diff --git a/vllm/v1/core/encoder_cache_manager.py b/vllm/v1/core/encoder_cache_manager.py
index 651bc01aa5cf..13ad14e45b32 100644
--- a/vllm/v1/core/encoder_cache_manager.py
+++ b/vllm/v1/core/encoder_cache_manager.py
@@ -54,7 +54,7 @@ def free_encoder_input(self, request: Request, input_id: int) -> None:
 
     def free(self, request: Request) -> None:
         """Free all cached input ids for the request."""
-        input_ids = self.get_cached_input_ids(request)
+        input_ids = self.get_cached_input_ids(request).copy()
         for input_id in input_ids:
             self.free_encoder_input(request, input_id)