Looks like a couple of workers might have gotten killed – can you try with fewer of them (see the message at the start of the test run)?
The NVML failure is not a problem.
So apart from the workers getting killed, almost everything seems to be working just fine. It has been a while since I tested WSL myself, but there’s no reason to expect CUDA.jl wouldn’t work on it.