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

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

test/manual/test_triton_moe_wna16.py

@@ -4,7 +4,7 @@ import pytest
 import torch
 
 from sglang.srt.layers.activation import SiluAndMul
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_moe
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
 from sglang.srt.utils import get_device

test/registered/lora/test_lora_moe_runner.py

@@ -416,7 +416,7 @@ def test_lora_moe_runner_multi_expert(
         enable_deterministic_inference = False
 
     with patch(
-        "sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_config.get_global_server_args",
+        "sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_config.get_global_server_args",
         return_value=MockServerArgs(),
     ):
         runner = MoeRunner(MoeRunnerBackend.TRITON, config, lora_enabled=True)
@@ -566,7 +566,7 @@ def test_lora_moe_runner_virtual_experts(
         enable_deterministic_inference = False
 
     with patch(
-        "sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_config.get_global_server_args",
+        "sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_config.get_global_server_args",
         return_value=MockServerArgs(),
     ):
         runner = MoeRunner(MoeRunnerBackend.TRITON, config, lora_enabled=True)
@@ -758,7 +758,7 @@ def test_lora_moe_runner_marlin(num_tokens, top_k_num, num_experts, max_lora_ran
         enable_deterministic_inference = False
 
     with patch(
-        "sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_config.get_global_server_args",
+        "sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_config.get_global_server_args",
         return_value=MockServerArgs(),
     ):
         runner = MoeRunner(MoeRunnerBackend.MARLIN, config, lora_enabled=True)

test/registered/lora/test_marlin_lora_correctness.py

@@ -256,7 +256,7 @@ def test_marlin_vs_triton_lora_correctness(num_tokens, top_k):
         enable_deterministic_inference = False
 
     with patch(
-        "sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_config.get_global_server_args",
+        "sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_config.get_global_server_args",
         return_value=MockServerArgs(),
     ):
         marlin_runner = MoeRunner(MoeRunnerBackend.MARLIN, config, lora_enabled=True)

test/registered/moe/test_fused_moe.py

@@ -4,7 +4,7 @@ import torch
 from tqdm import tqdm
 
 from sglang.srt.layers.activation import SiluAndMul
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_moe
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.layers.quantization.fp8_kernel import is_fp8_fnuz
 from sglang.srt.layers.quantization.fp8_utils import normalize_e4m3fn_to_e4m3fnuz

test/registered/moe/test_triton_moe_channel_fp8_kernel.py

@@ -4,7 +4,7 @@ import unittest
 import torch
 
 from sglang.srt.layers.activation import SiluAndMul
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_moe
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.layers.quantization.fp8_kernel import scaled_fp8_quant
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler

test/registered/quant/test_block_int8.py

@@ -4,7 +4,7 @@ import unittest
 import torch
 
 from sglang.srt.layers.activation import SiluAndMul
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_moe
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

test/registered/quant/test_int8_kernel.py

@@ -4,7 +4,7 @@ import unittest
 import torch
 
 from sglang.srt.layers.activation import SiluAndMul
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_moe
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.layers.quantization.int8_kernel import per_token_quant_int8
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler