diff --git a/3rdparty/amd/tuning/benchmark_moe_rocm.py b/3rdparty/amd/tuning/benchmark_moe_rocm.py
index 131b25270..d7ea67c5f 100644
--- a/3rdparty/amd/tuning/benchmark_moe_rocm.py
+++ b/3rdparty/amd/tuning/benchmark_moe_rocm.py
@@ -10,7 +10,7 @@ import triton.language as tl
 from tqdm import tqdm
 from transformers import AutoConfig
 
-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,
     get_config_file_name,
 )
diff --git a/benchmark/kernels/fused_moe_triton/benchmark_sglang_fused_moe_triton.py b/benchmark/kernels/fused_moe_triton/benchmark_sglang_fused_moe_triton.py
index eaec7712a..4515ff53b 100644
--- a/benchmark/kernels/fused_moe_triton/benchmark_sglang_fused_moe_triton.py
+++ b/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,
diff --git a/benchmark/kernels/fused_moe_triton/benchmark_torch_compile_fused_moe.py b/benchmark/kernels/fused_moe_triton/benchmark_torch_compile_fused_moe.py
index e94c66f4d..e6fdfa8a7 100644
--- a/benchmark/kernels/fused_moe_triton/benchmark_torch_compile_fused_moe.py
+++ b/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
diff --git a/benchmark/kernels/fused_moe_triton/benchmark_vllm_vs_sglang_fused_moe_triton.py b/benchmark/kernels/fused_moe_triton/benchmark_vllm_vs_sglang_fused_moe_triton.py
index 24fb580dc..fc100ce50 100644
--- a/benchmark/kernels/fused_moe_triton/benchmark_vllm_vs_sglang_fused_moe_triton.py
+++ b/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,
 )
 
diff --git a/benchmark/kernels/fused_moe_triton/common_utils.py b/benchmark/kernels/fused_moe_triton/common_utils.py
index d08d2bb75..4f4d09b68 100644
--- a/benchmark/kernels/fused_moe_triton/common_utils.py
+++ b/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
diff --git a/benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton.py b/benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton.py
index 4cc397f65..f134c56ef 100644
--- a/benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton.py
+++ b/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,
diff --git a/benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton_sep.py b/benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton_sep.py
index 8d4afbe84..cf9be7eb4 100644
--- a/benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton_sep.py
+++ b/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,
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/__init__.py b/python/sglang/srt/layers/moe/fused_moe_triton/__init__.py
index be3ed3af4..f398626da 100644
--- a/python/sglang/srt/layers/moe/fused_moe_triton/__init__.py
+++ b/python/sglang/srt/layers/moe/fused_moe_triton/__init__.py
@@ -1,35 +1,16 @@
-from contextlib import contextmanager
-from typing import Any, Dict, Optional
-
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_experts
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_config import (
-    get_config_file_name,
-    try_get_optimal_moe_config,
-)
 from sglang.srt.layers.moe.fused_moe_triton.layer import (
     FusedMoE,
     FusedMoeWeightScaleSupported,
 )
-from sglang.srt.layers.moe.fused_moe_triton.moe_align_block_size import (
+from sglang.srt.layers.moe.moe_runner.triton_utils import (
+    fused_experts,
+    get_config,
+    get_config_file_name,
     moe_align_block_size,
+    override_config,
+    try_get_optimal_moe_config,
 )
 
-_config: Optional[Dict[str, Any]] = None
-
-
-@contextmanager
-def override_config(config):
-    global _config
-    old_config = _config
-    _config = config
-    yield
-    _config = old_config
-
-
-def get_config() -> Optional[Dict[str, Any]]:
-    return _config
-
-
 __all__ = [
     "FusedMoE",
     "FusedMoeWeightScaleSupported",
diff --git a/python/sglang/srt/layers/moe/moe_runner/triton.py b/python/sglang/srt/layers/moe/moe_runner/triton.py
index fcae2cadf..840d70b0b 100644
--- a/python/sglang/srt/layers/moe/moe_runner/triton.py
+++ b/python/sglang/srt/layers/moe/moe_runner/triton.py
@@ -1,12 +1,9 @@
 from __future__ import annotations
 
