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

diffusers/schedulers/scheduling_ddim_inverse.py

@@ -33,16 +33,16 @@ class DDIMSchedulerOutput(BaseOutput):
     Output class for the scheduler's `step` function output.
 
     Args:
-        prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+        prev_sample (`torch.Tensor` of shape `(batch_size, num_channels, height, width)` for images):
             Computed sample `(x_{t-1})` of previous timestep. `prev_sample` should be used as next model input in the
             denoising loop.
-        pred_original_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+        pred_original_sample (`torch.Tensor` of shape `(batch_size, num_channels, height, width)` for images):
             The predicted denoised sample `(x_{0})` based on the model output from the current timestep.
             `pred_original_sample` can be used to preview progress or for guidance.
     """
 
-    prev_sample: torch.FloatTensor
-    pred_original_sample: Optional[torch.FloatTensor] = None
+    prev_sample: torch.Tensor
+    pred_original_sample: Optional[torch.Tensor] = None
 
 
 # Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar
@@ -80,7 +80,7 @@ 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):
@@ -97,11 +97,11 @@ def rescale_zero_terminal_snr(betas):
 
 
     Args:
-        betas (`torch.FloatTensor`):
+        betas (`torch.Tensor`):
             the betas that the scheduler is being initialized with.
 
     Returns:
-        `torch.FloatTensor`: rescaled betas with zero terminal SNR
+        `torch.Tensor`: rescaled betas with zero terminal SNR
     """
     # Convert betas to alphas_bar_sqrt
     alphas = 1.0 - betas
@@ -207,7 +207,7 @@ class DDIMInverseScheduler(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__}")
 
         # Rescale for zero SNR
         if rescale_betas_zero_snr:
@@ -231,19 +231,19 @@ class DDIMInverseScheduler(SchedulerMixin, ConfigMixin):
         self.timesteps = torch.from_numpy(np.arange(0, num_train_timesteps).copy().astype(np.int64))
 
     # Copied from diffusers.schedulers.scheduling_ddim.DDIMScheduler.scale_model_input
-    def scale_model_input(self, sample: torch.FloatTensor, timestep: Optional[int] = None) -> torch.FloatTensor:
+    def scale_model_input(self, sample: torch.Tensor, timestep: Optional[int] = None) -> 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.
             timestep (`int`, *optional*):
                 The current timestep in the diffusion chain.
 
         Returns:
-            `torch.FloatTensor`:
+            `torch.Tensor`:
                 A scaled input sample.
         """
         return sample
@@ -288,9 +288,9 @@ class DDIMInverseScheduler(SchedulerMixin, ConfigMixin):
 
     def step(
         self,
-        model_output: torch.FloatTensor,
+        model_output: torch.Tensor,
         timestep: int,
-        sample: torch.FloatTensor,
+        sample: torch.Tensor,
         return_dict: bool = True,
     ) -> Union[DDIMSchedulerOutput, Tuple]:
         """
@@ -298,11 +298,11 @@ class DDIMInverseScheduler(SchedulerMixin, ConfigMixin):
         process from the learned model outputs (most often the predicted noise).
 
         Args:
-            model_output (`torch.FloatTensor`):
+            model_output (`torch.Tensor`):
                 The direct output from learned diffusion model.
             timestep (`float`):
                 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.
             eta (`float`):
                 The weight of noise for added noise in diffusion step.
@@ -311,7 +311,7 @@ class DDIMInverseScheduler(SchedulerMixin, ConfigMixin):
                 because predicted original sample is clipped to [-1, 1] when `self.config.clip_sample` is `True`. If no
                 clipping has happened, "corrected" `model_output` would coincide with the one provided as input and
                 `use_clipped_model_output` has no effect.
-            variance_noise (`torch.FloatTensor`):
+            variance_noise (`torch.Tensor`):
                 Alternative to generating noise with `generator` by directly providing the noise for the variance
                 itself. Useful for methods such as [`CycleDiffusion`].
             return_dict (`bool`, *optional*, defaults to `True`):

diffusers/schedulers/scheduling_ddim_parallel.py

@@ -35,16 +35,16 @@ class DDIMParallelSchedulerOutput(BaseOutput):
     Output class for the scheduler's `step` function output.
 
