Merge branch 'main' of https://github.com/huggingface/diffusers into main

models/vision/ddim/modeling_ddim.py

@@ -34,49 +34,68 @@ class DDIM(DiffusionPipeline):
         inference_step_times = range(0, num_trained_timesteps, num_trained_timesteps // num_inference_steps)
 
         self.unet.to(torch_device)
+
+        # Sample gaussian noise to begin loop
         image = self.noise_scheduler.sample_noise(
             (batch_size, self.unet.in_channels, self.unet.resolution, self.unet.resolution),
             device=torch_device,
             generator=generator,
         )
 
+        # See formulas (9), (10) and (7) of DDIM paper https://arxiv.org/pdf/2010.02502.pdf
+        # Ideally, read DDIM paper in-detail understanding
+
+        # Notation (<variable name> -> <name in paper>
+        # - pred_noise_t -> e_theta(x_t, t)
+        # - pred_original_image -> f_theta(x_t, t) or x_0
+        # - std_dev_t -> sigma_t
+        # - eta -> η
+        # - pred_image_direction -> "direction pointingc to x_t"
+        # - pred_prev_image -> "x_t-1"
         for t in tqdm.tqdm(reversed(range(num_inference_steps)), total=num_inference_steps):
-            # get actual t and t-1
+            # 1. predict noise residual
+            with torch.no_grad():
+                pred_noise_t = self.unet(image, inference_step_times[t])
+
+            # 2. get actual t and t-1
             train_step = inference_step_times[t]
             prev_train_step = inference_step_times[t - 1] if t > 0 else -1
 
-            # compute alphas
+            # 3. compute alphas, betas
             alpha_prod_t = self.noise_scheduler.get_alpha_prod(train_step)
             alpha_prod_t_prev = self.noise_scheduler.get_alpha_prod(prev_train_step)
-            alpha_prod_t_rsqrt = 1 / alpha_prod_t.sqrt()
-            alpha_prod_t_prev_rsqrt = 1 / alpha_prod_t_prev.sqrt()
-            beta_prod_t_sqrt = (1 - alpha_prod_t).sqrt()
-            beta_prod_t_prev_sqrt = (1 - alpha_prod_t_prev).sqrt()
-
-            # compute relevant coefficients
-            coeff_1 = (
-                (alpha_prod_t_prev - alpha_prod_t).sqrt()
-                * alpha_prod_t_prev_rsqrt
-                * beta_prod_t_prev_sqrt
-                / beta_prod_t_sqrt
-                * eta
-            )
-            coeff_2 = ((1 - alpha_prod_t_prev) - coeff_1**2).sqrt()
-
-            # model forward
-            with torch.no_grad():
-                noise_residual = self.unet(image, train_step)
+            beta_prod_t = (1 - alpha_prod_t)
+            beta_prod_t_prev = (1 - alpha_prod_t_prev)
+
+            # 4. Compute predicted previous image from predicted noise
 
-            # predict mean of prev image
-            pred_mean = alpha_prod_t_rsqrt * (image - beta_prod_t_sqrt * noise_residual)
-            pred_mean = torch.clamp(pred_mean, -1, 1)
-            pred_mean = (1 / alpha_prod_t_prev_rsqrt) * pred_mean + coeff_2 * noise_residual
+            # First: compute predicted original image from predicted noise also called
+            # "predicted x_0" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+            pred_original_image = (image - beta_prod_t.sqrt() * pred_noise_t) / alpha_prod_t.sqrt()
 
-            # if eta > 0.0 add noise. Note eta = 1.0 essentially corresponds to DDPM
+            # Second: Clip "predicted x_0"
+            pred_original_image = torch.clamp(pred_original_image, -1, 1)
+
+            # Third: Compute variance: "sigma_t(η)" -> see formula (16)
+            # σ_t = sqrt((1 − α_t−1)/(1 − α_t)) * sqrt(1 − α_t/α_t−1)
+            std_dev_t = (beta_prod_t_prev / beta_prod_t).sqrt() * (1 - alpha_prod_t / alpha_prod_t_prev).sqrt()
+            std_dev_t = eta * std_dev_t
+
+            # Fourth: Compute "direction pointing to x_t" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+            pred_image_direction = (1 - alpha_prod_t_prev - std_dev_t**2).sqrt() * pred_noise_t
+
+            # Fifth: Compute x_t without "random noise" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+            pred_prev_image = alpha_prod_t_prev.sqrt() * pred_original_image + pred_image_direction
+
+            # 5. Sample x_t-1 image optionally if η > 0.0 by adding noise to pred_prev_image
+            # Note: eta = 1.0 essentially corresponds to DDPM
             if eta > 0.0:
                 noise = self.noise_scheduler.sample_noise(image.shape, device=image.device, generator=generator)
-                image = pred_mean + coeff_1 * noise
+                prev_image = pred_prev_image + std_dev_t * noise
             else:
-                image = pred_mean
+                prev_image = pred_prev_image
+
+            # 6. Set current image to prev_image: x_t -> x_t-1
+            image = prev_image
 
         return image

models/vision/ddpm/modeling_ddpm.py

@@ -30,43 +30,43 @@ class DDPM(DiffusionPipeline):
             torch_device = "cuda" if torch.cuda.is_available() else "cpu"
 
