Skip to content

Add K-LMS scheduler from k-diffusion #185

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
anton-l merged 11 commits into from
Aug 16, 2022
Merged

Add K-LMS scheduler from k-diffusion #185

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
3691a1e
test LMS with LDM
anton-l Aug 11, 2022
f5017b7
test LMS with LDM
anton-l Aug 15, 2022
c644dc8
Merge remote-tracking branch 'upstream/main' into lms-scheduler
anton-l Aug 15, 2022
8be3114
Interchangeable sigma and timestep. Added dummy objects
anton-l Aug 15, 2022
eb1aee4
Debug
anton-l Aug 16, 2022
ad7d92f
Merge master
anton-l Aug 16, 2022
ab50168
cuda generator
anton-l Aug 16, 2022
5fe7895
Fix derivatives
anton-l Aug 16, 2022
79efde4
Merge remote-tracking branch 'upstream/main' into lms-scheduler
anton-l Aug 16, 2022
1be0071
Update tests
anton-l Aug 16, 2022
4aa07f5
Rename Lms->LMS
anton-l Aug 16, 2022
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Update tests
  • Loading branch information
@anton-l
anton-l committed Aug 16, 2022
commit 1be007191eb4c10b1cdaaa89a5237b605d175ab1
5 changes: 4 additions & 1 deletion src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,7 +125,10 @@ def __call__(
noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

# compute the previous noisy sample x_t -> x_t-1
latents = self.scheduler.step(noise_pred, i, latents, **extra_step_kwargs)["prev_sample"]
if isinstance(self.scheduler, LmsDiscreteScheduler):
latents = self.scheduler.step(noise_pred, i, latents, **extra_step_kwargs)["prev_sample"]
Copy link
Contributor

@patrickvonplaten patrickvonplaten Aug 16, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

shouldn't we maybe pass the sigma here?
Copy link
Contributor

@patrickvonplaten patrickvonplaten Aug 16, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok we need it anyway - good to leave as is for me!
Copy link
Member Author

@anton-l anton-l Aug 16, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The i is needed inside the step for other calculations too, and we can't reverse the sigma into i, so we'll need it like this until a more full refactor.
else:
latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs)["prev_sample"]
Copy link
Member Author

@anton-l anton-l Aug 16, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This can be made cleaner by removing the branch with t->i, but for now I'm not touching the timesteps from k-lms for ease of debugging.

# scale and decode the image latents with vae
latents = 1 / 0.18215 * latents
Expand Down
7 changes: 4 additions & 3 deletions tests/test_modeling_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -842,7 +842,7 @@ def test_ldm_text2img_fast(self):
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@slow
@unittest.skipIf(torch_device == "cpu", "Stable diffusion is suppused to run on GPU")
@unittest.skipIf(torch_device == "cpu", "Stable diffusion is supposed to run on GPU")
def test_stable_diffusion(self):
# make sure here that pndm scheduler skips prk
sd_pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-1-diffusers")
Expand All @@ -863,7 +863,7 @@ def test_stable_diffusion(self):
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@slow
@unittest.skipIf(torch_device == "cpu", "Stable diffusion is suppused to run on GPU")
@unittest.skipIf(torch_device == "cpu", "Stable diffusion is supposed to run on GPU")
def test_stable_diffusion_fast_ddim(self):
sd_pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-1-diffusers")

Expand Down Expand Up @@ -979,7 +979,8 @@ def test_karras_ve_pipeline(self):
expected_slice = np.array([0.26815, 0.1581, 0.2658, 0.23248, 0.1550, 0.2539, 0.1131, 0.1024, 0.0837])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

# @slow
@slow
@unittest.skipIf(torch_device == "cpu", "Stable diffusion is supposed to run on GPU")
def test_lms_stable_diffusion_pipeline(self):
model_id = "CompVis/stable-diffusion-v1-1-diffusers"
pipe = StableDiffusionPipeline.from_pretrained(model_id, use_auth_token=True)
Expand Down