opt, 0)
 end
diff --git a/test/Project.toml b/test/Project.toml
@@ -65,7 +65,7 @@ MLUtils = "0.4.3"
 NNlib = "0.9.26"
 Octavian = "0.3.28"
 OneHotArrays = "0.2.5"
-Optimisers = "0.4.1"
+Optimisers = "0.4.6"
 Pkg = "1.10"
 Preferences = "1.4.3"
 Random = "1.10"
diff --git a/test/layers/embedding_tests.jl b/test/layers/embedding_tests.jl
@@ -116,7 +116,7 @@ end
     end
 end
 
-@testitem "Reactant SinusoidalPositionalEmbedding" setup = [
+@testitem "Reactant: SinusoidalPositionalEmbedding" setup = [
     SharedTestSetup, SharedReactantLayersTestSetup
 ] tags = [:reactant] begin
     using Reactant, Lux
@@ -234,7 +234,7 @@ end
     end
 end
 
-@testitem "Reactant RotaryPositionalEmbedding" setup = [
+@testitem "Reactant: RotaryPositionalEmbedding" setup = [
     SharedTestSetup, SharedReactantLayersTestSetup
 ] tags = [:reactant] begin
     using Reactant, Lux
diff --git a/test/reactant/training_tests.jl b/test/reactant/training_tests.jl
