diff --git a/vllm/v1/worker/gpu/model_runner.py b/vllm/v1/worker/gpu/model_runner.py
index 0ce0288f31..fe423d6e38 100644
--- a/vllm/v1/worker/gpu/model_runner.py
+++ b/vllm/v1/worker/gpu/model_runner.py
@@ -1128,7 +1128,6 @@ class GPUModelRunner(LoRAModelRunnerMixin):
 
             if self.pooling_runner is not None:
                 assert new_req_data.pooling_params is not None
-                assert new_req_data.prompt_token_ids is not None
                 self.pooling_runner.add_request(
                     req_id,
                     req_index,
diff --git a/vllm/v1/worker/gpu/pool/pooling_runner.py b/vllm/v1/worker/gpu/pool/pooling_runner.py
index 839edc8b8c..ce94764851 100644
--- a/vllm/v1/worker/gpu/pool/pooling_runner.py
+++ b/vllm/v1/worker/gpu/pool/pooling_runner.py
@@ -64,7 +64,7 @@ class PoolingRunner:
         req_id: str,
         req_index: int,
         pooling_params: PoolingParams,
-        prompt_token_ids: list[int],
+        prompt_token_ids: list[int] | None,
     ) -> None:
         task = pooling_params.task
         if task not in self.supported_tasks:
@@ -77,6 +77,7 @@ class PoolingRunner:
         self.pooling_params[req_index] = pooling_params
         self.pooling_states[req_index] = PoolingStates()
         if pooling_params.requires_token_ids:
+            assert prompt_token_ids is not None
             self.prompt_token_ids[req_index] = torch.tensor(
                 prompt_token_ids, dtype=torch.int64
             )