-import functools
-import os
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any, List, Optional
 
 import torch
-import triton.language as tl
 
 from sglang.srt.layers.moe.moe_runner.base import (
     MoeQuantInfo,
@@ -19,7 +16,6 @@ from sglang.srt.layers.moe.moe_runner.base import (
     register_pre_permute,
 )
 from sglang.srt.layers.moe.utils import MoeRunnerBackend
-from sglang.srt.utils import cpu_has_amx_support, is_cpu, is_cuda, is_hip, is_xpu
 
 if TYPE_CHECKING:
     from sglang.srt.layers.moe.token_dispatcher.standard import (
@@ -28,47 +24,6 @@ if TYPE_CHECKING:
     )
 
 
-_is_hip = is_hip()
-_is_cuda = is_cuda()
-_is_cpu_amx_available = cpu_has_amx_support()
-_is_cpu = is_cpu()
-_use_aiter = bool(int(os.getenv("SGLANG_USE_AITER", "0")))
-_is_xpu = is_xpu()
-_MOE_PADDING_SIZE = 128 if bool(int(os.getenv("SGLANG_MOE_PADDING", "0"))) else 0
-
-
-if _is_cuda or _is_hip:
-    from sgl_kernel import gelu_and_mul, silu_and_mul
-
-    if _is_hip:
-        _has_vllm = False
-        if _use_aiter:
-            try:
-                from aiter import moe_sum
-            except ImportError:
-                raise ImportError(
-                    "aiter is required when SGLANG_USE_AITER is set to True"
-                )
-        else:
-            try:
-                from vllm import _custom_ops as vllm_ops  # moe_sum
-
-                _has_vllm = True
-            except ImportError:
-                # Fallback: vllm not available, will use triton moe_sum
-                _has_vllm = False
-elif _is_cpu and _is_cpu_amx_available:
-    pass
-elif _is_xpu:
-    from sgl_kernel import moe_sum_reduce, silu_and_mul
-
-
-if _is_cuda or _is_hip or _is_xpu:
-    from sgl_kernel import (  # noqa: F401
-        moe_align_block_size as sgl_moe_align_block_size,
-    )
-
-
 @dataclass
 class TritonRunnerInput(RunnerInput):
 
@@ -126,246 +81,54 @@ class TritonRunnerCore(MoeRunnerCore):
         running_state: dict,
         hooks: Optional[Any] = None,
     ) -> TritonRunnerOutput:
-
-        # TODO: move these functions to the triton runner
-        from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
-            _swiglu_gpt_oss_sigmoid_alpha,
-            _swiglu_silu_clamp_mul,
-            invoke_fused_moe_kernel,
-            moe_sum_reduce_torch_compile,
-            moe_sum_reduce_triton,
+        from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import (
+            _fused_moe_kernel_sequence,
         )
 
