fix prefix and latent split

perceiverar/module.py

@@ -1,4 +1,3 @@
-import pdb
 from typing import List, Optional, Tuple
 
 import torch
@@ -411,7 +410,7 @@ class PerceiverARModel(nn.Module):
         perciever_input = torch.cat((lags, features), dim=-1)
 
         prefix, x = (
-            perciever_input,
+            perciever_input[:, : self.context_length - 1, ...],,
             perciever_input[:, self.context_length - 1 :, ...],
         )