stable_diffusion example: Mac cpu-only failed when click Generate New Image (0.1.33) about reflex-examples HOT 6 CLOSED

Hi @milochen0418 great this is a great point. I am just about to submit a PR that will also allow you to run this with only a CPU on a mac. It will take about 4/5 minutes per image if running on a cpu only and running for 30 inference steps.

To speed this up further you can make the image smaller, by changing the width parameter in image_style inside of styles.py, or you can reduce the number of inference steps further.
Hope this helps.

from reflex-examples.

Some question related to this issue
https://stackoverflow.com/questions/72313510/runtimeerror-the-mps-backend-is-supported-on-macos-12-3-current-os-version-can

from reflex-examples.

More reference
oobabooga/text-generation-webui#1337

from reflex-examples.

After click Generate New Image, it will use img2img_mac.py
Right now, the current code only support mps

Apple Metal Performance Shaders (MPS)
The MPS backend extends the PyTorch framework by providing scripts and capabilities to set up and run operations on Mac. MPS optimizes compute performance with kernels that are fine-tuned for the unique characteristics of each Metal GPU family.

Hi @tgberkeley , Is it possible to run on the CPU?

from reflex-examples.

Hi @milochen0418 great this is a great point. I am just about to submit a PR that will also allow you to run this with only a CPU on a mac. It will take about 4/5 minutes per image if running on a cpu only and running for 30 inference steps.

To speed this up further you can make the image smaller, by changing the width parameter in image_style inside of styles.py, or you can reduce the number of inference steps further. Hope this helps.

Hi @tgberkeley,
You are so nice.
It's wonderful to have your code. Let everyone enjoy the stable diffusion of pynecone app.
It works on my old Mac with CPU only from this PR

• #117

And the following is my testing result

• #117 (review)

You say it takes 0.8 minutes for 30 inference steps.
In my case, I set Number of inference steps as 100 and Strength of diffusion as 70 in the running app.
Does that mean it needs 100*70 inference steps?
Is my understanding correct?
Thanks

from reflex-examples.

Bug solved by

• #117

from reflex-examples.