-        hidden_states = runner_input.hidden_states
-        topk_weights = runner_input.topk_weights
-        topk_ids = runner_input.topk_ids
-        sorted_token_ids = runner_input.sorted_token_ids
-        expert_ids = runner_input.expert_ids
-        num_tokens_post_padded = runner_input.num_tokens_post_padded
-
-        w13 = quant_info.w13_weight
-        w2 = quant_info.w2_weight
-        b13 = quant_info.b13
-        b2 = quant_info.b2
-        a13_scale = quant_info.a13_scale
-        a2_scale = quant_info.a2_scale
-        w13_scale = quant_info.w13_scale
-        w2_scale = quant_info.w2_scale
-        w13_zp = quant_info.w13_zp
-        w2_zp = quant_info.w2_zp
-        block_shape = quant_info.block_shape
-        per_channel_quant = quant_info.per_channel_quant
-        use_fp8_w8a8 = quant_info.use_fp8_w8a8
-        use_int8_w8a8 = quant_info.use_int8_w8a8
-        use_int8_w8a16 = quant_info.use_int8_w8a16
-        use_int4_w4a16 = quant_info.use_int4_w4a16
-
-        activation = self.config.activation
-        no_combine = self.config.no_combine
-        inplace = self.config.inplace
-        gemm1_alpha = self.config.gemm1_alpha
-        gemm1_limit = self.config.gemm1_clamp_limit
-        routed_scaling_factor = self.config.routed_scaling_factor
-        apply_router_weight_on_input = self.config.apply_router_weight_on_input
-
-        assert self.config.is_gated, "Only gated MoEs are supported for Triton runner"
-
-        M = hidden_states.shape[0]
-        E, N, _ = w13.shape
-        compute_type = (
-            tl.bfloat16 if hidden_states.dtype == torch.bfloat16 else tl.float16
+        filter_expert = (
+            self.config.num_experts is None
+            or self.config.num_experts != self.config.num_local_experts
         )
 
-        intermediate_cache1 = torch.empty(
-            (M, topk_ids.shape[1], N),
-            device=hidden_states.device,
-            dtype=hidden_states.dtype,
-        )
-
-        invoke_fused_moe_kernel(
-            hidden_states,
-            w13,
-            b13,
-            intermediate_cache1,
-            a13_scale,
-            w13_scale,
-            w13_zp,
-            topk_weights,
-            topk_ids,
-            sorted_token_ids,
-            expert_ids,
-            num_tokens_post_padded,
-            apply_router_weight_on_input,
-            topk_ids.shape[1],
+        out = _fused_moe_kernel_sequence(
+            runner_input.hidden_states,
+            quant_info.w13_weight,
+            quant_info.w2_weight,
+            runner_input.topk_weights,
+            runner_input.topk_ids,
+            runner_input.sorted_token_ids,
+            runner_input.expert_ids,
+            runner_input.num_tokens_post_padded,
             running_state["config"],
-            compute_type=compute_type,
-            use_fp8_w8a8=use_fp8_w8a8,
-            use_int8_w8a8=use_int8_w8a8,
-            use_int8_w8a16=use_int8_w8a16,
-            use_int4_w4a16=use_int4_w4a16,
-            per_channel_quant=per_channel_quant,
-            block_shape=block_shape,
+            running_state.get("down_config"),
+            running_state.get("down_moe_use_tma", False),
+            b1=quant_info.b13,
+            b2=quant_info.b2,
+            use_fp8_w8a8=quant_info.use_fp8_w8a8,
+            use_int8_w8a8=quant_info.use_int8_w8a8,
+            use_int8_w8a16=quant_info.use_int8_w8a16,
+            use_int4_w4a16=quant_info.use_int4_w4a16,
+            per_channel_quant=quant_info.per_channel_quant,
+            w1_scale=quant_info.w13_scale,
+            w2_scale=quant_info.w2_scale,
+            w1_zp=quant_info.w13_zp,
+            w2_zp=quant_info.w2_zp,
+            a1_scale=quant_info.a13_scale,
+            a2_scale=quant_info.a2_scale,
+            block_shape=quant_info.block_shape,
+            activation=self.config.activation,
+            is_gated=self.config.is_gated,
+            no_combine=self.config.no_combine,
+            inplace=self.config.inplace,
+            apply_router_weight_on_input=self.config.apply_router_weight_on_input,
+            routed_scaling_factor=self.config.routed_scaling_factor,
+            gemm1_alpha=self.config.gemm1_alpha,
+            gemm1_limit=self.config.gemm1_clamp_limit,
+            filter_expert=filter_expert,
+            hooks=hooks,
         )
 
