Michael Abbott
Michael Abbott
Yes you can, xref https://github.com/JuliaLang/julia/pull/56662 I think they are from inside Enzyme somewhere... ok I found where I looked, and the pointer was this line: https://github.com/EnzymeAD/Enzyme.jl/blob/main/src/absint.jl#L98 on 23 November, at...
The present recursion roughly assumes the existence of the default constructor. This instead assumes the existence of a keyword constructor, which is much less common: ```julia julia> RNN( cell =...
This code is not so far from everything else in this package, and now that it has extensions (thanks to #151) it's fine to have one more for this package....
The error I see on 1.11 looks a little different: ``` julia> using Flux: Flux julia> using Enzyme: Enzyme julia> using CUDA: CUDA julia> using cuDNN: cuDNN julia> function reproducer(;...
Ok. IDK whether my error is the same as that issue, or an independent problem. I tried both versions of Julia just because this thread didn't specify.
I like friendly errors, but I think this is going to produce ambiguities with Flux, method 7: ``` julia> using Flux, Optimisers julia> methods(Optimisers.update!) # 8 methods for generic function...
I don't like the auto-broadcast either but here we are. The built-in opt-out is this function -- which perhaps Reactant needs to know how to handle anyway? ```julia julia> Base.Broadcast.broadcast_preserving_zero_d(sin,...
But what this function is fed is some special fake 0D array which Reactant invents? My hope is that it can also be made to understand that `broadcast_preserving_zero_d(sin, x::ZeroDimTrackedArray):: ZeroDimTrackedArray...
This looks right to me, at first glance. I see that https://github.com/JuliaDiff/DualNumbers.jl/pull/104 recently adopted the other convention. But I have never used this package and am unsure who does, or...
One implementation which makes few assumptions about the data/gradient format is https://github.com/Jutho/OptimKit.jl However, the mismatch is that it wants to control when to call the function/model, while in Optimisers.jl you...