Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

outputsize(::Dense) #1262

Open
lcontento opened this issue Mar 14, 2025 · 1 comment
Open

outputsize(::Dense) #1262

lcontento opened this issue Mar 14, 2025 · 1 comment
Labels
bug Something isn't working

Comments

@lcontento
Copy link

The Dense layer does not take into account additional dimensions (other than batch dimension) when computing outputsize.

outputsize(d::Dense, _, ::AbstractRNG) = (d.out_dims,)

This is in contrast with what the fallback for outpusize does, leading to the following somewhat contradictory result:

julia> d = Lux.Dense(2, 10)
Dense(2 => 10)      # 30 parameters

julia> Lux.outputsize(d, rand(2, 3, 4), Random.default_rng())
(10,)

julia> Lux.outputsize(Chain(d), rand(2, 3, 4), Random.default_rng())
(10, 3)

My feeling is that the expected behaviour should be the one displayed by Chain and that the current implementation of outputsize for Dense should be fixed.

@avik-pal avik-pal added the bug Something isn't working label Mar 14, 2025
@avik-pal
Copy link
Member

yeah this is a bug

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

No branches or pull requests

2 participants