-        if hooks and hooks.after_gate_up:
-            hooks.after_gate_up(
-                hidden_states, intermediate_cache1, topk_weights, topk_ids
-            )
-
-        intermediate_cache2 = torch.empty(
-            (M * topk_ids.shape[1], N // 2),
-            device=hidden_states.device,
-            dtype=hidden_states.dtype,
-        )
-
-        if activation == "silu":
-            if gemm1_alpha is not None:
-                assert gemm1_limit is not None
-                intermediate_cache2 = _swiglu_gpt_oss_sigmoid_alpha(
-                    intermediate_cache1.view(-1, N), gemm1_alpha, gemm1_limit
-                )
-            elif gemm1_limit is not None:
-                intermediate_cache2 = _swiglu_silu_clamp_mul(
-                    intermediate_cache1.view(-1, N), gemm1_limit
-                )
-            elif _is_cuda or _is_hip or _is_xpu:
-                silu_and_mul(intermediate_cache1.view(-1, N), intermediate_cache2)
-            else:
-                vllm_ops.silu_and_mul(
-                    intermediate_cache2, intermediate_cache1.view(-1, N)
-                )
-        elif activation == "gelu":
-            assert gemm1_alpha is None, "gemm1_alpha is not supported for gelu"
-            assert gemm1_limit is None, "gemm1_limit is not supported for gelu"
-            if _is_cuda or _is_hip:
-                gelu_and_mul(intermediate_cache1.view(-1, N), intermediate_cache2)
-            else:
-                vllm_ops.gelu_and_mul(
-                    intermediate_cache2, intermediate_cache1.view(-1, N)
-                )
-        else:
-            raise ValueError(f"Unsupported activation: {activation=}")
-
-        intermediate_cache3 = torch.empty(
-            (M, topk_ids.shape[1], w2.shape[1]),
-            device=hidden_states.device,
-            dtype=hidden_states.dtype,
-        )
-
-        if no_combine:
-            assert not inplace
-            out_hidden_states = torch.empty(
-                (M, topk_ids.shape[1], w2.shape[1]),
-                device=hidden_states.device,
-                dtype=hidden_states.dtype,
-            )
-        elif inplace:
-            out_hidden_states = hidden_states
-        else:
-            out_hidden_states = torch.empty_like(hidden_states)
-
-        # When LoRA hooks are present, always write to intermediate_cache3
-        # so the hook can modify it before reduction.
-        _use_intermediate = not no_combine and (topk_ids.shape[1] != 1 or hooks)
-        invoke_fused_moe_kernel(
-            intermediate_cache2,
-            w2,
-            b2,
-            (
-                intermediate_cache3
-                if _use_intermediate
-                else out_hidden_states.unsqueeze(0)
-            ),
-            a2_scale,
-            w2_scale,
-            w2_zp,
-            topk_weights,
-            topk_ids,
-            sorted_token_ids,
-            expert_ids,
-            num_tokens_post_padded,
-            not apply_router_weight_on_input,
-            1,
-            running_state["config"],
-            compute_type=compute_type,
-            use_fp8_w8a8=use_fp8_w8a8,
-            use_int8_w8a8=use_int8_w8a8,
-            use_int8_w8a16=use_int8_w8a16,
-            use_int4_w4a16=use_int4_w4a16,
-            per_channel_quant=per_channel_quant,
-            block_shape=block_shape,
-        )
-
-        if hooks and hooks.after_down:
-            hooks.after_down(
-                intermediate_cache2, intermediate_cache3, topk_weights, topk_ids
-            )
-
-        if routed_scaling_factor is None:
-            routed_scaling_factor = 1.0
-
-        if no_combine:
-            pass
-        elif _is_cuda:
-            if (
-                topk_ids.shape[1] == 1
-                and routed_scaling_factor == 1.0
-                and not _use_intermediate
-            ):
-                pass  # we wrote directly into out_hidden_states
-            elif topk_ids.shape[1] == 2 and routed_scaling_factor == 1.0:
-                torch.add(
-                    intermediate_cache3[:, 0],
-                    intermediate_cache3[:, 1],
-                    out=out_hidden_states,
-                ).squeeze(dim=1)
-            else:
-                # According to micro benchmark results, torch.compile can get better performance for small token.
-                if M <= 32:
-                    moe_sum_reduce_torch_compile(
-                        intermediate_cache3.view(*intermediate_cache3.shape),
-                        out_hidden_states,
-                        routed_scaling_factor,
-                    )
-                else:
-                    moe_sum_reduce_triton(
-                        intermediate_cache3.view(*intermediate_cache3.shape),
-                        out_hidden_states,
-                        routed_scaling_factor,
-                    )
-        elif _is_hip:
-            if _use_aiter:
-                moe_sum(
-                    intermediate_cache3.view(*intermediate_cache3.shape),
-                    out_hidden_states,
-                )
-            elif _has_vllm:
-                vllm_ops.moe_sum(
-                    intermediate_cache3.view(*intermediate_cache3.shape),
-                    out_hidden_states,
-                )
-            else:
-                # Fallback: use triton moe_sum when vllm is not available
-                moe_sum_reduce_triton(
-                    intermediate_cache3.view(*intermediate_cache3.shape),
-                    out_hidden_states,
-                    routed_scaling_factor,
-                )
-        elif _is_xpu:
-            moe_sum_reduce(
-                intermediate_cache3.view(*intermediate_cache3.shape),
-                out_hidden_states,
-                routed_scaling_factor,
-            )
-        else:
-            vllm_ops.moe_sum(
-                intermediate_cache3.view(*intermediate_cache3.shape),
-                out_hidden_states,
-            )
-
-        return TritonRunnerOutput(
-            hidden_states=out_hidden_states,
-        )
+        return TritonRunnerOutput(hidden_states=out)
 
