diffusers 0.27.1__py3-none-any.whl → 0.32.2__py3-none-any.whl

diffusers/schedulers/scheduling_deis_multistep.py

@@ -22,10 +22,14 @@ import numpy as np
 import torch
 
 from ..configuration_utils import ConfigMixin, register_to_config
-from ..utils import deprecate
+from ..utils import deprecate, is_scipy_available
 from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin, SchedulerOutput
 
 
+if is_scipy_available():
+    import scipy.stats
+
+
 # Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar
 def betas_for_alpha_bar(
     num_diffusion_timesteps,
@@ -61,7 +65,7 @@ def betas_for_alpha_bar(
             return math.exp(t * -12.0)
 
     else:
-        raise ValueError(f"Unsupported alpha_tranform_type: {alpha_transform_type}")
+        raise ValueError(f"Unsupported alpha_transform_type: {alpha_transform_type}")
 
     betas = []
     for i in range(num_diffusion_timesteps):
@@ -111,6 +115,11 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
         use_karras_sigmas (`bool`, *optional*, defaults to `False`):
              Whether to use Karras sigmas for step sizes in the noise schedule during the sampling process. If `True`,
              the sigmas are determined according to a sequence of noise levels {σi}.
+        use_exponential_sigmas (`bool`, *optional*, defaults to `False`):
+            Whether to use exponential sigmas for step sizes in the noise schedule during the sampling process.
+        use_beta_sigmas (`bool`, *optional*, defaults to `False`):
+            Whether to use beta sigmas for step sizes in the noise schedule during the sampling process. Refer to [Beta
+            Sampling is All You Need](https://huggingface.co/papers/2407.12173) for more information.
         timestep_spacing (`str`, defaults to `"linspace"`):
             The way the timesteps should be scaled. Refer to Table 2 of the [Common Diffusion Noise Schedules and
             Sample Steps are Flawed](https://huggingface.co/papers/2305.08891) for more information.
@@ -138,9 +147,19 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
         solver_type: str = "logrho",
         lower_order_final: bool = True,
         use_karras_sigmas: Optional[bool] = False,
+        use_exponential_sigmas: Optional[bool] = False,
+        use_beta_sigmas: Optional[bool] = False,
+        use_flow_sigmas: Optional[bool] = False,
+        flow_shift: Optional[float] = 1.0,
         timestep_spacing: str = "linspace",
         steps_offset: int = 0,
     ):
+        if self.config.use_beta_sigmas and not is_scipy_available():
+            raise ImportError("Make sure to install scipy if you want to use beta sigmas.")
+        if sum([self.config.use_beta_sigmas, self.config.use_exponential_sigmas, self.config.use_karras_sigmas]) > 1:
+            raise ValueError(
+                "Only one of `config.use_beta_sigmas`, `config.use_exponential_sigmas`, `config.use_karras_sigmas` can be used."
+            )
         if trained_betas is not None:
             self.betas = torch.tensor(trained_betas, dtype=torch.float32)
         elif beta_schedule == "linear":
@@ -152,7 +171,7 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)
         else:
-            raise NotImplementedError(f"{beta_schedule} does is not implemented for {self.__class__}")
+            raise NotImplementedError(f"{beta_schedule} is not implemented for {self.__class__}")
 
         self.alphas = 1.0 - self.betas
         self.alphas_cumprod = torch.cumprod(self.alphas, dim=0)
@@ -170,13 +189,13 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
             if algorithm_type in ["dpmsolver", "dpmsolver++"]:
                 self.register_to_config(algorithm_type="deis")
             else:
-                raise NotImplementedError(f"{algorithm_type} does is not implemented for {self.__class__}")
+                raise NotImplementedError(f"{algorithm_type} is not implemented for {self.__class__}")
 
         if solver_type not in ["logrho"]:
             if solver_type in ["midpoint", "heun", "bh1", "bh2"]:
                 self.register_to_config(solver_type="logrho")
             else:
-                raise NotImplementedError(f"solver type {solver_type} does is not implemented for {self.__class__}")
+                raise NotImplementedError(f"solver type {solver_type} is not implemented for {self.__class__}")
 
         # setable values
         self.num_inference_steps = None
@@ -191,7 +210,7 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
     @property
     def step_index(self):
         """
-        The index counter for current timestep. It will increae 1 after each scheduler step.
+        The index counter for current timestep. It will increase 1 after each scheduler step.
         """
         return self._step_index
 
