[MoE Refactor] EPLB refactoring for FusedMoE (#41055)

Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
This commit is contained in:
bnellnm
2026-05-12 14:16:31 -04:00
committed by GitHub
co-authored by Robert Shaw
parent 4d591db470
commit d9b4990783
17 changed files with 83 additions and 98 deletions
@@ -10,6 +10,7 @@ import torch
from tests.kernels.moe.utils import make_test_quant_config
from vllm.config import VllmConfig, set_current_vllm_config
from vllm.distributed.eplb.eplb_state import EplbLayerState
from vllm.distributed.eplb.rebalance_execute import rearrange_expert_weights_inplace
from vllm.distributed.parallel_state import (
ensure_model_parallel_initialized,
@@ -201,7 +202,7 @@ def _test_eplb_fml(env, world_size: int, test_config: TestConfig):
dtype=torch.int32,
device=device,
)
fml.enable_eplb = True
fml.eplb_state = EplbLayerState()
fml.set_eplb_state(
lidx,
torch.zeros(