AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

[2737]: Add DPMSolverMultistepScheduler to CLIP guided community pipeline (#2779)

Browse Source 
[2737]: Add DPMSolverMultistepScheduler to CLIP guided community pipelines

Co-authored-by: njindal <njindal@adobe.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
This commit is contained in:
Nipun Jindal
2023-03-23 18:50:24 +05:30
committed by GitHub
parent 2ef9bdd76f
commit 055c90f589
2 changed files with 8 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/community/clip_guided_stable_diffusion.py
View File

@@ -11,6 +11,7 @@ from diffusers import (
AutoencoderKL,
DDIMScheduler,
DiffusionPipeline,
DPMSolverMultistepScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
UNet2DConditionModel,
@@ -63,7 +64,7 @@ class CLIPGuidedStableDiffusion(DiffusionPipeline):
clip_model: CLIPModel,
tokenizer: CLIPTokenizer,
unet: UNet2DConditionModel,
scheduler: Union[PNDMScheduler, LMSDiscreteScheduler, DDIMScheduler],
scheduler: Union[PNDMScheduler, LMSDiscreteScheduler, DDIMScheduler, DPMSolverMultistepScheduler],
feature_extractor: CLIPImageProcessor,
):
super().__init__()
@@ -125,17 +126,12 @@ class CLIPGuidedStableDiffusion(DiffusionPipeline):
):
latents = latents.detach().requires_grad_()
if isinstance(self.scheduler, LMSDiscreteScheduler):
sigma = self.scheduler.sigmas[index]
# the model input needs to be scaled to match the continuous ODE formulation in K-LMS
latent_model_input = latents / ((sigma**2 + 1) ** 0.5)
else:
latent_model_input = latents
latent_model_input = self.scheduler.scale_model_input(latents, timestep)
# predict the noise residual
noise_pred = self.unet(latent_model_input, timestep, encoder_hidden_states=text_embeddings).sample
if isinstance(self.scheduler, (PNDMScheduler, DDIMScheduler)):
if isinstance(self.scheduler, (PNDMScheduler, DDIMScheduler, DPMSolverMultistepScheduler)):
alpha_prod_t = self.scheduler.alphas_cumprod[timestep]
beta_prod_t = 1 - alpha_prod_t
# compute predicted original sample from predicted noise also called

12
examples/community/clip_guided_stable_diffusion_img2img.py
View File

@@ -13,6 +13,7 @@ from diffusers import (
AutoencoderKL,
DDIMScheduler,
DiffusionPipeline,
DPMSolverMultistepScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
UNet2DConditionModel,
@@ -140,7 +141,7 @@ class CLIPGuidedStableDiffusion(DiffusionPipeline):
clip_model: CLIPModel,
tokenizer: CLIPTokenizer,
unet: UNet2DConditionModel,
scheduler: Union[PNDMScheduler, LMSDiscreteScheduler, DDIMScheduler],
scheduler: Union[PNDMScheduler, LMSDiscreteScheduler, DDIMScheduler, DPMSolverMultistepScheduler],
feature_extractor: CLIPFeatureExtractor,
):
super().__init__()
@@ -263,17 +264,12 @@ class CLIPGuidedStableDiffusion(DiffusionPipeline):
):
latents = latents.detach().requires_grad_()
if isinstance(self.scheduler, LMSDiscreteScheduler):
sigma = self.scheduler.sigmas[index]
# the model input needs to be scaled to match the continuous ODE formulation in K-LMS
latent_model_input = latents / ((sigma**2 + 1) ** 0.5)
else:
latent_model_input = latents
latent_model_input = self.scheduler.scale_model_input(latents, timestep)
# predict the noise residual
noise_pred = self.unet(latent_model_input, timestep, encoder_hidden_states=text_embeddings).sample
if isinstance(self.scheduler, (PNDMScheduler, DDIMScheduler)):
if isinstance(self.scheduler, (PNDMScheduler, DDIMScheduler, DPMSolverMultistepScheduler)):
alpha_prod_t = self.scheduler.alphas_cumprod[timestep]
beta_prod_t = 1 - alpha_prod_t
# compute predicted original sample from predicted noise also called