Please support Lux.jl

[schlichtanders]

The SciML ecosystem strongly prefers Lux.jl (see https://github.com/SciML/DiffEqFlux.jl README)

it would be nice if Transformers.jl could also be put on top of Lux.jl

[chengchingwen]

What does it need to make a layer Lux compatible?

[tobefreeman]

Why do SciML people favour Lux over Flux? Also, what is your evidence for this, given that Flux is so much more popular overall?

…

On Fri, Mar 1, 2024 at 3:24 PM Peter ***@***.***> wrote: What does it need to make a layer Lux compatible? — Reply to this email directly, view it on GitHub <#170 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAXOFIXXFIOGE7AEKKUHYS3YWCFSNAVCNFSM6AAAAABD4AMT6OVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNZTGI4TKOJZG4> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- dr tobe freeman - +41 79 377 8595 wordup development AG <https://wordupdevelopment.com/>

[schlichtanders]

DiffEqFlux.jl is the central package for combining deep learning with differential equation. Take a look there - they even changed the official name to DiffEq For Lux.

Flux ist just older, hence there may be more packages referring to it. But actually for me Lux is already the go to package for deep learning in Julia

[schlichtanders]

I myself asked once about the advantages of Lux.jl vs Flux.jl for DiffEqFlux.jl

here the important answer SciML/DiffEqFlux.jl#815 (comment)

and meanwhile the issue got closed by moving to Lux.jl completely

[chengchingwen]

I'm totally fine with supporting Lux.jl. But IIUC Lux.transform can wrap a Flux layer into a Lux compatible layer already?

[schlichtanders]

I am not experienced with the pitfalls of Lux.transform. sounds like it works kind of good, as DiffEqFlux is relying on it, but it probably has some drawbacks.

[zahachtah]

simple test of Lux.transform didn't work. Just to add, Lux support would be nice.