diff --git a/torch/distributed/fsdp/_fully_shard/_fsdp_state.py b/torch/distributed/fsdp/_fully_shard/_fsdp_state.py
index 237f59673828..e8ba77e8fa0e 100644
--- a/torch/distributed/fsdp/_fully_shard/_fsdp_state.py
+++ b/torch/distributed/fsdp/_fully_shard/_fsdp_state.py
@@ -230,7 +230,7 @@ class FSDPState(_State):
         self, module: nn.Module, args: tuple[Any, ...], kwargs: dict[str, Any]
     ) -> tuple[tuple[Any, ...], dict[str, Any]]:
         # When composing with module-hook-based activation checkpointing, the
-        # the pre-backward hook is responsible for the unshard
+        # pre-backward hook is responsible for the unshard
         if self._training_state == TrainingState.PRE_BACKWARD:
             return args, kwargs
         self._training_state = TrainingState.FORWARD