@@ -249,12 +268,28 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
             )
 
         sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5)
+        log_sigmas = np.log(sigmas)
         if self.config.use_karras_sigmas:
-            log_sigmas = np.log(sigmas)
             sigmas = np.flip(sigmas).copy()
             sigmas = self._convert_to_karras(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
             timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas]).round()
             sigmas = np.concatenate([sigmas, sigmas[-1:]]).astype(np.float32)
+        elif self.config.use_exponential_sigmas:
+            sigmas = np.flip(sigmas).copy()
+            sigmas = self._convert_to_exponential(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+            timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas])
+            sigmas = np.concatenate([sigmas, sigmas[-1:]]).astype(np.float32)
+        elif self.config.use_beta_sigmas:
+            sigmas = np.flip(sigmas).copy()
+            sigmas = self._convert_to_beta(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+            timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas])
+            sigmas = np.concatenate([sigmas, sigmas[-1:]]).astype(np.float32)
+        elif self.config.use_flow_sigmas:
+            alphas = np.linspace(1, 1 / self.config.num_train_timesteps, num_inference_steps + 1)
+            sigmas = 1.0 - alphas
+            sigmas = np.flip(self.config.flow_shift * sigmas / (1 + (self.config.flow_shift - 1) * sigmas))[:-1].copy()
+            timesteps = (sigmas * self.config.num_train_timesteps).copy()
+            sigmas = np.concatenate([sigmas, sigmas[-1:]]).astype(np.float32)
         else:
             sigmas = np.interp(timesteps, np.arange(0, len(sigmas)), sigmas)
             sigma_last = ((1 - self.alphas_cumprod[0]) / self.alphas_cumprod[0]) ** 0.5
@@ -276,7 +311,7 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
         self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication
 
     # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler._threshold_sample
-    def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
+    def _threshold_sample(self, sample: torch.Tensor) -> torch.Tensor:
         """
         "Dynamic thresholding: At each sampling step we set s to a certain percentile absolute pixel value in xt0 (the
         prediction of x_0 at timestep t), and if s > 1, then we threshold xt0 to the range [-s, s] and then divide by
@@ -335,13 +370,17 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
 
     # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler._sigma_to_alpha_sigma_t
     def _sigma_to_alpha_sigma_t(self, sigma):
-        alpha_t = 1 / ((sigma**2 + 1) ** 0.5)
-        sigma_t = sigma * alpha_t
+        if self.config.use_flow_sigmas:
+            alpha_t = 1 - sigma
+            sigma_t = sigma
+        else:
+            alpha_t = 1 / ((sigma**2 + 1) ** 0.5)
+            sigma_t = sigma * alpha_t
 
         return alpha_t, sigma_t
 
     # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_karras
-    def _convert_to_karras(self, in_sigmas: torch.FloatTensor, num_inference_steps) -> torch.FloatTensor:
+    def _convert_to_karras(self, in_sigmas: torch.Tensor, num_inference_steps) -> torch.Tensor:
         """Constructs the noise schedule of Karras et al. (2022)."""
 
         # Hack to make sure that other schedulers which copy this function don't break
@@ -366,26 +405,80 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
         sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho
         return sigmas
 
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_exponential
+    def _convert_to_exponential(self, in_sigmas: torch.Tensor, num_inference_steps: int) -> torch.Tensor:
+        """Constructs an exponential noise schedule."""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        sigmas = np.exp(np.linspace(math.log(sigma_max), math.log(sigma_min), num_inference_steps))
+        return sigmas
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_beta
+    def _convert_to_beta(
+        self, in_sigmas: torch.Tensor, num_inference_steps: int, alpha: float = 0.6, beta: float = 0.6
+    ) -> torch.Tensor:
+        """From "Beta Sampling is All You Need" [arXiv:2407.12173] (Lee et. al, 2024)"""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        sigmas = np.array(
+            [
+                sigma_min + (ppf * (sigma_max - sigma_min))
+                for ppf in [
+                    scipy.stats.beta.ppf(timestep, alpha, beta)
+                    for timestep in 1 - np.linspace(0, 1, num_inference_steps)
+                ]
+            ]
+        )
+        return sigmas
+
     def convert_model_output(
         self,
-        model_output: torch.FloatTensor,
+        model_output: torch.Tensor,
         *args,
-        sample: torch.FloatTensor = None,
+        sample: torch.Tensor = None,
         **kwargs,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         """
         Convert the model output to the corresponding type the DEIS algorithm needs.
 
         Args:
-            model_output (`torch.FloatTensor`):
+            model_output (`torch.Tensor`):
                 The direct output from the learned diffusion model.
             timestep (`int`):
                 The current discrete timestep in the diffusion chain.
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 A current instance of a sample created by the diffusion process.
 
         Returns:
-            `torch.FloatTensor`:
+            `torch.Tensor`:
                 The converted model output.
         """
         timestep = args[0] if len(args) > 0 else kwargs.pop("timestep", None)
@@ -409,10 +502,13 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
             x0_pred = model_output
         elif self.config.prediction_type == "v_prediction":
             x0_pred = alpha_t * sample - sigma_t * model_output
+        elif self.config.prediction_type == "flow_prediction":
+            sigma_t = self.sigmas[self.step_index]
+            x0_pred = sample - sigma_t * model_output
         else:
             raise ValueError(
-                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample`, or"
-                " `v_prediction` for the DEISMultistepScheduler."
+                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample`, "
+                "`v_prediction`, or `flow_prediction` for the DEISMultistepScheduler."
             )
 
         if self.config.thresholding:
@@ -425,26 +521,26 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
 
     def deis_first_order_update(
         self,
-        model_output: torch.FloatTensor,
+        model_output: torch.Tensor,
         *args,
-        sample: torch.FloatTensor = None,
+        sample: torch.Tensor = None,
         **kwargs,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         """
         One step for the first-order DEIS (equivalent to DDIM).
 
