From 8d2aa65402c829583e26cdf2c336b8d3057657d6 Mon Sep 17 00:00:00 2001 From: Volpeon Date: Fri, 5 May 2023 10:51:14 +0200 Subject: Update --- pipelines/stable_diffusion/vlpn_stable_diffusion.py | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-) (limited to 'pipelines/stable_diffusion') diff --git a/pipelines/stable_diffusion/vlpn_stable_diffusion.py b/pipelines/stable_diffusion/vlpn_stable_diffusion.py index aa3dbc6..aa446ec 100644 --- a/pipelines/stable_diffusion/vlpn_stable_diffusion.py +++ b/pipelines/stable_diffusion/vlpn_stable_diffusion.py @@ -386,7 +386,7 @@ class VlpnStableDiffusion(DiffusionPipeline): def decode_latents(self, latents): latents = 1 / self.vae.config.scaling_factor * latents - image = self.vae.decode(latents.to(dtype=self.vae.dtype)).sample + image = self.vae.decode(latents.to(dtype=self.vae.dtype), return_dict=False)[0] image = (image / 2 + 0.5).clamp(0, 1) # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16 image = image.cpu().permute(0, 2, 3, 1).float().numpy() @@ -545,7 +545,8 @@ class VlpnStableDiffusion(DiffusionPipeline): t, encoder_hidden_states=prompt_embeds, cross_attention_kwargs=cross_attention_kwargs, - ).sample + return_dict=False, + )[0] # perform guidance if do_classifier_free_guidance: @@ -567,7 +568,8 @@ class VlpnStableDiffusion(DiffusionPipeline): ) uncond_emb, _ = prompt_embeds.chunk(2) # forward and give guidance - degraded_pred = self.unet(degraded_latents, t, encoder_hidden_states=uncond_emb).sample + degraded_pred = self.unet( + degraded_latents, t, encoder_hidden_states=uncond_emb, return_dict=False)[0] noise_pred += sag_scale * (noise_pred_uncond - degraded_pred) else: # DDIM-like prediction of x0 @@ -579,11 +581,12 @@ class VlpnStableDiffusion(DiffusionPipeline): pred_x0, cond_attn, t, self.pred_epsilon(latents, noise_pred, t) ) # forward and give guidance - degraded_pred = self.unet(degraded_latents, t, encoder_hidden_states=prompt_embeds).sample + degraded_pred = self.unet( + degraded_latents, t, encoder_hidden_states=prompt_embeds, return_dict=False)[0] noise_pred += sag_scale * (noise_pred - degraded_pred) # compute the previous noisy sample x_t -> x_t-1 - latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample + latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] # call the callback, if provided if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): -- cgit v1.2.3-54-g00ecf