         self.unet.to(torch_device)
-        # 1. Sample gaussian noise
+
+        # Sample gaussian noise to begin loop
         image = self.noise_scheduler.sample_noise(
             (batch_size, self.unet.in_channels, self.unet.resolution, self.unet.resolution),
             device=torch_device,
             generator=generator,
         )
-        for t in tqdm.tqdm(reversed(range(len(self.noise_scheduler))), total=len(self.noise_scheduler)):
-            # i) define coefficients for time step t
-            clipped_image_coeff = 1 / torch.sqrt(self.noise_scheduler.get_alpha_prod(t))
-            clipped_noise_coeff = torch.sqrt(1 / self.noise_scheduler.get_alpha_prod(t) - 1)
-            image_coeff = (
-                (1 - self.noise_scheduler.get_alpha_prod(t - 1))
-                * torch.sqrt(self.noise_scheduler.get_alpha(t))
-                / (1 - self.noise_scheduler.get_alpha_prod(t))
-            )
-            clipped_coeff = (
-                torch.sqrt(self.noise_scheduler.get_alpha_prod(t - 1))
-                * self.noise_scheduler.get_beta(t)
-                / (1 - self.noise_scheduler.get_alpha_prod(t))
-            )
 
-            # ii) predict noise residual
+        for t in tqdm.tqdm(reversed(range(len(self.noise_scheduler))), total=len(self.noise_scheduler)):
+            # 1. predict noise residual
             with torch.no_grad():
                 noise_residual = self.unet(image, t)
 
-            # iii) compute predicted image from residual
+            # 2. compute alphas, betas
+            alpha_prod_t = self.noise_scheduler.get_alpha_prod(t)
+            alpha_prod_t_prev = self.noise_scheduler.get_alpha_prod(t - 1)
+            beta_prod_t = 1 - alpha_prod_t
+            beta_prod_t_prev = 1 - alpha_prod_t_prev
+
+            # 3. compute predicted image from residual
             # See 2nd formula at https://github.com/hojonathanho/diffusion/issues/5#issue-896554416 for comparison
-            pred_mean = clipped_image_coeff * image - clipped_noise_coeff * noise_residual
+            # First: Compute inner formula
+            pred_mean = (1 / alpha_prod_t.sqrt()) * (image - beta_prod_t.sqrt() * noise_residual)
+            # Second: Clip
             pred_mean = torch.clamp(pred_mean, -1, 1)
-            prev_image = clipped_coeff * pred_mean + image_coeff * image
+            # Third: Compute outer coefficients
+            pred_mean_coeff = (alpha_prod_t_prev.sqrt() * self.noise_scheduler.get_beta(t)) / beta_prod_t
+            image_coeff = (beta_prod_t_prev * self.noise_scheduler.get_alpha(t).sqrt()) / beta_prod_t 
+            # Fourth: Compute outer formula
+            prev_image = pred_mean_coeff * pred_mean + image_coeff * image
 
-            # iv) sample variance
+            # 4. sample variance
             prev_variance = self.noise_scheduler.sample_variance(
                 t, prev_image.shape, device=torch_device, generator=generator
             )
 
-            # v) sample  x_{t-1} ~ N(prev_image, prev_variance)
+            # 5. sample  x_{t-1} ~ N(prev_image, prev_variance) = add variance to predicted image
             sampled_prev_image = prev_image + prev_variance
             image = sampled_prev_image
 

src/diffusers/pipeline_utils.py

@@ -40,6 +40,7 @@ LOADABLE_CLASSES = {
     },
     "transformers": {
         "PreTrainedTokenizer": ["save_pretrained", "from_pretrained"],
+        "PreTrainedModel": ["save_pretrained", "from_pretrained"],
     },
 }
 
@@ -82,24 +83,25 @@ class DiffusionPipeline(ConfigMixin):
         model_index_dict.pop("_diffusers_version")
         model_index_dict.pop("_module")
 
-        for name, (library_name, class_name) in model_index_dict.items():
-            importable_classes = LOADABLE_CLASSES[library_name]
-
-            # TODO: Suraj
-            if library_name == self.__module__:
-                library_name = self
-
-            library = importlib.import_module(library_name)
-            class_obj = getattr(library, class_name)
-            class_candidates = {c: getattr(library, c) for c in importable_classes.keys()}
+        for pipeline_component_name in model_index_dict.keys():
+            sub_model = getattr(self, pipeline_component_name)
+            model_cls = sub_model.__class__
 
             save_method_name = None
-            for class_name, class_candidate in class_candidates.items():
-                if issubclass(class_obj, class_candidate):
-                    save_method_name = importable_classes[class_name][0]
-
-            save_method = getattr(getattr(self, name), save_method_name)
-            save_method(os.path.join(save_directory, name))
+            # search for the model's base class in LOADABLE_CLASSES
+            for library_name, library_classes in LOADABLE_CLASSES.items():
+                library = importlib.import_module(library_name)
+                for base_class, save_load_methods in library_classes.items():
+                    class_candidate = getattr(library, base_class)
+                    if issubclass(model_cls, class_candidate):
+                        # if we found a suitable base class in LOADABLE_CLASSES then grab its save method
+                        save_method_name = save_load_methods[0]
+                        break
+                if save_method_name is not None:
+                    break
+
+            save_method = getattr(sub_model, save_method_name)
+            save_method(os.path.join(save_directory, pipeline_component_name))
 
     @classmethod
     def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], **kwargs):