Open nikopj opened 2 weeks ago
There are 3 possible solutions:
LuxCUDADevice
(https://github.com/LuxDL/Lux.jl/blob/68238456942b016f03dc40d47d7f312a7587c7a5/ext/LuxMPINCCLExt.jl#L38-L42), do the reim
and complex
if the buffer type is ::AbstractArray{<:Complex}
.
The NCCL backend in distributed utils does not support complex values see issue. Can we add a conveinent wrapper in the NCCLEXT to support broadcast, reduce, etc., likely via using
reim
andcomplex
? I'm happy to get started on it but would like some feedback on a preferred location and so on.