     Args:
-        prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+        prev_sample (`torch.Tensor` of shape `(batch_size, num_channels, height, width)` for images):
             Computed sample `(x_{t-1})` of previous timestep. `prev_sample` should be used as next model input in the
             denoising loop.
-        pred_original_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+        pred_original_sample (`torch.Tensor` of shape `(batch_size, num_channels, height, width)` for images):
             The predicted denoised sample `(x_{0})` based on the model output from the current timestep.
             `pred_original_sample` can be used to preview progress or for guidance.
     """
 
-    prev_sample: torch.FloatTensor
-    pred_original_sample: Optional[torch.FloatTensor] = None
+    prev_sample: torch.Tensor
+    pred_original_sample: Optional[torch.Tensor] = None
 
 
 # Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar
@@ -82,7 +82,7 @@ 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):
@@ -99,11 +99,11 @@ def rescale_zero_terminal_snr(betas):
 
 
     Args:
-        betas (`torch.FloatTensor`):
+        betas (`torch.Tensor`):
             the betas that the scheduler is being initialized with.
 
     Returns:
-        `torch.FloatTensor`: rescaled betas with zero terminal SNR
+        `torch.Tensor`: rescaled betas with zero terminal SNR
     """
     # Convert betas to alphas_bar_sqrt
     alphas = 1.0 - betas
@@ -218,7 +218,7 @@ class DDIMParallelScheduler(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__}")
 
         # Rescale for zero SNR
         if rescale_betas_zero_snr:
@@ -241,19 +241,19 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
         self.timesteps = torch.from_numpy(np.arange(0, num_train_timesteps)[::-1].copy().astype(np.int64))
 
     # Copied from diffusers.schedulers.scheduling_ddim.DDIMScheduler.scale_model_input
-    def scale_model_input(self, sample: torch.FloatTensor, timestep: Optional[int] = None) -> torch.FloatTensor:
+    def scale_model_input(self, sample: torch.Tensor, timestep: Optional[int] = None) -> 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.
             timestep (`int`, *optional*):
                 The current timestep in the diffusion chain.
 
         Returns:
-            `torch.FloatTensor`:
+            `torch.Tensor`:
                 A scaled input sample.
         """
         return sample
@@ -283,7 +283,7 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
         return variance
 
     # 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
@@ -364,13 +364,13 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
 
     def step(
         self,
-        model_output: torch.FloatTensor,
+        model_output: torch.Tensor,
         timestep: int,
-        sample: torch.FloatTensor,
+        sample: torch.Tensor,
         eta: float = 0.0,
         use_clipped_model_output: bool = False,
         generator=None,
-        variance_noise: Optional[torch.FloatTensor] = None,
+        variance_noise: Optional[torch.Tensor] = None,
         return_dict: bool = True,
     ) -> Union[DDIMParallelSchedulerOutput, Tuple]:
         """
@@ -378,9 +378,9 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
         process from the learned model outputs (most often the predicted noise).
 
         Args:
-            model_output (`torch.FloatTensor`): direct output from learned diffusion model.
+            model_output (`torch.Tensor`): direct output from learned diffusion model.
             timestep (`int`): current discrete timestep in the diffusion chain.
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 current instance of sample being created by diffusion process.
             eta (`float`): weight of noise for added noise in diffusion step.
             use_clipped_model_output (`bool`): if `True`, compute "corrected" `model_output` from the clipped
@@ -388,7 +388,7 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
                 `self.config.clip_sample` is `True`. If no clipping has happened, "corrected" `model_output` would
                 coincide with the one provided as input and `use_clipped_model_output` will have not effect.
             generator: random number generator.
-            variance_noise (`torch.FloatTensor`): instead of generating noise for the variance using `generator`, we
+            variance_noise (`torch.Tensor`): instead of generating noise for the variance using `generator`, we
                 can directly provide the noise for the variance itself. This is useful for methods such as
                 CycleDiffusion. (https://arxiv.org/abs/2210.05559)
             return_dict (`bool`): option for returning tuple rather than DDIMParallelSchedulerOutput class
@@ -480,18 +480,21 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
             prev_sample = prev_sample + variance
 
         if not return_dict:
-            return (prev_sample,)
+            return (
+                prev_sample,
+                pred_original_sample,
+            )
 
         return DDIMParallelSchedulerOutput(prev_sample=prev_sample, pred_original_sample=pred_original_sample)
 
     def batch_step_no_noise(
         self,
-        model_output: torch.FloatTensor,
+        model_output: torch.Tensor,
         timesteps: List[int],
-        sample: torch.FloatTensor,
+        sample: torch.Tensor,
         eta: float = 0.0,
         use_clipped_model_output: bool = False,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         """
         Batched version of the `step` function, to be able to reverse the SDE for multiple samples/timesteps at once.
         Also, does not add any noise to the predicted sample, which is necessary for parallel sampling where the noise
@@ -501,10 +504,10 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
         process from the learned model outputs (most often the predicted noise).
 
