biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[BugFix] Fix async EPLB hang with DeepEP LL all2all backend (#32860)

Browse Source 
Signed-off-by: ilmarkov <markovilya197@gmail.com>
This commit is contained in:
Ilya Markov
2026-02-10 23:34:47 +01:00
committed by GitHub
parent 67132945bb
commit bb2fc8b5e7
2 changed files with 56 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/worker/gpu_worker.py
View File

@@ -22,6 +22,7 @@ from vllm.distributed import (
set_custom_all_reduce,
)
from vllm.distributed.ec_transfer import ensure_ec_transfer_initialized
from vllm.distributed.eplb.eplb_utils import override_envs_for_eplb
from vllm.distributed.kv_transfer import (
ensure_kv_transfer_initialized,
ensure_kv_transfer_shutdown,
@@ -1035,6 +1036,7 @@ def init_worker_distributed_environment(
from vllm.model_executor.layers.batch_invariant import init_batch_invariance
init_batch_invariance(attention_config.backend)
override_envs_for_eplb(parallel_config)
set_custom_all_reduce(not parallel_config.disable_custom_all_reduce)
init_method = distributed_init_method or "env://"