fix pause reqs

Author: shihaobai

lightllm/server/router/manager.py

@@ -307,7 +307,7 @@ async def _step(self):
             paused_reqs = select_paused_reqs(
                 self.running_batch, self.pause_strategy, self.req_queue, self.max_total_token_num
             )
-            await self._pause_reqs(self.running_batch, paused_reqs)
+            await self._pause_reqs(paused_reqs)
             logger.debug(f"pasued req num: {self.req_queue.get_paused_req_num()}")
             self.has_wait_tokens = 0
             return
@@ -342,9 +342,9 @@ async def _decode_batch(self, batch: Batch):
         )
         return
 
-    async def _pause_reqs(self, batch: Batch, pasue_reqs):
+    async def _pause_reqs(self, pasue_reqs):
         pasue_req_ids = [r.request_id for r in pasue_reqs]
-        await self.model_rpc_client.pause_reqs(batch.batch_id, pasue_req_ids)
+        await self.model_rpc_client.pause_reqs(pasue_req_ids)
         return
 
     def _filter_runing_batch(self):

lightllm/server/router/model_infer/mode_backend/chunked_prefill/impl.py

@@ -54,6 +54,7 @@ def post_handel(self, run_reqs: List[InferReq], next_token_ids, next_token_logpr
 
             req_obj.cur_kv_len = len(req_obj.get_chuncked_input_token_ids())
             if req_obj.cur_kv_len < req_obj.get_cur_total_len():
+                req_obj.shm_req.shm_cur_kv_len = req_obj.cur_kv_len
                 continue
 
             req_obj.set_next_gen_token_id(next_token_id, next_token_logprob)