feat(ui): Expose tile size as advanced upscaling option.

[hipsterusername]

Summary

This exposes the tile size and overlap parameter for more controllable upscaling settings. This helps with manually solving for AMD OOM issues without changing functionality for others.

Related Issues / Discussions

#6981

QA Instructions

Test upscaling SD 1.5 on an AMD card.

Merge Plan

N/A

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Tests added / updated (if applicable)
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

[heathen711]

Worked for me.

My 2c: Maybe we need to add a ROCM section to the FAQ for these kinds of things to be tracked? I'm sure I'll find other things in the future...

[hipsterusername]

Worked for me.

My 2c: Maybe we need to add a ROCM section to the FAQ for these kinds of things to be tracked? I'm sure I'll find other things in the future...

Could see that being super helpful. Do you have a list of the ROCm stuff that you’ve navigated through already?

[heathen711]

Worked for me.
My 2c: Maybe we need to add a ROCM section to the FAQ for these kinds of things to be tracked? I'm sure I'll find other things in the future...

Could see that being super helpful. Do you have a list of the ROCm stuff that you’ve navigated through already?

So far its been install issues, but I'm addressing those in the rocm docker PR, there's two versions in there, the full version will need it's own documentation once it's approved, so I'd want to link that there and/or at the docker guide.
#8152

I also found that https://invoke-ai.github.io/InvokeAI/features/low-vram/#pytorch-cuda-allocator-config does not work for rocm (for obvious reasons) but it's not obvious without looking at the logs (the UI just says it failed), so maybe a note on that for rocm users.

These all seem like they have homes in other pages, but I was thinking this is minor and should just be a FAQ. Writing this makes me feel like the guide need to branch, cpu vs cuda vs rocm. Which is out of scope for this PR. So just the simple FAQ to start, and polish as time goes on.

[hipsterusername]

Worked for me.
My 2c: Maybe we need to add a ROCM section to the FAQ for these kinds of things to be tracked? I'm sure I'll find other things in the future...

Could see that being super helpful. Do you have a list of the ROCm stuff that you’ve navigated through already?

So far its been install issues, but I'm addressing those in the rocm docker PR, there's two versions in there, the full version will need it's own documentation once it's approved, so I'd want to link that there and/or at the docker guide. #8152

I also found that https://invoke-ai.github.io/InvokeAI/features/low-vram/#pytorch-cuda-allocator-config does not work for rocm (for obvious reasons) but it's not obvious without looking at the logs (the UI just says it failed), so maybe a note on that for rocm users.

These all seem like they have homes in other pages, but I was thinking this is minor and should just be a FAQ. Writing this makes me feel like the guide need to branch, cpu vs cuda vs rocm. Which is out of scope for this PR. So just the simple FAQ to start, and polish as time goes on.

Aye - this will also probably change a bit as ROCm gets live for windows and deeper torch support

Open to you tackling some of the ROCm stuff you’d recommend - else we’ll just be summarizing what you’ve found already hehe

[psychedelicious]

Looks good just 2 minor things