     @property
     def runner_backend(self) -> MoeRunnerBackend:
@@ -378,7 +141,7 @@ def fused_experts_none_to_triton(
     quant_info: TritonMoeQuantInfo,
     runner_config: MoeRunnerConfig,
 ) -> StandardCombineInput:
-    from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_experts
+    from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_experts
     from sglang.srt.layers.moe.token_dispatcher.standard import StandardCombineInput
 
     output = fused_experts(
@@ -419,10 +182,8 @@ def pre_permute_standard_to_triton(
     # NOTE: this is dead code as a fused func for standard format is registered.
     # This is left here for testing and examples.
 
-    from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
-        get_config_dtype_str,
-        moe_align_block_size,
-        try_get_optimal_moe_config,
+    from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import (
+        _prepare_fused_moe_run,
     )
     from sglang.srt.layers.moe.topk import TopKOutputChecker
 
@@ -433,47 +194,29 @@ def pre_permute_standard_to_triton(
 
     assert TopKOutputChecker.format_is_standard(topk_output)
 
-    num_tokens = hidden_states.shape[0]
-    num_local_experts = runner_config.num_local_experts
-
-    if (
-        not (quant_info.use_fp8_w8a8 or quant_info.use_int8_w8a8)
-        or quant_info.block_shape is not None
-        or _use_aiter
-    ):
-        padding_size = 0
-    else:
-        padding_size = _MOE_PADDING_SIZE
-
-    config_dtype = get_config_dtype_str(
+    (
+        config,
+        down_config,
+        down_moe_use_tma,
+        sorted_token_ids,
+        expert_ids,
+        num_tokens_post_padded,
+    ) = _prepare_fused_moe_run(
+        hidden_states,
+        quant_info.w13_weight,
+        quant_info.w2_weight,
+        topk_output.topk_ids,
         use_fp8_w8a8=quant_info.use_fp8_w8a8,
         use_int8_w8a8=quant_info.use_int8_w8a8,
         use_int8_w8a16=quant_info.use_int8_w8a16,
         use_int4_w4a16=quant_info.use_int4_w4a16,
-        dtype=hidden_states.dtype,
-    )
-
-    get_config_func = functools.partial(
-        try_get_optimal_moe_config,
-        quant_info.w13_weight.shape,
-        (
-            num_local_experts,
-            quant_info.w2_weight.shape[1],
-            quant_info.w2_weight.shape[2] - padding_size,
-        ),
-        topk_output.topk_ids.shape[1],
-        config_dtype,
-        block_shape=quant_info.block_shape,
         per_channel_quant=quant_info.per_channel_quant,
-    )
-
-    config = get_config_func(num_tokens)
-
-    sorted_token_ids, expert_ids, num_tokens_post_padded = moe_align_block_size(
-        topk_output.topk_ids, config["BLOCK_SIZE_M"], num_local_experts
+        block_shape=quant_info.block_shape,
     )
 
