[Feature] Support EPLB for DeepSeek v4 Mega Moe (#43339)

Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
Co-authored-by: Wei Zhao (Engrg-Hardware 1) <weizha@login-lyris01.lyris.clusters.nvidia.com>
This commit is contained in:
Wei Zhao
2026-06-02 10:56:44 -07:00
committed by GitHub
co-authored by Wei Zhao
parent fe32e7830b
commit 2427094152
4 changed files with 232 additions and 46 deletions
+4 -1
View File
@@ -1147,7 +1147,10 @@ def init_worker_distributed_environment(
from vllm.model_executor.layers.batch_invariant import init_batch_invariance
init_batch_invariance()
override_envs_for_eplb(parallel_config)
override_envs_for_eplb(
parallel_config,
moe_backend=getattr(vllm_config.kernel_config, "moe_backend", None),
)
set_custom_all_reduce(not parallel_config.disable_custom_all_reduce)
init_method = distributed_init_method or "env://"