Update custom_pipeline.py

custom_pipeline.py

@@ -44,7 +44,6 @@ from diffusers.utils.torch_utils import randn_tensor
 from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin
 from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
 
-
 if is_invisible_watermark_available():
     from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
 
@@ -88,7 +87,6 @@ EXAMPLE_DOC_STRING = """
         ```
 """
 
-
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
 def retrieve_latents(
     encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
@@ -773,7 +771,7 @@ class CosStableDiffusionXLInstructPix2PixPipeline(
         # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
         # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
         # corresponds to doing no classifier free guidance.
-        do_classifier_free_guidance = guidance_scale > 10.0 and image_guidance_scale >= 5.0
+        do_classifier_free_guidance = guidance_scale > 1.0 and image_guidance_scale >= 1.0
 
         # 3. Encode input prompt
         text_encoder_lora_scale = (
@@ -815,9 +813,7 @@ class CosStableDiffusionXLInstructPix2PixPipeline(
             device,
             do_classifier_free_guidance,
         )
-        
-        image_latents = image_latents * self.vae.config.scaling_factor
-        
+
         # 7. Prepare latent variables
         num_channels_latents = self.vae.config.latent_channels
         latents = self.prepare_latents(
@@ -859,7 +855,8 @@ class CosStableDiffusionXLInstructPix2PixPipeline(
             dtype=prompt_embeds.dtype,
             text_encoder_projection_dim=text_encoder_projection_dim,
         )
-
+        add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1)
+        
         if do_classifier_free_guidance:
             # The extra concat similar to how it's done in SD InstructPix2Pix.
             prompt_embeds = torch.cat([prompt_embeds, negative_prompt_embeds, negative_prompt_embeds], dim=0)
@@ -870,35 +867,19 @@ class CosStableDiffusionXLInstructPix2PixPipeline(
 
         prompt_embeds = prompt_embeds.to(device)
         add_text_embeds = add_text_embeds.to(device)
-        add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1)
 
         # 11. Denoising loop
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
-        if denoising_end is not None and isinstance(denoising_end, float) and denoising_end > 0 and denoising_end < 1:
-            discrete_timestep_cutoff = int(
-                round(
-                    self.scheduler.config.num_train_timesteps
-                    - (denoising_end * self.scheduler.config.num_train_timesteps)
-                )
-            )
-            num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps)))
-            timesteps = timesteps[:num_inference_steps]
-
         with self.progress_bar(total=num_inference_steps) as progress_bar:
             for i, t in enumerate(timesteps):
-                # Expand the latents if we are doing classifier free guidance.
-                # The latents are expanded 3 times because for pix2pix the guidance
-                # is applied for both the text and the input image.
+                # expand the latents if we are doing classifier free guidance
                 latent_model_input = torch.cat([latents] * 3) if do_classifier_free_guidance else latents
-
-                # concat latents, image_latents in the channel dimension
-                scaled_latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
-                scaled_latent_model_input = torch.cat([scaled_latent_model_input, image_latents], dim=1)
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
 
                 # predict the noise residual
                 added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
                 noise_pred = self.unet(
-                    scaled_latent_model_input,
+                    torch.cat([latent_model_input, image_latents], dim=1),
                     t,
                     encoder_hidden_states=prompt_embeds,
                     cross_attention_kwargs=cross_attention_kwargs,
@@ -911,7 +892,7 @@ class CosStableDiffusionXLInstructPix2PixPipeline(
                     noise_pred_text, noise_pred_image, noise_pred_uncond = noise_pred.chunk(3)
                     noise_pred = (
                         noise_pred_uncond
-                        + guidance_scale * (noise_pred_text - noise_pred_image)
+                        + guidance_scale * (noise_pred_text - noise_pred_uncond)
                         + image_guidance_scale * (noise_pred_image - noise_pred_uncond)
                     )
 
@@ -920,12 +901,7 @@ class CosStableDiffusionXLInstructPix2PixPipeline(
                     noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
 
                 # compute the previous noisy sample x_t -> x_t-1
-                latents_dtype = latents.dtype
                 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
-                if latents.dtype != latents_dtype:
-                    if torch.backends.mps.is_available():
-                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
-                        latents = latents.to(latents_dtype)
 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
@@ -934,41 +910,8 @@ class CosStableDiffusionXLInstructPix2PixPipeline(
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
-                if XLA_AVAILABLE:
-                    xm.mark_step()
-
         if not output_type == "latent":
-            # make sure the VAE is in float32 mode, as it overflows in float16
-            needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
-
-            if needs_upcasting:
-                self.upcast_vae()
-                latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
-            elif latents.dtype != self.vae.dtype:
-                if torch.backends.mps.is_available():
-                    # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
-                    self.vae = self.vae.to(latents.dtype)
-
-            # unscale/denormalize the latents
-            # denormalize with the mean and std if available and not None
-            has_latents_mean = hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None
-            has_latents_std = hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None
-            if has_latents_mean and has_latents_std:
-                latents_mean = (
-                    torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1).to(latents.device, latents.dtype)
-                )
-                latents_std = (
-                    torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1).to(latents.device, latents.dtype)
-                )
-                latents = latents * latents_std / self.vae.config.scaling_factor + latents_mean
-            else:
-                latents = latents / self.vae.config.scaling_factor
-
-            image = self.vae.decode(latents, return_dict=False)[0]
-
-            # cast back to fp16 if needed
-            if needs_upcasting:
-                self.vae.to(dtype=torch.float16)
+            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
         else:
             return StableDiffusionXLPipelineOutput(images=latents)