remove some v0.13 deprecations (#2493)

* remove v0.13 deprecations

.buildkite/pipeline.yml

@@ -46,7 +46,7 @@ steps:
           using Pkg
           Pkg.resolve()'
     commands: |
-      printf "[MLDataDevices]\ngpu_backend = \"Metal\"\n" > LocalPreferences.toml
+      printf "[Flux]\ngpu_backend = \"Metal\"\n" > LocalPreferences.toml
 
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
@@ -63,7 +63,7 @@ steps:
   - label: "AMD GPU with Julia 1"
     plugins:
       - JuliaCI/julia#v1:
-          version: "1.10"
+          version: "1"
       - JuliaCI/julia-test#v1:
       - JuliaCI/julia-coverage#v1:
           dirs:
@@ -74,7 +74,7 @@ steps:
       rocm: "*"
       rocmgpu: "*"
     commands: |
-      printf "[MLDataDevices]\ngpu_backend = \"AMDGPU\"\n" > LocalPreferences.toml
+      printf "[Flux]\ngpu_backend = \"AMDGPU\"\n" > LocalPreferences.toml
     timeout_in_minutes: 60
     env:
       JULIA_AMDGPU_CORE_MUST_LOAD: "1"

.github/workflows/ci.yml

@@ -9,16 +9,23 @@ on:
       - master
     tags: '*'
 
+# needed for julia-actions/cache to delete old caches
+permissions:
+  actions: write
+  contents: read
+
 jobs:
   test:
-    name: Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }}
+    name: Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }}
     runs-on: ${{ matrix.os }}
     strategy:
       fail-fast: false
       matrix:
         version:
-          # - '1.9' # Uncomment when 1.10 is out. Replace this with the minimum Julia version that your package supports. 
-          - '1'
+          - '1.10' # Replace this with the minimum Julia version that your package supports. 
+          - '1'    # latest stable 1.x releas
+          # - 'pre'  # latest stable prerelease
+          - 'nightly' # latest nightly release
         os: [ubuntu-latest]
         arch: [x64]
         include:
@@ -29,7 +36,7 @@ jobs:
             version: '1'
             arch: aarch64
     steps:
-      - uses: actions/checkout@v4.2.1
+      - uses: actions/checkout@v4
       - uses: julia-actions/setup-julia@v2
         with:
           version: ${{ matrix.version }}
@@ -47,17 +54,16 @@ jobs:
       - uses: julia-actions/julia-buildpkg@v1
       - name: "Run test without coverage report"
         uses: julia-actions/julia-runtest@v1
-        if: ${{ !contains(fromJson('["1", "1.9"]'), matrix.version) || matrix.os != 'ubuntu-latest' }}
+        if: matrix.version != '1' || matrix.os != 'ubuntu-latest'
         with:
           coverage: false
-
       - name: "Run test with coverage report"
         uses: julia-actions/julia-runtest@v1
-        if: contains(fromJson('["1", "1.9"]'), matrix.version) && matrix.os == 'ubuntu-latest'
+        if: matrix.version == '1' && matrix.os == 'ubuntu-latest'
       - uses: julia-actions/julia-processcoverage@v1
-        if: contains(fromJson('["1", "1.9"]'), matrix.version) && matrix.os == 'ubuntu-latest'
+        if: matrix.version == '1' && matrix.os == 'ubuntu-latest'
       - uses: codecov/codecov-action@v4
-        if: contains(fromJson('["1", "1.9"]'), matrix.version) && matrix.os == 'ubuntu-latest'
+        if: matrix.version == '1' && matrix.os == 'ubuntu-latest'
         with:
           file: lcov.info
 

Project.toml

@@ -67,4 +67,4 @@ SpecialFunctions = "2.1.2"
 Statistics = "1"
 Zygote = "0.6.67"
 cuDNN = "1"
-julia = "1.9"
+julia = "1.10"

ext/FluxAMDGPUExt/FluxAMDGPUExt.jl

@@ -6,7 +6,7 @@ import Flux
 import Flux: FluxCPUAdaptor, FluxAMDGPUAdaptor, _amd, adapt_storage, fmap
 import Flux: DenseConvDims, Conv, ConvTranspose, conv, conv_reshape_bias
 import NNlib
-
+using MLDataDevices: MLDataDevices
 using AMDGPU
 using Adapt
 using Random

ext/FluxCUDAExt/FluxCUDAExt.jl

@@ -10,6 +10,7 @@ using ChainRulesCore
 using Random
 using Adapt
 import Adapt: adapt_storage
+using MLDataDevices: MLDataDevices
 
 
 const USE_CUDA = Ref{Union{Nothing, Bool}}(nothing)

ext/FluxCUDAExt/functor.jl

@@ -57,5 +57,5 @@ function _cuda(id::Union{Nothing, Int}, x)
 end
 
 function Flux._get_device(::Val{:CUDA}, id::Int)
-    return MLDataUtils.gpu_device(id+1, force=true)
+    return MLDataDevices.gpu_device(id+1, force=true)
 end

ext/FluxEnzymeExt/FluxEnzymeExt.jl

@@ -16,10 +16,6 @@ _applyloss(loss, model, d...) = loss(model, d...)
 