         Args:
-            model_output (`torch.FloatTensor`): direct output from learned diffusion model.
+            model_output (`torch.Tensor`): direct output from learned diffusion model.
             timesteps (`List[int]`):
                 current discrete timesteps in the diffusion chain. This is now a list of integers.
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 current instance of sample being created by diffusion process.
             eta (`float`): weight of noise for added noise in diffusion step.
             use_clipped_model_output (`bool`): if `True`, compute "corrected" `model_output` from the clipped
@@ -513,7 +516,7 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
                 coincide with the one provided as input and `use_clipped_model_output` will have not effect.
 
         Returns:
-            `torch.FloatTensor`: sample tensor at previous timestep.
+            `torch.Tensor`: sample tensor at previous timestep.
 
         """
         if self.num_inference_steps is None:
@@ -595,10 +598,10 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
     # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.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 alphas_cumprod and timestep have same device and dtype as original_samples
         # Move the self.alphas_cumprod to device to avoid redundant CPU to GPU data movement
         # for the subsequent add_noise calls
@@ -620,9 +623,7 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
         return noisy_samples
 
     # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.get_velocity
-    def get_velocity(
-        self, sample: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor
-    ) -> torch.FloatTensor:
+    def get_velocity(self, sample: torch.Tensor, noise: torch.Tensor, timesteps: torch.IntTensor) -> torch.Tensor:
         # Make sure alphas_cumprod and timestep have same device and dtype as sample
         self.alphas_cumprod = self.alphas_cumprod.to(device=sample.device)
         alphas_cumprod = self.alphas_cumprod.to(dtype=sample.dtype)

diffusers/schedulers/scheduling_ddpm.py

@@ -33,16 +33,16 @@ class DDPMSchedulerOutput(BaseOutput):
     Output class for the scheduler's `step` function output.
 
     Args:
-        prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+        prev_sample (`torch.Tensor` of shape `(batch_size, num_channels, height, width)` for images):
             Computed sample `(x_{t-1})` of previous timestep. `prev_sample` should be used as next model input in the
             denoising loop.
-        pred_original_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+        pred_original_sample (`torch.Tensor` of shape `(batch_size, num_channels, height, width)` for images):
             The predicted denoised sample `(x_{0})` based on the model output from the current timestep.
             `pred_original_sample` can be used to preview progress or for guidance.
     """
 
-    prev_sample: torch.FloatTensor
-    pred_original_sample: Optional[torch.FloatTensor] = None
+    prev_sample: torch.Tensor
+    pred_original_sample: Optional[torch.Tensor] = None
 
 
 def betas_for_alpha_bar(
@@ -79,7 +79,7 @@ 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):
@@ -96,11 +96,11 @@ def rescale_zero_terminal_snr(betas):
 
 
     Args:
-        betas (`torch.FloatTensor`):
+        betas (`torch.Tensor`):
             the betas that the scheduler is being initialized with.
 
     Returns:
-        `torch.FloatTensor`: rescaled betas with zero terminal SNR
+        `torch.Tensor`: rescaled betas with zero terminal SNR
     """
     # Convert betas to alphas_bar_sqrt
     alphas = 1.0 - betas
@@ -194,7 +194,7 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
         sample_max_value: float = 1.0,
         timestep_spacing: str = "leading",
         steps_offset: int = 0,
-        rescale_betas_zero_snr: int = False,
+        rescale_betas_zero_snr: bool = False,
     ):
         if trained_betas is not None:
             self.betas = torch.tensor(trained_betas, dtype=torch.float32)
@@ -211,7 +211,7 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
             betas = torch.linspace(-6, 6, num_train_timesteps)
             self.betas = torch.sigmoid(betas) * (beta_end - beta_start) + beta_start
         else:
-            raise NotImplementedError(f"{beta_schedule} does is not implemented for {self.__class__}")
+            raise NotImplementedError(f"{beta_schedule} is not implemented for {self.__class__}")
 
         # Rescale for zero SNR
         if rescale_betas_zero_snr:
@@ -231,19 +231,19 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
 
         self.variance_type = variance_type
 
-    def scale_model_input(self, sample: torch.FloatTensor, timestep: Optional[int] = None) -> torch.FloatTensor:
+    def scale_model_input(self, sample: torch.Tensor, timestep: Optional[int] = None) -> 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.
             timestep (`int`, *optional*):
                 The current timestep in the diffusion chain.
 
