diff --git a/vllm/v1/worker/gpu_input_batch.py b/vllm/v1/worker/gpu_input_batch.py
index 5db843e99d6a..bc1186e5feb7 100644
--- a/vllm/v1/worker/gpu_input_batch.py
+++ b/vllm/v1/worker/gpu_input_batch.py
@@ -251,7 +251,6 @@ def __init__(
 
         # Cached reference to the GPU tensor of previously sampled tokens
         self.prev_sampled_token_ids: Optional[torch.Tensor] = None
-        self.prev_sampled_token_ids_invalid_indices: Optional[set[int]] = None
         self.prev_req_id_to_index: Optional[dict[str, int]] = None
 
     @property
diff --git a/vllm/v1/worker/gpu_model_runner.py b/vllm/v1/worker/gpu_model_runner.py
index 0a5b56d95224..b5687c75fc3e 100644
--- a/vllm/v1/worker/gpu_model_runner.py
+++ b/vllm/v1/worker/gpu_model_runner.py
@@ -2305,9 +2305,6 @@ def _bookkeeping_sync(
             # These will be copied into input_ids in the next step
             # when preparing inputs.
             self.input_batch.prev_sampled_token_ids = sampled_token_ids
-            self.input_batch.prev_sampled_token_ids_invalid_indices = (
-                invalid_req_indices_set
-            )
             self.input_batch.prev_req_id_to_index = {
                 req_id: i
                 for i, req_id in enumerate(self.input_batch.req_ids)