[2/N] Elastic EP Milestone 2: Integrating NIXL-EP (#35627)

Signed-off-by: Itay Alroy <ialroy@nvidia.com>
Co-authored-by: Yongji Wu <wuyongji317@gmail.com>
Co-authored-by: Ron Tourgeman <rtourgeman@nvidia.com>
This commit is contained in:
Itay Alroy
2026-03-13 09:25:33 -04:00
committed by GitHub
co-authored by Yongji Wu Ron Tourgeman
parent 82f836d976
commit d5af196c18
14 changed files with 635 additions and 11 deletions
@@ -896,7 +896,9 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
# batched activation format. As self.fused_experts is not
# initialized at this point, we resort to checking the MoE config
# directly.
is_batched_moe = self.moe.use_deepep_ll_kernels
is_batched_moe = (
self.moe.use_deepep_ll_kernels or self.moe.use_nixl_ep_kernels
)
if is_batched_moe:
num_warps = 4 if envs.VLLM_MOE_DP_CHUNK_SIZE <= 512 else 8
else: