refactor(moe): de-duplicate triton MoE runner path into shared helpers (#23019)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

benchmark/kernels/fused_moe_triton/benchmark_sglang_fused_moe_triton.py

@@ -12,13 +12,13 @@ from sglang.srt.distributed.parallel_state import (
     init_distributed_environment,
     initialize_model_parallel,
 )
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
-    fused_moe as fused_moe_sglang,
-)
 from sglang.srt.layers.moe.fused_moe_triton.triton_kernels_moe import (
     triton_kernel_moe_forward,
 )
 from sglang.srt.layers.moe.moe_runner import MoeRunnerConfig
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import (
+    fused_moe as fused_moe_sglang,
+)
 from sglang.srt.layers.moe.topk import (
     TopK,
     TopKConfig,

benchmark/kernels/fused_moe_triton/benchmark_torch_compile_fused_moe.py

@@ -7,7 +7,7 @@ from torch.nn import functional as F
 from transformers import AutoConfig
 
 from sglang.benchmark.bench_utils import run_bench
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import (
     fused_moe as fused_moe_triton,
 )
 from sglang.srt.model_executor.cuda_graph_runner import set_torch_compile_config

benchmark/kernels/fused_moe_triton/benchmark_vllm_vs_sglang_fused_moe_triton.py

@@ -12,7 +12,7 @@ from sglang.srt.distributed.parallel_state import (
     init_distributed_environment,
     initialize_model_parallel,
 )
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import (
     fused_moe as fused_moe_sglang,
 )
 

benchmark/kernels/fused_moe_triton/common_utils.py

@@ -3,8 +3,8 @@ from typing import Dict, List, TypedDict
 
 import torch
 
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import get_config_dtype_str
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_config import (
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import get_config_dtype_str
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_config import (
     get_config_file_name,
 )
 from sglang.srt.utils import is_hip

benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton.py

@@ -20,13 +20,13 @@ from common_utils import (
 from ray.experimental.tqdm_ray import tqdm
 
 from sglang.srt.layers.moe.fused_moe_triton import override_config
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_config import (
+from sglang.srt.layers.moe.moe_runner import MoeRunnerConfig
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_moe
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_config import (
     get_config_dtype_str,
     get_default_config,
     get_moe_configs,
 )
-from sglang.srt.layers.moe.moe_runner import MoeRunnerConfig
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.server_args import (
     ServerArgs,

benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton_sep.py

@@ -22,15 +22,15 @@ from common_utils import (
 )
 from ray.experimental.tqdm_ray import tqdm
 
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
+from sglang.srt.layers.moe.moe_runner import MoeRunnerConfig
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import (
     get_config_dtype_str,
     invoke_fused_moe_kernel,
     moe_align_block_size,
 )
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_config import (
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_config import (
     get_config_file_name,
 )
-from sglang.srt.layers.moe.moe_runner import MoeRunnerConfig
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.server_args import (
     ServerArgs,