Skip to content

Conversation

@mcabbott
Copy link
Member

@mcabbott mcabbott commented Dec 3, 2024

  • I think it's too obscure to say "cuDNN.jl has to be installed". We should just write the Pkg.add command out.
  • The line model(noisy |> gpu) |> cpu is doing too many things, simplify?
  • I don't like the "device = ..." pattern, but Carlo seems set on it, and added it here (instead of gpu). This at least tidies up a bit... most importantly never re-using x, y for things of different type.
  • Maybe the name of that should be a verb not a noun? Like move = get_device()? Not done yet.

@CarloLucibello
Copy link
Member

CarloLucibello commented Dec 4, 2024

Maybe the name of that should be a verb not a noun? Like move = get_device()? Not done yet

I would stick to device, it is very familiar to pytorch users, and x |> device is nice, x |> move is not, although move(x) could be better than device(x) . to_device = get_device() could be another option, bu I still preferer device, also because get_device() returns an XDevice object.

@mcabbott mcabbott merged commit 74c3a63 into FluxML:master Dec 5, 2024
7 of 8 checks passed
@mcabbott mcabbott deleted the tweak_quickstart branch December 5, 2024 03:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants