[Feature Request]: Selectively integrate Fooocus OG PRs

[iwr-redmond]

Is there an existing issue for this?

• I have searched the existing issues and checked the recent builds/commits

What would your feature do?

There are some useful PRs in the original Fooocus repository, and it may be helpful to integrate some of this code into FooocusPlus.

Proposed PR review list

1. Colorize AR list by ratio type (fooocus#2926)
2. Fix VRAM issues on AMD (fooocus#3263)
3. Remove console print of clipboard contents (fooocus#3751)
4. Fix inpainting on Mac M series (fooocus#3872)
5. Inline LoRA support, wildcard improvements (fooocus#3892) [large PR]
6. File management features (fooocus#4018) [large PR]

Additional information

No response

[DavidDragonsage]

@iwr-redmond Thank you for this quick summary. It has been in the back of my mind to go through some of those old PRs and bug reports because in most case we have inherited those issues.

That VRAM issues on AMD looks particularly relevant. On Pure Fooocus, we recently had an AMD user who has 12GB VRAM but only 4GB is recognized. This happens on Ruined too. I asked him if the same thing happened on original Fooocus but never received any further response, so I had to drop the issue. Edited to add: I see its not the same issue at all, oh well.

[iwr-redmond]

12GB VRAM but only 4GB is recognized.

That may be an issue with DirectML on Windows. Truly, if AMD were a person he'd hang out more often than his Mama’s washing. Intel got into the GPU game not so long ago and they already have full XPU support in Pytorch and Bitsandbytes on both Windows and Linux.

[DavidDragonsage]

Yes, I'm quite sure it is. There was a complaint at the Microsoft DirectML support forum about this problem with his specific card, a AMD Radeon 12GB RX 6700 XT. Of course the Microsoft rep. politely said that DirectML is "no longer in active development".

I found a post where he did try original Fooocus and it only saw 1GB of VRAM! 😲

It appears that running on Linux is the only solution.

[iwr-redmond]

Yes, at least until AMD's TheRock builds are stable (torchruntime#117).