Abhirath Anand

Oh, you're right! That is indeed the issue. I'll see how I can make `chunk` work on the GPU

Is the `default_rng_value()` function in a release yet? I can change the Metalhead API and bump compat if it is

P.S. there really needs to be a way to _just_ run docs previews if it's only a docs change - the Metalhead CI load is quite heavy

> Separate it into a different `.yml`? Though I think the documentation job should run in parallel? You don't get a green check mark immediately, but the preview job should...

Some weird errors pop up in the docs CI: https://github.com/FluxML/Metalhead.jl/runs/8228349979?check_suite_focus=true#step:6:18 doesn't seem to be the right error for the pre-trained ResNet loading - this one seems to suggest that there's...

The RNG error is odd. I'll try and test on 1.6 to see if I can narrow something down. > Maybe try `using Flux`? Also I think it would be...

> @theabhirath I think the appropriate fix here is to always save the pre-trained weights using the minimum Julia compat. So in this case, re-save them under Julia 1.6. Gotcha,...

Oh, it is indeed 🤦🏽 Still adjusting to the Julian syntax v/s the Pythonic one

Now that there is a [consistent code style for Metalhead](https://github.com/FluxML/Metalhead.jl/pull/168), could this be extended to the Flux repo as well?

I was about to set that up for Metalhead but decided to hold off until https://github.com/domluna/JuliaFormatter.jl/issues/604 is resolved