[MoE Refactor] Remove MoE DP chunking (#39107)

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
2026-04-14 09:48:05 -04:00
committed by GitHub
co-authored by Robert Shaw
parent f7e62e3d66
commit e1e318af01
20 changed files with 76 additions and 389 deletions
@@ -46,6 +46,7 @@ from vllm.utils.import_utils import (
has_deep_gemm,
has_mori,
)
from vllm.utils.math_utils import next_power_of_2
from .mk_objects import (
TestMoEQuantConfig,
@@ -604,13 +605,6 @@ def make_modular_kernel(
vllm_config: VllmConfig,
quant_config: FusedMoEQuantConfig,
) -> mk.FusedMoEKernel:
def next_power_of_2(x):
import math
if x == 0:
return 1
return 2 ** math.ceil(math.log2(x))
# make moe config
moe_parallel_config: FusedMoEParallelConfig = FusedMoEParallelConfig.make(
tp_size_=get_tensor_model_parallel_world_size(),
@@ -126,7 +126,7 @@ def parallel_launch_with_config(
world_size: int,
worker: Callable[Concatenate[ProcessGroupInfo, VllmConfig, Any, P], None],
vllm_config: VllmConfig,
env_dict: dict[Any, Any],
env_dict: dict[Any, Any] | None,
*args: P.args,
**kwargs: P.kwargs,
) -> None: