Tor Erlend Fjelde
Tor Erlend Fjelde
@yebai should also probably have a look at this :+1:
Softmax isn't bijective. The one we have now is (maps from d to d-1 dimensional)
Unfortately doesn't seem to address the issue in https://github.com/TuringLang/DynamicPPL.jl/issues/643 nor the failing one herehttps://github.com/torfjelde/EnzymeCon2024-Turing.jl-benchmarks/blob/main/src/models/fails/satellite-ssm-matrix.jl 😕
I think I'll have to defer to @wsmoses for your comments @devmotion . From convo with him, I sort of got the impression that we wanted to avoid `typeof` and...
Converted to DRAFT though because it wasn't my intention to have this be merged into master atm; just for Enzyme testing
> I recommended Tor see if propagating the use of the actual runtime type would resolve it here, but if not Tor you're going to have to see what other...
@mhauru @wsmoses
Yeah I would be very surprised if Zygote.jl worked with threads like this. You should probably look into something like Transducers.jl or something that defines a parallel way to perform...
From the snippet you shared, it doesn't seem like you're using the same RNG? If so, that could just be the cause of it. Second thought, though this seems unlikely...
> I had a rather quick look. There's many things that should not be here (but maybe some them could be moved or fixed somewhere else). Specializations for LazyArrays could...