[BugFix] Remove to(device) for clip_value usage as it's a buffer already

ghstack-source-id: ec50cf8
Pull-Request-resolved: #2960

Author: Vincent Moens

torchrl/objectives/ppo.py

@@ -677,7 +677,7 @@ def loss_critic(self, tensordict: TensorDictBase) -> torch.Tensor:
             loss_value, clip_fraction = _clip_value_loss(
                 old_state_value,
                 state_value,
-                self.clip_value.to(state_value.device),
+                self.clip_value,
                 target_return,
                 loss_value,
                 self.loss_critic_type,