Can't get Flux (v0.10.3) working for Custom Loss function

[compleathorseplayer]

I have a very simple problem I am trying to solve using Flux 0.10.3 using methods
which worked with the 0.9.0 version of Flux but which fail now.

xin is a 39K by 50 input matrix (rows are observations, columns are variables)
yout is a 39K by 1 outcome variable.

I run it including a step to verify that the loss function works for the dataset:

julia> model=Chain(Dense(50,1,sigmoid))
Chain(Dense(50, 1, σ))

julia> loss1(x,y)=mean(-(model(x).*y)')
loss1 (generic function with 2 methods)

julia> loss1(x::Tuple)=loss1(x[1],x[2])
loss1 (generic function with 2 methods)

julia> evalcb = () -> @show(loss1((xin',yout'))) #note transposition for compatibility
#63 (generic function with 1 method)

julia> dataset = [(xin',yout'),(xin',yout')];

julia> println(loss1(dataset[1]))
0.0016364615651065868

All of the above seems fine, but when I try to run Flux:

julia> Flux.train!(loss1,params(model),dataset,ADAM(),cb = throttle(evalcb, 5))
ERROR: MethodError: no method matching *(::Dense{typeof(σ),Array{Float32,2},Array{Float32,1}}, ::Float64)
Closest...

Perhaps this is only an issue having to do with change of convention
and calling syntax with the upgrade?

In any case, I would be very grateful for any help.

Actually, it has been noted elsewhere that the more recent (0.10.x) versions of Flux are not
backwards compatible with previous, but I cannot seem to find
any documentation explaining the changes in usage required on the user side.

[I had previously resolved such issues by downgrading to the 0.9.0 version
but had inadvertently included Flux in a general package update to resolve another issue
recently and now find that downgrading to 0.9.0 is broken so I am 'stuck' with the current one]

[ToucheSir]

I just tried this MWE on 0.12.4 and didn't encounter any issues. Feel free to comment with an updated version (including dummy data) if you're still running into this @compleathorseplayer, but otherwise I think it's safe to close.