diff --git a/server/lorax_server/models/flash_causal_lm.py b/server/lorax_server/models/flash_causal_lm.py
index 4bc02387f..91be95a86 100644
--- a/server/lorax_server/models/flash_causal_lm.py
+++ b/server/lorax_server/models/flash_causal_lm.py
@@ -941,8 +941,6 @@ def decode(self, generated_ids: Union[torch.Tensor, List[int]]) -> str:
         )
 
     def forward(self, batch: FlashCausalLMBatch, adapter_data: AdapterBatchData) -> Tuple[torch.Tensor, torch.Tensor]:
-        global CACHE_MANAGER
-
         prefill = batch.cu_seqlen_prefill is not None
         model = self.model
         if (