fix

Author: hiworldwzj

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

@@ -54,7 +54,8 @@ 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
+                if self.tp_rank < self.dp_size:
+                    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)