feat(training-utils): support device and dtype params in compute_density_for_timestep_sampling

Author: badayvedat

src/diffusers/training_utils.py

@@ -248,7 +248,13 @@ def _set_state_dict_into_text_encoder(
 
 
 def compute_density_for_timestep_sampling(
-    weighting_scheme: str, batch_size: int, logit_mean: float = None, logit_std: float = None, mode_scale: float = None
+    weighting_scheme: str,
+    batch_size: int,
+    logit_mean: float = None,
+    logit_std: float = None,
+    mode_scale: float = None,
+    device: torch.device = "cpu",
+    generator: Optional[torch.Generator] = None,
 ):
     """
     Compute the density for sampling the timesteps when doing SD3 training.
@@ -258,14 +264,13 @@ def compute_density_for_timestep_sampling(
     SD3 paper reference: https://arxiv.org/abs/2403.03206v1.
     """
     if weighting_scheme == "logit_normal":
-        # See 3.1 in the SD3 paper ($rf/lognorm(0.00,1.00)$).
-        u = torch.normal(mean=logit_mean, std=logit_std, size=(batch_size,), device="cpu")
+        u = torch.normal(mean=logit_mean, std=logit_std, size=(batch_size,), device=device, generator=generator)
         u = torch.nn.functional.sigmoid(u)
     elif weighting_scheme == "mode":
-        u = torch.rand(size=(batch_size,), device="cpu")
+        u = torch.rand(size=(batch_size,), device=device, generator=generator)
         u = 1 - u - mode_scale * (torch.cos(math.pi * u / 2) ** 2 - 1 + u)
     else:
-        u = torch.rand(size=(batch_size,), device="cpu")
+        u = torch.rand(size=(batch_size,), device=device, generator=generator)
     return u