[Qwen4Exp] support Sequence Parallelism for Qwen3.8-Flash-Next (#56322)

Signed-off-by: huanghaoyan.hhy <huanghaoyan.hhy@alibaba-inc.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Harry Huang
2026-10-12 20:56:44 +08:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 258b97ff6b
commit c75cb63cd2
23 changed files with 870 additions and 178 deletions
+3
View File
@@ -538,6 +538,7 @@ class EngineArgs:
data_parallel_external_lb: bool = False
data_parallel_multi_port_external_lb: bool = False
data_parallel_backend: DataParallelBackend = ParallelConfig.data_parallel_backend
enable_hc_sp: bool = ParallelConfig.enable_hc_sp
enable_expert_parallel: bool = ParallelConfig.enable_expert_parallel
enable_batch_sharded_sampling: bool | None = (
ParallelConfig.enable_batch_sharded_sampling
@@ -1258,6 +1259,7 @@ class EngineArgs:
"server per local data parallel rank and exposes aggregated health on "
"a supervisor port.",
)
parallel_group.add_argument("--enable-hc-sp", **parallel_kwargs["enable_hc_sp"])
parallel_group.add_argument(
"--enable-expert-parallel",
"-ep",
@@ -2532,6 +2534,7 @@ class EngineArgs:
data_parallel_backend=self.data_parallel_backend,
data_parallel_hybrid_lb=self.data_parallel_hybrid_lb,
is_moe_model=model_config.is_moe,
enable_hc_sp=self.enable_hc_sp,
enable_expert_parallel=self.enable_expert_parallel,
enable_batch_sharded_sampling=self.enable_batch_sharded_sampling,
enable_ep_weight_filter=self.enable_ep_weight_filter,