[MoE] Move GPT OSS Triton kernel experts into fused_moe/experts/ (#39007)

Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Signed-off-by: Jackmin801 <ongjackm@gmail.com>
Co-authored-by: Robert Shaw <robertgshaw2@gmail.com>
This commit is contained in:
Jackmin801
2026-04-14 19:27:39 +00:00
committed by GitHub
co-authored by Robert Shaw
parent ecf5ff7ce3
commit 1a9353bb02
8 changed files with 16 additions and 12 deletions
@@ -25,7 +25,7 @@ from triton_kernels.tensor_details import layout
from triton_kernels.testing import assert_close
from vllm.model_executor.layers.fused_moe.config import mxfp4_w4a16_moe_quant_config
from vllm.model_executor.layers.fused_moe.gpt_oss_triton_kernels_moe import (
from vllm.model_executor.layers.fused_moe.experts.gpt_oss_triton_kernels_moe import (
triton_kernel_moe_forward,
)
from vllm.utils.math_utils import round_up
@@ -29,7 +29,7 @@ from vllm.model_executor.layers.fused_moe.all2all_utils import (
maybe_make_prepare_finalize,
)
from vllm.model_executor.layers.fused_moe.config import mxfp4_w4a16_moe_quant_config
from vllm.model_executor.layers.fused_moe.gpt_oss_triton_kernels_moe import (
from vllm.model_executor.layers.fused_moe.experts.gpt_oss_triton_kernels_moe import (
OAITritonExperts,
UnfusedOAITritonExperts,
)
@@ -29,18 +29,22 @@ class TestTritonMoeForwardExpertMap:
torch.tensor([0, -1, 1, -1], device=device) if expert_map_present else None
)
from vllm.utils.import_utils import import_triton_kernels
import_triton_kernels()
with (
patch("triton_kernels.topk.topk") as mock_topk,
patch(
"vllm.model_executor.layers.fused_moe."
"vllm.model_executor.layers.fused_moe.experts."
"gpt_oss_triton_kernels_moe.make_routing_data"
) as mock_make_routing,
patch(
"vllm.model_executor.layers.fused_moe."
"vllm.model_executor.layers.fused_moe.experts."
"gpt_oss_triton_kernels_moe.triton_kernel_fused_experts"
) as mock_fused_experts,
):
from vllm.model_executor.layers.fused_moe.gpt_oss_triton_kernels_moe import ( # noqa: E501
from vllm.model_executor.layers.fused_moe.experts.gpt_oss_triton_kernels_moe import ( # noqa: E501
triton_kernel_moe_forward,
)