Fix UniPC scheduler for 1D (huggingface#5276)

src/diffusers/schedulers/scheduling_ddim.py

@@ -276,25 +276,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample

src/diffusers/schedulers/scheduling_ddim_parallel.py

@@ -298,25 +298,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample

src/diffusers/schedulers/scheduling_ddpm.py

@@ -330,25 +330,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample

src/diffusers/schedulers/scheduling_ddpm_parallel.py

@@ -344,25 +344,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample

src/diffusers/schedulers/scheduling_deis_multistep.py

@@ -268,25 +268,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample

src/diffusers/schedulers/scheduling_dpmsolver_multistep.py

@@ -288,25 +288,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample

src/diffusers/schedulers/scheduling_dpmsolver_multistep_inverse.py

@@ -298,25 +298,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample

src/diffusers/schedulers/scheduling_dpmsolver_singlestep.py

@@ -302,25 +302,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample

src/diffusers/schedulers/scheduling_unipc_multistep.py

@@ -282,25 +282,24 @@ def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
         https://arxiv.org/abs/2205.11487
         """
         dtype = sample.dtype
-        batch_size, channels, height, width = sample.shape
+        batch_size, channels, *remaining_dims = sample.shape
 
         if dtype not in (torch.float32, torch.float64):
             sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
 
         # Flatten sample for doing quantile calculation along each image
-        sample = sample.reshape(batch_size, channels * height * width)
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
 
         abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
 
         s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
         s = torch.clamp(
             s, min=1, max=self.config.sample_max_value
         )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
-
         s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
         sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
 
-        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
         sample = sample.to(dtype)
 
         return sample