         Returns:
-            `torch.FloatTensor`:
+            `torch.Tensor`:
                 A scaled input sample.
         """
         return sample
@@ -363,7 +363,7 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
 
         return variance
 
-    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
@@ -398,9 +398,9 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
 
     def step(
         self,
-        model_output: torch.FloatTensor,
+        model_output: torch.Tensor,
         timestep: int,
-        sample: torch.FloatTensor,
+        sample: torch.Tensor,
         generator=None,
         return_dict: bool = True,
     ) -> Union[DDPMSchedulerOutput, Tuple]:
@@ -409,11 +409,11 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
         process from the learned model outputs (most often the predicted noise).
 
         Args:
-            model_output (`torch.FloatTensor`):
+            model_output (`torch.Tensor`):
                 The direct output from learned diffusion model.
             timestep (`float`):
                 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.
             generator (`torch.Generator`, *optional*):
                 A random number generator.
@@ -492,16 +492,19 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
         pred_prev_sample = pred_prev_sample + variance
 
         if not return_dict:
-            return (pred_prev_sample,)
+            return (
+                pred_prev_sample,
+                pred_original_sample,
+            )
 
         return DDPMSchedulerOutput(prev_sample=pred_prev_sample, pred_original_sample=pred_original_sample)
 
     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 alphas_cumprod and timestep have same device and dtype as original_samples
         # Move the self.alphas_cumprod to device to avoid redundant CPU to GPU data movement
         # for the subsequent add_noise calls
@@ -522,9 +525,7 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
         noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise
         return noisy_samples
 
-    def get_velocity(
-        self, sample: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor
-    ) -> torch.FloatTensor:
+    def get_velocity(self, sample: torch.Tensor, noise: torch.Tensor, timesteps: torch.IntTensor) -> torch.Tensor:
         # Make sure alphas_cumprod and timestep have same device and dtype as sample
         self.alphas_cumprod = self.alphas_cumprod.to(device=sample.device)
         alphas_cumprod = self.alphas_cumprod.to(dtype=sample.dtype)
@@ -547,16 +548,12 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
         return self.config.num_train_timesteps
 
     def previous_timestep(self, timestep):
-        if self.custom_timesteps:
+        if self.custom_timesteps or self.num_inference_steps:
             index = (self.timesteps == timestep).nonzero(as_tuple=True)[0][0]
             if index == self.timesteps.shape[0] - 1:
                 prev_t = torch.tensor(-1)
             else:
                 prev_t = self.timesteps[index + 1]
         else:
-            num_inference_steps = (
-                self.num_inference_steps if self.num_inference_steps else self.config.num_train_timesteps
-            )
-            prev_t = timestep - self.config.num_train_timesteps // num_inference_steps
-
+            prev_t = timestep - 1
         return prev_t

diffusers/schedulers/scheduling_ddpm_flax.py

@@ -222,9 +222,13 @@ class FlaxDDPMScheduler(FlaxSchedulerMixin, ConfigMixin):
         t = timestep
 
         if key is None:
-            key = jax.random.PRNGKey(0)
+            key = jax.random.key(0)
 
-        if model_output.shape[1] == sample.shape[1] * 2 and self.config.variance_type in ["learned", "learned_range"]:
+        if (
+            len(model_output.shape) > 1
+            and model_output.shape[1] == sample.shape[1] * 2
+            and self.config.variance_type in ["learned", "learned_range"]
+        ):
             model_output, predicted_variance = jnp.split(model_output, sample.shape[1], axis=1)
         else:
             predicted_variance = None
@@ -264,7 +268,7 @@ class FlaxDDPMScheduler(FlaxSchedulerMixin, ConfigMixin):
 
         # 6. Add noise
         def random_variance():
-            split_key = jax.random.split(key, num=1)
+            split_key = jax.random.split(key, num=1)[0]
             noise = jax.random.normal(split_key, shape=model_output.shape, dtype=self.dtype)
             return (self._get_variance(state, t, predicted_variance=predicted_variance) ** 0.5) * noise