EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[BugFix] Fix the port conflict bug of running external dp with disaggregated-prefill. (#3416)

Browse Source 
This PR fixes the port conflict bug of running external dp in
disaggregated-prefill scenario.

- vLLM version: v0.11.0

Signed-off-by: whx-sjtu <2952154980@qq.com>
This commit is contained in:
whx
2025-10-14 16:37:10 +08:00
committed by GitHub
parent 9eb62935b8
commit ee25a517d1
1 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm_ascend/distributed/llmdatadist_c_mgr_connector.py
View File

@@ -180,7 +180,10 @@ class LLMDataDistCMgrConnectorScheduler():
# Can not retrieve the parallel config since it is not initialized. # Can not retrieve the parallel config since it is not initialized.
self.local_dp_rank = None self.local_dp_rank = None
self.tp_size = None self.tp_size = None
dp_rank_local = self.vllm_config.parallel_config.data_parallel_rank_local if vllm_config.parallel_config.data_parallel_external_lb:
dp_rank_local = vllm_config.parallel_config.data_parallel_rank
else:
dp_rank_local = vllm_config.parallel_config.data_parallel_rank_local
tp_size = self.vllm_config.parallel_config.tensor_parallel_size tp_size = self.vllm_config.parallel_config.tensor_parallel_size
self.port = dp_rank_local * tp_size + envs_ascend.VLLM_ASCEND_LLMDD_RPC_PORT if dp_rank_local is not None else tp_size + envs_ascend.VLLM_ASCEND_LLMDD_RPC_PORT self.port = dp_rank_local * tp_size + envs_ascend.VLLM_ASCEND_LLMDD_RPC_PORT if dp_rank_local is not None else tp_size + envs_ascend.VLLM_ASCEND_LLMDD_RPC_PORT
@@ -312,6 +315,9 @@ class LLMDataDistCMgrConnectorWorker():
vllm_config.parallel_config.data_parallel_size_local * vllm_config.parallel_config.data_parallel_size_local *
vllm_config.parallel_config.tensor_parallel_size) vllm_config.parallel_config.tensor_parallel_size)
self.local_rank = get_world_group().local_rank self.local_rank = get_world_group().local_rank
if vllm_config.parallel_config.data_parallel_external_lb:
self.local_dp_rank = vllm_config.parallel_config.data_parallel_rank
else:
self.local_dp_rank = vllm_config.parallel_config.data_parallel_rank_local self.local_dp_rank = vllm_config.parallel_config.data_parallel_rank_local
self.tp_size = vllm_config.parallel_config.tensor_parallel_size self.tp_size = vllm_config.parallel_config.tensor_parallel_size
self.tp_rank = get_tp_group().rank_in_group self.tp_rank = get_tp_group().rank_in_group