fix

Author: kevssim

src/twinkle/model/transformers/transformers.py

@@ -955,9 +955,6 @@ def _get_full_state_dict(self) -> dict:
         the local expert shards across the EP group to reconstruct the
         full expert tensor (all num_experts on dim-0).
         """
-        import torch
-        import torch.distributed as dist
-
         model = self.strategy.unwrap_model(self.model)
         state_dict = {}