Error with Flux GRU layer

[jgreener64]

I am on Julia 1.10.0, Enzyme main (ba58bcb) and Flux 0.14.10.

using Enzyme, Flux
loss(model, x) = sum(model(x))
model = GRU(3 => 5)
x = randn(Float32, 3, 10)
d_model = Flux.fmap(model) do x
    x isa Array ? zero(x) : x
end
Enzyme.autodiff(Reverse, loss, Active, Duplicated(model, d_model), Const(x))

See the error.txt and the printall_error.txt.

[wsmoses]

Are you able to minimize this by chance (and potentially open issues for the other problems you found)?

[jgreener64]

I think this is the minimal error. On Enzyme main (5e4e2ef) and Julia 1.10.2:

using Enzyme

loss(gxs, ghs) = sum(gxs .+ ghs)

gxs = rand(5, 10)
ghs = rand(5)

# Comment out the next two lines and it works
gxs = view(gxs, 1:5, 1:10)
ghs = view(ghs, 1:5)

loss(gxs, ghs)

autodiff(
    Reverse,
    loss,
    Active,
    Duplicated(gxs, gxs),
    Duplicated(ghs, ghs),
)

The error.txt and printall_error.txt are attached.

[wsmoses]

@jgreener64 @CarloLucibello the last example now works on main, going to close this. Please reopen if it persists.