When trying to use Lux with CUDA/cuDNN (to train with Zygote GPU), WeightInitializers fails to precompile:
julia> using WeightInitializers
[ Info: Precompiling CUDAExt [0d0e9cde-cfc5-56bb-a277-295c1cc0b950](cache misses: wrong dep version loaded (2), incompatible header (2))
Info Given CUDAExt was explicitly requested, output will be shown live
ERROR: LoadError: UndefVarError: `RNG` not defined in `cuRAND`
Suggestion: check for spelling errors or missing imports.
Hint: a global variable of this name also exists in CUDA.
This happens with with latest CUDA releases (v6); perhaps some incompatibility with the CUDA rework into sub-packages?
Pkg status:
Status `~/evovest/NeuroTabModels.jl/adhoc/Project.toml`
[052768ef] CUDA v6.1.0
[b2108857] Lux v1.31.4
[d49dbf32] WeightInitializers v1.3.3
[02a925ec] cuDNN v6.1.0
When trying to use Lux with CUDA/cuDNN (to train with Zygote GPU),
WeightInitializersfails to precompile:This happens with with latest CUDA releases (v6); perhaps some incompatibility with the CUDA rework into sub-packages?
Pkg status: