Vedant Puri

Error with `Zygote.gradient` for `foldl, sum`

2

MWE ```julia using Zygote, LinearAlgebra N = 4 u0 = rand(N) ps = rand(N) mats = (rand(N,N), rand(N,N),) # (A, B,) nums = (rand(), rand(),) # (α, β,) loss_m =...

Fix constructor for GPU arrays

4

```julia julia> c = ComponentArray(a=CUDA.ones(2,3), b=CUDA.ones(4,5)) ``` ```julia ERROR: CuArray only supports element types that are stored inline Stacktrace: [1] error(s::String) @ Base ./error.jl:35 [2] CuArray{Vector{Any}, 2, CUDA.Mem.DeviceBuffer}(#unused#::UndefInitializer, dims::Tuple{Int 64,...

ComponentArray with scalars on GPU

2

```julia julia> using ComponentArrays, Lux julia> c = ComponentArray(a=ones(2,3), b=0.0) |> gpu ComponentVector{Float32, CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, Tuple{Axis{(a = ViewAxis(1: 6, ShapedAxis((2, 3), NamedTuple())), b = 7)}}}(a = Float32[1.0 1.0 1.0;...

Error when `prob.u0 isa ComponentArray` with GPU data

4

This line is erroring when I form `u0` as follows: ```julia u0 = ComponentArray(; vx=rand(N,10), vy=rand(N,10)) |> gpu ``` https://github.com/SciML/SciMLSensitivity.jl/blob/6b23999c640f2f08966c9545c1e383aaba9d154c/src/concrete_solve.jl#L348-L350 MWE: https://github.com/vpuri3/PDEInterfaces.jl/blob/1b5c1261ddef84bec9ef0f1f26561b18e68f9a4b/examples/ad/ca.jl#L1 stacktrace: ```julia julia> include("examples/ad/ca.jl") (10+ results) [934/1822] fwd...

`EnzymeVJP` failing for `SplitFunction`s

2

this line fails for `SplitODEProblem`s as `SplitFunction`s don't have the field `f`. https://github.com/SciML/DiffEqSensitivity.jl/blob/f06856ff850dcb6e566bddac133902127ac35660/src/adjoint_common.jl#L187 Error: ```julia julia> include("examples/opt.jl") fwd 1340.6743837932922 bwd ERROR: LoadError: type SplitFunction has no field f Stacktrace: [1]...

sundials doesn't GPU

4

```julia using CUDA, Sundials N = 4 f = (u, p, t) -> zero(u) u0 = CUDA.rand(N) ts = (0f0, 1f0) prob = ODEProblem(f, u0, ts) alg = CVODE_BDF() sol...

`substitute` didn't work on `Vector`

1

```julia using Symbolics julia> @variables x[1:10]; substitute(sin.(π .* x), Dict(x => rand(10))) (broadcast(sin, broadcast(*, π, x)))[1:10] ``` @shashi

remove functions that are predefined with `abstractarray` interface

6

unless there's performance diff

`mul!(::Matrix, ::ArrayPartition, ::Adjoint(ArrayPartition))`

4

https://github.com/JuliaNLSolvers/Optim.jl/pull/972 MWE ```julia using RecursiveArrayTools, LinearAlgebra u = ArrayPartition(rand(2), rand(1)) v = copy(u) H = u * v' # works mul!(H, u, v') # error ``` ```julia julia> mul!(H,u,v') ERROR:...

should use broadcast for vector space calculations?

1

https://github.com/SciML/RecursiveArrayTools.jl/blob/569a07e9f0ecaacef167ee52e8101122a25c9b4a/test/partitions_test.jl#L78-L82 What's the use case for the vector space interface? using broadcast, rather than overloading `+,-` would be consistent with AbstractArray interface. ```julia julia> 2 + rand(2) ERROR: MethodError: no...