 EnzymeRules.inactive(::typeof(Flux.Losses._check_sizes), args...) = true
 
-using Flux: _old_to_new  # from src/deprecations.jl
-train!(loss, model::Duplicated, data, opt::Optimise.AbstractOptimiser; cb=nothing) =
-  train!(loss, model, data, _old_to_new(opt); cb)
-
 function train!(loss, model::Duplicated, data, rule::Optimisers.AbstractRule; cb = nothing)
   train!(loss, model, data, _rule_to_state(model, rule); cb)
 end

ext/FluxMetalExt/FluxMetalExt.jl

@@ -4,7 +4,7 @@ import Flux
 import Flux: FluxCPUAdaptor, FluxMetalAdaptor, _metal, _isleaf, adapt_storage, fmap
 import NNlib
 using ChainRulesCore
-
+using MLDataDevices: MLDataDevices
 using Metal
 using Adapt
 using Random

ext/FluxMetalExt/functor.jl

@@ -35,6 +35,6 @@ end
 
 function Flux._get_device(::Val{:Metal}, id::Int)
     @assert id == 0 "Metal backend only supports one device at the moment"
-    return MLDataDevices.gpu_device()
+    return MLDataDevices.gpu_device(force=true)
 end
 

src/Flux.jl

@@ -18,13 +18,14 @@ using Zygote: Params, @adjoint, gradient, pullback
 using Zygote.ForwardDiff: value
 export gradient
 
-@reexport using MLDataDevices: MLDataDevices, gpu_backend!, supported_gpu_backends, reset_gpu_device!,
+@reexport using MLDataDevices: MLDataDevices, supported_gpu_backends, reset_gpu_device!,
                     default_device_rng,
                     gpu_device, cpu_device, xla_device,
                     CPUDevice,
                     CUDADevice, AMDGPUDevice, MetalDevice, oneAPIDevice,
                     XLADevice,
                     # get_device, # we define get_device here for retrocompatibility
+                    # gpu_backend!, # have to define here due to https://github.com/JuliaPackaging/Preferences.jl/issues/39
                     get_device_type,
                     DeviceIterator
 
@@ -104,7 +105,7 @@ include("losses/Losses.jl")
 using .Losses
 
 include("devices.jl")
-export get_device
+export get_device, gpu_backend!
 
 # Distributed Training
 include("distributed/backend.jl")

src/deprecations.jl

@@ -1,47 +1,6 @@
 
 # v0.13 deprecations
 
-function Broadcast.broadcasted(f::Recur, args...)
-  # This had an explicit @adjoint rule, calling Zygote.∇map(__context__, f, args...), until v0.12
-  Base.depwarn("""Broadcasting is not safe to use with RNNs, as it does not guarantee an iteration order.
-    Re-writing this as a comprehension would be better.""", :broadcasted)
-  map(f, args...)  # map isn't really safe either, but 
-end
-
-@deprecate frequencies(xs) group_counts(xs)
-
-struct Zeros
-  function Zeros()
-    Base.depwarn("Flux.Zeros is no more, has ceased to be, is bereft of life, is an ex-boondoggle... please use bias=false instead", :Zeros)
-    false
-  end
-end
-Zeros(args...) = Zeros()  # was used both Dense(10, 2, initb = Zeros) and Dense(rand(2,10), Zeros())
-
-function Optimise.update!(x::AbstractArray, x̄)
-  Base.depwarn("`Flux.Optimise.update!(x, x̄)` was not used internally and has been removed. Please write `x .-= x̄` instead.", :update!)
-  x .-= x̄
-end
-
-function Diagonal(size::Integer...; kw...)
-  Base.depwarn("Flux.Diagonal is now Flux.Scale, and also allows an activation function.", :Diagonal)
-  Scale(size...; kw...)
-end
-function Diagonal(size::Tuple; kw...)
-  Base.depwarn("Flux.Diagonal is now Flux.Scale, and also allows an activation function.", :Diagonal)
-  Scale(size...; kw...)
-end
-
-# Deprecate this eventually once saving models w/o structure is no more
-function loadparams!(m, xs)
-  Base.depwarn("loadparams! will be deprecated eventually. Use loadmodel! instead.", :loadparams!)
-  for (p, x) in zip(params(m), xs)
-    size(p) == size(x) ||
-      error("Expected param size $(size(p)), got $(size(x))")
-    copyto!(p, x)
-  end
-end
-
 # Channel notation: Changed to match Conv, but very softly deprecated!
 # Perhaps change to @deprecate for v0.15, but there is no plan to remove these.
 Dense(in::Integer, out::Integer, σ = identity; kw...) =
@@ -56,32 +15,6 @@ LSTMCell(in::Integer, out::Integer; kw...) = LSTMCell(in => out; kw...)
 GRUCell(in::Integer, out::Integer; kw...) = GRUCell(in => out; kw...)
 GRUv3Cell(in::Integer, out::Integer; kw...) = GRUv3Cell(in => out; kw...)
 