     running_state["config"] = config
+    running_state["down_config"] = down_config
+    running_state["down_moe_use_tma"] = down_moe_use_tma
 
     return TritonRunnerInput(
         hidden_states=hidden_states,
diff --git a/python/sglang/srt/layers/moe/moe_runner/triton_utils/__init__.py b/python/sglang/srt/layers/moe/moe_runner/triton_utils/__init__.py
new file mode 100644
index 000000000..7a0f6ab1d
--- /dev/null
+++ b/python/sglang/srt/layers/moe/moe_runner/triton_utils/__init__.py
@@ -0,0 +1,36 @@
+from contextlib import contextmanager
+from typing import Any, Dict, Optional
+
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_experts
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_config import (
+    get_config_file_name,
+    try_get_optimal_moe_config,
+)
+from sglang.srt.layers.moe.moe_runner.triton_utils.moe_align_block_size import (
+    moe_align_block_size,
+)
+
+_config: Optional[Dict[str, Any]] = None
+
+
+@contextmanager
+def override_config(config):
+    global _config
+    old_config = _config
+    _config = config
+    yield
+    _config = old_config
+
+
+def get_config() -> Optional[Dict[str, Any]]:
+    return _config
+
+
+__all__ = [
+    "override_config",
+    "get_config",
+    "fused_experts",
+    "get_config_file_name",
+    "moe_align_block_size",
+    "try_get_optimal_moe_config",
+]
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/README.md b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/README.md
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/README.md
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/README.md
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=144,N=512,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=144,N=512,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=144,N=512,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=144,N=512,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3200,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3200,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3200,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3200,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=6400,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=6400,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=6400,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=6400,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=20,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=20,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=20,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=20,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=24,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=24,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=24,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=24,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_A800-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L20,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L20,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L20,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L20,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L40S,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L40S,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L40S,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L40S,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A800-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A800-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A800-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A800-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A800-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A800-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A800-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A800-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI300X.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI325X.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI325X.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI325X.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Radeon_Graphics.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Radeon_Graphics.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Radeon_Graphics.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Radeon_Graphics.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI300X.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI325X.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI325X.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI325X.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Radeon_Graphics.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Radeon_Graphics.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Radeon_Graphics.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Radeon_Graphics.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI300X.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI325X.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI325X.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI325X.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Radeon_Graphics.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Radeon_Graphics.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Radeon_Graphics.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Radeon_Graphics.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_L40S.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_L40S.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_L40S.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_L40S.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI300X.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI325X.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI325X.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI325X.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Radeon_Graphics.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Radeon_Graphics.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Radeon_Graphics.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Radeon_Graphics.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_A800-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_A800-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_A800-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_A800-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H20.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H20.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H20.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H20.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_A800-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_A800-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_A800-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_A800-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=96,device_name=NVIDIA_H20.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=96,device_name=NVIDIA_H20.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=96,device_name=NVIDIA_H20.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=128,N=96,device_name=NVIDIA_H20.