@@ -174,8 +174,7 @@ class ConditionalCFM(BASECFM):
# random timestep
t = torch.rand([b, 1, 1], device=mu.device, dtype=mu.dtype)
- if self.t_scheduler == 'cosine':
- t = 1 - torch.cos(t * 0.5 * torch.pi)
+
# sample noise p(x_0)
z = torch.randn_like(x1)