diff --git a/vllm_ascend/envs.py b/vllm_ascend/envs.py
index 74e9c19..02ecd66 100644
--- a/vllm_ascend/envs.py
+++ b/vllm_ascend/envs.py
@@ -111,15 +111,6 @@ env_variables: Dict[str, Callable[[], Any]] = {
     #   1: enable moe_all2all_buffer.
     "MOE_ALL2ALL_BUFFER":
     lambda: bool(int(os.getenv("MOE_ALL2ALL_BUFFER", '0'))),
-    # VLLM_ASCEND_ACL_OP_INIT_MODE:
-    #   0: default, normal init.
-    #   1: delay init until launch aclops.
-    #   2: forbid aclops init and launch.
-    # Find more details at https://gitee.com/ascend/pytorch/pulls/18094
-    # We set this var default to `1` in vllm-ascend to avoid segment fault when
-    # enable `pin_memory` while creating a tensor using `torch.tensor`.
-    "VLLM_ASCEND_ACL_OP_INIT_MODE":
-    lambda: os.getenv("VLLM_ASCEND_ACL_OP_INIT_MODE", '0'),
     # Some models are optimized by vllm ascend. While in some case, e.g. rlhf
     # training, the optimized model may not be suitable. In this case, set this
     # value to False to disable the optimized model.
diff --git a/vllm_ascend/platform.py b/vllm_ascend/platform.py
index 5a45e9e..b9233da 100644
--- a/vllm_ascend/platform.py
+++ b/vllm_ascend/platform.py
@@ -27,7 +27,6 @@ from torch.distributed.distributed_c10d import PrefixStore
 from vllm.logger import logger
 from vllm.platforms import Platform, PlatformEnum
 
-import vllm_ascend.envs as ascend_envs
 from vllm_ascend.ascend_config import check_ascend_config, init_ascend_config
 from vllm_ascend.utils import ASCEND_QUATIZATION_METHOD, update_aclgraph_sizes
 
@@ -39,8 +38,6 @@ else:
     VllmConfig = None
     FlexibleArgumentParser = None
 
-os.environ["ACL_OP_INIT_MODE"] = ascend_envs.VLLM_ASCEND_ACL_OP_INIT_MODE
-
 
 class NPUPlatform(Platform):
 
@@ -188,6 +185,9 @@ class NPUPlatform(Platform):
             if envs.VLLM_USE_V1:
                 parallel_config.worker_cls = "vllm_ascend.worker.worker_v1.NPUWorker"
             elif vllm_config.speculative_config:
+                # NOTE: We set this var to `1` in vllm-ascend to avoid segment
+                # fault when using spec decode with V0 engine.
+                os.environ["ACL_OP_INIT_MODE"] = "1"
                 parallel_config.worker_cls = "vllm.spec_decode.spec_decode_worker.create_spec_worker"
                 parallel_config.sd_worker_cls = "vllm_ascend.worker.worker.NPUWorker"
             elif vllm_config.scheduler_config.is_multi_step: