diff --git a/examples/disaggregated_prefill_v1/load_balance_proxy_layerwise_server_example.py b/examples/disaggregated_prefill_v1/load_balance_proxy_layerwise_server_example.py
index ea0c5be..67c34ee 100644
--- a/examples/disaggregated_prefill_v1/load_balance_proxy_layerwise_server_example.py
+++ b/examples/disaggregated_prefill_v1/load_balance_proxy_layerwise_server_example.py
@@ -98,7 +98,6 @@ from typing import List
 import httpx
 from fastapi import FastAPI, Request
 from fastapi.responses import StreamingResponse
-from transformers import AutoTokenizer
 from vllm.logger import init_logger
 
 logger = init_logger(__name__)
@@ -153,8 +152,6 @@ class ProxyState:
         heapq.heapify(self.decoder_heap)
         self.req_id_future = {}
         self.req_data_dict = {}
-        self.tokenizer = AutoTokenizer.from_pretrained(
-            global_args.tokenizer_dir)
 
     def _update_prefiller_priority(self, server_idx: int):
         """Update the priority of a prefiller server in the heap."""
@@ -281,10 +278,6 @@ def parse_args():
                         nargs="+",
                         default=["localhost"])
     parser.add_argument("--decoder-ports", type=int, nargs="+", default=[8002])
-    parser.add_argument("--tokenizer-dir",
-                        type=str,
-                        default="/mnt/weight/Qwen3-235B-A22B-W8A8",
-                        help="Maximum number of retries for HTTP requests")
     parser.add_argument("--max-retries",
                         type=int,
                         default=3,
diff --git a/vllm_ascend/distributed/mooncake_layerwise_connector.py b/vllm_ascend/distributed/mooncake_layerwise_connector.py
index e500f53..7f7c464 100644
--- a/vllm_ascend/distributed/mooncake_layerwise_connector.py
+++ b/vllm_ascend/distributed/mooncake_layerwise_connector.py
@@ -656,7 +656,7 @@ class MooncakeLayerwiseConnectorWorker:
         self.device_id = device_ids[self.tp_rank]  # type: ignore
 
         if vllm_config.kv_transfer_config.get_from_extra_config(
-                'use_ascend_direct', True):
+                'use_ascend_direct', False):
             hostname = self.side_channel_host
         else:
             hostname = f"{self.side_channel_host}:0:npu_{self.device_id}"