[Cherry Pick from pr#3981][0.11.0][P/D]Make kv-transfer env variable take effect & Fix load-balance proxy (#3983)

### What this PR does / why we need it?
Make kv-transfer env variable take effect & Fix load-balance proxy.
Cherry Pick from #3981

---------
Signed-off-by: nwpu-zxr <zhouxuerong2@huawei.com>
This commit is contained in:
zxr2333
2025-11-08 13:52:33 +08:00
committed by GitHub
parent 8e72758645
commit d4e2a44307
7 changed files with 35 additions and 9 deletions

View File

@@ -30,6 +30,7 @@ from vllm.v1.core.sched.output import SchedulerOutput
from vllm.v1.request import Request, RequestStatus
import vllm_ascend.envs as envs_ascend
from vllm_ascend.distributed.utils import get_transfer_timeout_value
from vllm_ascend.utils import AscendSocVersion, get_ascend_soc_version
TORCH_DTYPE_TO_NPU_DTYPE = {
@@ -411,7 +412,7 @@ class LLMDataDistCMgrConnectorWorker():
assert self.local_agent_metadata is not None
llm_config = LLMConfig()
llm_config.device_id = self.local_rank
llm_config.sync_kv_timeout = 20000
llm_config.sync_kv_timeout = get_transfer_timeout_value()
llm_config.enable_switch_role = True
llm_config.enable_cache_manager = True
llm_config.enable_remote_cache_accessible = True