-# Optimisers with old naming convention
-Base.@deprecate_binding ADAM Adam
-Base.@deprecate_binding NADAM NAdam
-Base.@deprecate_binding ADAMW AdamW
-Base.@deprecate_binding RADAM RAdam
-Base.@deprecate_binding OADAM OAdam
-Base.@deprecate_binding ADAGrad AdaGrad
-Base.@deprecate_binding ADADelta AdaDelta
-
-# Remove sub-module Data, while making sure Flux.Data.DataLoader keeps working
-Base.@deprecate_binding Data Flux false "Sub-module Flux.Data has been removed. The only thing it contained may be accessed as Flux.DataLoader"
-
-@deprecate paramtype(T,m) _paramtype(T,m) false  # internal method, renamed to make this clear
-
-@deprecate rng_from_array() Random.default_rng()
-
-function istraining()
-  Base.depwarn("Flux.istraining() is deprecated, use NNlib.within_gradient(x) instead", :istraining)
-  false
-end
-ChainRulesCore.rrule(::typeof(istraining)) = true, _ -> (NoTangent(),)
-
-function _isactive(m)
-  Base.depwarn("_isactive(m) is deprecated, use _isactive(m,x)", :_isactive, force=true)
-  _isactive(m, 1:0)
-end
 
 #=
   # Valid method in Optimise, old implicit style, is:
@@ -110,7 +43,6 @@ train!(loss, ps::Params, data, opt::Optimisers.AbstractRule; cb=nothing) = error
 train!(loss, model, data, opt::Optimise.AbstractOptimiser; cb=nothing) =
   train!(loss, model, data, _old_to_new(opt); cb)
 
-
 # Next, to use the new `setup` with the still-exported old-style `Adam` etc:
 import .Train: setup
 setup(rule::Optimise.AbstractOptimiser, model) = setup(_old_to_new(rule), model)
@@ -179,33 +111,6 @@ function update!(opt::Optimise.AbstractOptimiser, ::Params, grads::Union{Tuple,
     """)
 end
 
-""" 
-    trainmode!(m, active)
-
-!!! warning
-    This two-argument method is deprecated.
-
-Possible values of  `active` are:
-- `true` for training, or 
-- `false` for testing, same as [`testmode!`](@ref)`(m)`
-- `:auto` or `nothing` for Flux to detect training automatically.
-"""
-function trainmode!(m, active::Bool)
-  Base.depwarn("trainmode!(m, active::Bool) is deprecated", :trainmode)
-  testmode!(m, !active)
-end
-
-# Greek-letter keywords deprecated in Flux 0.13
-# Arguments (old => new, :function, "β" => "beta")
-function _greek_ascii_depwarn(βbeta::Pair, func = :loss, names = "" => "")
-  Base.depwarn(LazyString("function ", func, " no longer accepts greek-letter keyword ", names.first, """
-    please use ascii """, names.second, " instead"), func)
-  βbeta.first
-end
-_greek_ascii_depwarn(βbeta::Pair{Nothing}, _...) = βbeta.second
-
-ChainRulesCore.@non_differentiable _greek_ascii_depwarn(::Any...)
-
 
 # v0.14 deprecations
 @deprecate default_rng_value() Random.default_rng()

src/functor.jl

@@ -189,11 +189,19 @@ _isleaf(::Union{Transpose, Adjoint, PermutedDimsArray}) = false
 
 _isleaf(::AbstractRNG) = true
 
+# Remove when 
+# https://github.com/JuliaPackaging/Preferences.jl/issues/39
+# is resolved
+function gpu_backend!(backend::String)
+    @set_preferences!("gpu_backend" => backend)
+    MLDataDevices.gpu_backend!(backend)
+end
+
 # the order below is important
 const GPU_BACKENDS = ("CUDA", "AMDGPU", "Metal", "CPU")
-const GPU_BACKEND_ORDER = Dict(collect(zip(GPU_BACKENDS, 1:length(GPU_BACKENDS))))
-const GPU_BACKEND = load_preference(MLDataDevices, "gpu_backend", "CUDA")
-
+# const GPU_BACKEND = load_preference(MLDataDevices, "gpu_backend", "CUDA")
+# https://github.com/JuliaPackaging/Preferences.jl/issues/39
+const GPU_BACKEND = @load_preference("gpu_backend", "CUDA")
 
 """
     gpu(m)