Agata Skorupka

Hi @CarloLucibello I just wanted to check if it's ok to add `Graphs` to deps instead of weakdeps, or do you see any better solutions instead. I need it to...

Now dependent on & blocked by https://github.com/CarloLucibello/GraphNeuralNetworks.jl/pull/499

@CarloLucibello ready for another look

hi @CarloLucibello as per our discussion I've set up a working example for Flux and then used this example to extend to `GraphNeuralNetworks`. here is my code: ```julia using Flux,...

Thank you for the code @CarloLucibello. I've tried it after loading `NNlibEnzymeCore` extension but I'm not sure if the `Enzyme.autodiff` function call in the above example snippet is correct as...

Maybe let me paste here whole stacktrace. ```julia Stacktrace: [1] getproperty @ ./Base.jl:32 [inlined] [2] unwrap_unionall @ ./essentials.jl:379 [inlined] [3] fieldnames @ ./reflection.jl:169 [inlined] [4] augmented_julia_fieldnames_8170wrap @ ./reflection.jl:0 [5] macro...

Interestingly, this is first thing in the stacktrace causing the issue. ```julia function (l::GraphConv)(g::AbstractGNNGraph, x) check_num_nodes(g, x) xj, xi = expand_srcdst(g, x) m = propagate(copy_xj, g, l.aggr, xj = xj)...

Also linking some related issues/PRs for future testing purposes. [Flux.jl PR #2392](https://github.com/FluxML/Flux.jl/pull/2392) [EnzymeAD issue #805](https://github.com/EnzymeAD/Enzyme.jl/issues/805)

@CarloLucibello I was able to move it forward according to your suggestions and MPI example with training works :tada: (I still need to do some cleanup tho) details are in...

Update: both MPI and NCCL work. please run `mpiexecjl --project=@. -n 3 julia distributed_NCCL.jl` or `mpiexecjl --project=@. -n 3 julia distributed_MPI.jl` respectively from your terminal. still in the draft state,...