diff --git a/python/sglang/srt/managers/tokenizer_manager.py b/python/sglang/srt/managers/tokenizer_manager.py
index a3c5001e87a1..46345efdad54 100644
--- a/python/sglang/srt/managers/tokenizer_manager.py
+++ b/python/sglang/srt/managers/tokenizer_manager.py
@@ -479,6 +479,7 @@ def init_request_dispatcher(self):
 
         self.sampling_params_class = SamplingParams
         self.signal_handler_class = SignalHandler
+        self.req_state_class = ReqState
 
     async def generate_request(
         self,
@@ -1056,7 +1057,9 @@ def _send_one_request(
         trace_slice_start(RequestStage.TOKENIZER_DISPATCH, obj.rid)
         tokenized_obj.trace_context = trace_get_proc_propagate_context(obj.rid)
         self.send_to_scheduler.send_pyobj(tokenized_obj)
-        state = ReqState([], False, asyncio.Event(), obj, created_time=created_time)
+        state = self.req_state_class(
+            [], False, asyncio.Event(), obj, created_time=created_time
+        )
         state.request_sent_to_scheduler_ts = time.time()
         self.rid_to_state[obj.rid] = state
         trace_slice_end(
@@ -1082,7 +1085,7 @@ def _send_batch_request(
         # Create states for each individual request in the batch
         for i, tokenized_obj in enumerate(tokenized_objs):
             tmp_obj = obj[i]
-            state = ReqState(
+            state = self.req_state_class(
                 [], False, asyncio.Event(), tmp_obj, created_time=created_time
             )
             self.rid_to_state[tmp_obj.rid] = state