Valentin Churavy
Valentin Churavy
@wsmoses #537 did implement this right?
@ayeankit feel free to work on this, but I have found it best over the years to not assign work to new members (see the term cookie-licking). Currently no-one is...
Rebase after https://github.com/EnzymeAD/Enzyme.jl/pull/1104 ? there still seems to be some useful things in here.
I do think the name of the macro needs to change. It has nothing to do with `const`. IIUC it's `lhs::typeof(rhs) = rhs` and we call those typed globals. So...
@wsmoses ``` %"'ipc.i" = addrspacecast [1 x {} addrspace(10)*]* %5 to [1 x {} addrspace(10)*] addrspace(11)*, !dbg !18 %12 = insertvalue [2 x [1 x {} addrspace(10)*] addrspace(11)*] undef, [1...
While Enzyme shouldn't fail in the way you are seeing right now. Currently we do not support CUDA on the host side, that is ongoing work,
Why are you converting the Julia array to an `mx.NDArray` before passing it to `ArrayDataProvider`?
I am a bit worried that mx.Reshape get's confused between Julias ordering of dimensions: (W, H, Channel, Sample) and mxnet ordering (C, S, H, W). I would recommend doing the...
MXNet.jl will handle the transformation from Julia order to C/C++ order for you. `dataP =ArrayDataProvider(:data => reshape(data, (75, 75, 1, 66)))`
Your label is between 0 and 1? You can use `mx.MultiACE(2)` to see the logloss associated with each class. From a purely network design perspective, I would use `mx.ADAM` instead...