         Args:
-            model_output (`torch.FloatTensor`):
+            model_output (`torch.Tensor`):
                 The direct output from the learned diffusion model.
             timestep (`int`):
                 The current discrete timestep in the diffusion chain.
             prev_timestep (`int`):
                 The previous discrete timestep in the diffusion chain.
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 A current instance of a sample created by the diffusion process.
 
         Returns:
-            `torch.FloatTensor`:
+            `torch.Tensor`:
                 The sample tensor at the previous timestep.
         """
         timestep = args[0] if len(args) > 0 else kwargs.pop("timestep", None)
@@ -483,22 +579,22 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
 
     def multistep_deis_second_order_update(
         self,
-        model_output_list: List[torch.FloatTensor],
+        model_output_list: List[torch.Tensor],
         *args,
-        sample: torch.FloatTensor = None,
+        sample: torch.Tensor = None,
         **kwargs,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         """
         One step for the second-order multistep DEIS.
 
         Args:
-            model_output_list (`List[torch.FloatTensor]`):
+            model_output_list (`List[torch.Tensor]`):
                 The direct outputs from learned diffusion model at current and latter timesteps.
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 A current instance of a sample created by the diffusion process.
 
         Returns:
-            `torch.FloatTensor`:
+            `torch.Tensor`:
                 The sample tensor at the previous timestep.
         """
         timestep_list = args[0] if len(args) > 0 else kwargs.pop("timestep_list", None)
@@ -552,22 +648,22 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
 
     def multistep_deis_third_order_update(
         self,
-        model_output_list: List[torch.FloatTensor],
+        model_output_list: List[torch.Tensor],
         *args,
-        sample: torch.FloatTensor = None,
+        sample: torch.Tensor = None,
         **kwargs,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         """
         One step for the third-order multistep DEIS.
 
         Args:
-            model_output_list (`List[torch.FloatTensor]`):
+            model_output_list (`List[torch.Tensor]`):
                 The direct outputs from learned diffusion model at current and latter timesteps.
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 A current instance of a sample created by diffusion process.
 
         Returns:
-            `torch.FloatTensor`:
+            `torch.Tensor`:
                 The sample tensor at the previous timestep.
         """
 
@@ -673,9 +769,9 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
 
     def step(
         self,
-        model_output: torch.FloatTensor,
-        timestep: int,
-        sample: torch.FloatTensor,
+        model_output: torch.Tensor,
+        timestep: Union[int, torch.Tensor],
+        sample: torch.Tensor,
         return_dict: bool = True,
     ) -> Union[SchedulerOutput, Tuple]:
         """
@@ -683,11 +779,11 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
         the multistep DEIS.
 
         Args:
-            model_output (`torch.FloatTensor`):
+            model_output (`torch.Tensor`):
                 The direct output from learned diffusion model.
-            timestep (`float`):
+            timestep (`int`):
                 The current discrete timestep in the diffusion chain.
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 A current instance of a sample created by the diffusion process.
             return_dict (`bool`):
                 Whether or not to return a [`~schedulers.scheduling_utils.SchedulerOutput`] or `tuple`.
@@ -736,17 +832,17 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
 
         return SchedulerOutput(prev_sample=prev_sample)
 
-    def scale_model_input(self, sample: torch.FloatTensor, *args, **kwargs) -> torch.FloatTensor:
+    def scale_model_input(self, sample: torch.Tensor, *args, **kwargs) -> torch.Tensor:
         """
         Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
         current timestep.
 
         Args:
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 The input sample.
 
         Returns:
-            `torch.FloatTensor`:
+            `torch.Tensor`:
                 A scaled input sample.
         """
         return sample
@@ -754,10 +850,10 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
     # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.add_noise
     def add_noise(
         self,
-        original_samples: torch.FloatTensor,
-        noise: torch.FloatTensor,
+        original_samples: torch.Tensor,
+        noise: torch.Tensor,
         timesteps: torch.IntTensor,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         # Make sure sigmas and timesteps have the same device and dtype as original_samples
         sigmas = self.sigmas.to(device=original_samples.device, dtype=original_samples.dtype)
         if original_samples.device.type == "mps" and torch.is_floating_point(timesteps):
@@ -768,10 +864,14 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
             schedule_timesteps = self.timesteps.to(original_samples.device)
             timesteps = timesteps.to(original_samples.device)
 
-        # begin_index is None when the scheduler is used for training
+        # begin_index is None when the scheduler is used for training or pipeline does not implement set_begin_index
         if self.begin_index is None:
             step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timesteps]
+        elif self.step_index is not None:
+            # add_noise is called after first denoising step (for inpainting)
+            step_indices = [self.step_index] * timesteps.shape[0]
         else:
+            # add noise is called before first denoising step to create initial latent(img2img)
             step_indices = [self.begin_index] * timesteps.shape[0]
 
         sigma = sigmas[step_indices].flatten()