comment out casting to float32

src/euler_scheduler.py

@@ -120,7 +120,7 @@ class MyEulerAncestralDiscreteScheduler(EulerAncestralDiscreteScheduler):
         sigma = self.sigmas[self.step_index]
 
         # Upcast to avoid precision issues when computing prev_sample
-        sample = sample.to(torch.float32)
+        # sample = sample.to(torch.float32)
 
         # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
         if self.config.prediction_type == "epsilon":
@@ -226,7 +226,7 @@ class MyEulerAncestralDiscreteScheduler(EulerAncestralDiscreteScheduler):
         sigma = self.sigmas[self.step_index]
 
         # Upcast to avoid precision issues when computing prev_sample
-        sample = sample.to(torch.float32)
+        # sample = sample.to(torch.float32)
 
         # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
         if self.config.prediction_type == "epsilon":
@@ -394,7 +394,7 @@ class MyEulerAncestralDiscreteScheduler(EulerAncestralDiscreteScheduler):
 
     def get_all_sigmas(self) -> torch.FloatTensor:
         sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5)
-        sigmas = np.concatenate([sigmas[::-1], [0.0]]).astype(np.float32)
+        sigmas = np.concatenate([sigmas[::-1], [0.0]])#.astype(np.float32)
         return torch.from_numpy(sigmas)
 
     def add_noise_off_schedule(
@@ -408,7 +408,7 @@ class MyEulerAncestralDiscreteScheduler(EulerAncestralDiscreteScheduler):
         sigmas = sigmas.to(device=original_samples.device, dtype=original_samples.dtype)
         if original_samples.device.type == "mps" and torch.is_floating_point(timesteps):
             # mps does not support float64
-            timesteps = timesteps.to(original_samples.device, dtype=torch.float32)
+            timesteps = timesteps.to(original_samples)#.device, dtype=torch.float32)
         else:
             timesteps = timesteps.to(original_samples.device)