[MoE] Move various experts classes to fused_moe/experts/ (#41979)

Signed-off-by: Jackmin801 <ongjackm@gmail.com>
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Signed-off-by: Jackmin801 <56836461+Jackmin801@users.noreply.github.com>
Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Jackmin801 <ongjackm@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Robert Shaw <robertgshaw2@gmail.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Co-authored-by: Jackmin801 <56836461+Jackmin801@users.noreply.github.com>
This commit is contained in:
bnellnm
2026-05-11 07:54:33 +08:00
committed by GitHub
co-authored by Jackmin801 Claude Robert Shaw Robert Shaw Jackmin801
parent 21943d4c25
commit 1b57eb41f2
34 changed files with 586 additions and 553 deletions
@@ -20,7 +20,7 @@ from vllm.model_executor.layers.fused_moe.config import (
FusedMoEConfig,
FusedMoEQuantConfig,
)
from vllm.model_executor.layers.fused_moe.fused_marlin_moe import fused_marlin_moe
from vllm.model_executor.layers.fused_moe.experts.marlin_moe import fused_marlin_moe
from vllm.model_executor.layers.fused_moe.layer import (
FusedMoE,
FusedMoEMethodBase,
@@ -764,7 +764,7 @@ class AWQMarlinMoEMethod(FusedMoEMethodBase):
)
from vllm.model_executor.layers.fused_moe import modular_kernel as mk
from vllm.model_executor.layers.fused_moe.fused_marlin_moe import (
from vllm.model_executor.layers.fused_moe.experts.marlin_moe import (
BatchedMarlinExperts,
MarlinExperts,
)
@@ -17,7 +17,7 @@ from vllm.model_executor.layers.fused_moe.config import (
from vllm.model_executor.layers.fused_moe.experts.cutlass_moe import (
CutlassExpertsMxfp4,
)
from vllm.model_executor.layers.fused_moe.fused_marlin_moe import (
from vllm.model_executor.layers.fused_moe.experts.marlin_moe import (
MarlinExperts,
)
from vllm.model_executor.layers.fused_moe.oracle.mxfp4 import (
@@ -20,7 +20,7 @@ from vllm.model_executor.layers.fused_moe.config import (
FusedMoEQuantConfig,
int4_w4a16_moe_quant_config,
)
from vllm.model_executor.layers.fused_moe.fused_marlin_moe import (
from vllm.model_executor.layers.fused_moe.experts.marlin_moe import (
BatchedMarlinExperts,
MarlinExperts,
fused_marlin_moe,
@@ -26,7 +26,7 @@ from vllm.model_executor.layers.fused_moe.config import (
mxfp4_w4a16_moe_quant_config,
ocp_mx_moe_quant_config,
)
from vllm.model_executor.layers.fused_moe.fused_marlin_moe import fused_marlin_moe
from vllm.model_executor.layers.fused_moe.experts.marlin_moe import fused_marlin_moe
from vllm.model_executor.layers.fused_moe.oracle.mxfp4 import (
TRITON_BACKENDS,
Mxfp4MoeBackend,
@@ -444,7 +444,7 @@ class QuarkW8A8Fp8MoEMethod(QuarkMoEMethod):
shared_experts_input: torch.Tensor | None,
) -> torch.Tensor:
if self.rocm_aiter_moe_enabled:
from vllm.model_executor.layers.fused_moe.rocm_aiter_fused_moe import (
from vllm.model_executor.layers.fused_moe.experts.rocm_aiter_moe import (
rocm_aiter_fused_experts,
)
@@ -909,7 +909,7 @@ class QuarkW4A8Fp8MoEMethod(QuarkMoEMethod):
topk_ids: torch.Tensor,
shared_experts_input: torch.Tensor | None,
) -> torch.Tensor:
from vllm.model_executor.layers.fused_moe.rocm_aiter_fused_moe import (
from vllm.model_executor.layers.fused_moe.experts.rocm_aiter_moe import (
rocm_aiter_fused_experts,
)
@@ -1436,7 +1436,7 @@ class QuarkOCP_MX_MoEMethod(QuarkMoEMethod):
# AITER path
# TODO: Refactor this to use modular MOE kernel as well.
from vllm.model_executor.layers.fused_moe.rocm_aiter_fused_moe import (
from vllm.model_executor.layers.fused_moe.experts.rocm_aiter_moe import (
rocm_aiter_fused_experts,
)