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=129,N=352,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=129,N=352,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=129,N=352,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=129,N=352,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=160,N=320,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=160,N=320,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=160,N=320,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=160,N=320,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=161,N=192,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=161,N=192,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=161,N=192,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=161,N=192,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=264,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=264,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=64,device_name=NVIDIA_A800-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=272,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=272,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=288,N=64,device_name=NVIDIA_A800-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=288,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=288,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=288,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_2_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_0/E=16,N=1024,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_0/E=16,N=1024,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_0/E=16,N=1024,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_0/E=16,N=1024,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=352,device_name=NVIDIA_RTX_6000_Ada_Generation,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=352,device_name=NVIDIA_RTX_6000_Ada_Generation,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=352,device_name=NVIDIA_RTX_6000_Ada_Generation,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=352,device_name=NVIDIA_RTX_6000_Ada_Generation,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=160,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=160,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=320,device_name=NVIDIA_H20-3e.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=160,N=320,device_name=NVIDIA_H20-3e.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=320,device_name=NVIDIA_H20-3e.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=160,N=320,device_name=NVIDIA_H20-3e.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=160,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=160,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=160,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=160,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=384,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=384,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_3_1/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_L40S.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_L40S.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_L40S.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_L40S.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=352,device_name=NVIDIA_RTX_5880_Ada_Generation,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=352,device_name=NVIDIA_RTX_5880_Ada_Generation,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=352,device_name=NVIDIA_RTX_5880_Ada_Generation,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=352,device_name=NVIDIA_RTX_5880_Ada_Generation,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=928,device_name=NVIDIA_L40S.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=928,device_name=NVIDIA_L40S.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=928,device_name=NVIDIA_L40S.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=128,N=928,device_name=NVIDIA_L40S.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_B200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=129,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=129,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=160,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=160,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=160,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=160,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=161,N=384,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=161,N=384,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=161,N=384,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=161,N=384,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_H20.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_H20.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_H20.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_H20.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=64,device_name=NVIDIA_A100-SXM4-80GB.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=257,N=64,device_name=NVIDIA_A100-SXM4-80GB.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=64,device_name=NVIDIA_A100-SXM4-80GB.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=257,N=64,device_name=NVIDIA_A100-SXM4-80GB.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16_down.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16_down.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16_down.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16_down.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=384,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=384,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H20-3e.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H20-3e.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H20-3e.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H20-3e.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H20-3e.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H20-3e.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H20-3e.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H20-3e.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=352,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=352,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=352,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=352,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=352,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=352,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=352,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=352,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=704,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=704,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=704,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=704,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=704,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=704,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=704,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=704,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=2048,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=16,N=2048,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=2048,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=16,N=2048,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128]_down.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=40,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=40,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=40,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=40,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_B200.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_B200.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_B200.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json b/python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py b/python/sglang/srt/layers/moe/moe_runner/triton_utils/fused_moe.py
similarity index 56%
rename from python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/fused_moe.py
index 73bb93b55..cd9b3ad68 100644
--- a/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py
+++ b/python/sglang/srt/layers/moe/moe_runner/triton_utils/fused_moe.py
@@ -1,4 +1,3 @@
-# NOTE: this file will be separated into sglang/srt/layers/moe/moe_runner/triton_utils.py
 # Adapted from https://github.com/vllm-project/vllm/blob/a6221a144af772fd1a68fe7e627935dc53e81738/vllm/model_executor/layers/fused_moe/fused_moe.py
 
 """Fused MoE kernel."""
@@ -6,7 +5,7 @@
 from __future__ import annotations
 
 import functools
-from typing import TYPE_CHECKING, List, Optional
+from typing import TYPE_CHECKING, Any, Dict, List, Optional
 
 import torch
 import torch.nn.functional as F
@@ -320,6 +319,406 @@ def _down_moe_use_tma():
     return support_tensor_descriptor()
 
 
+def _prepare_fused_moe_run(
+    hidden_states: torch.Tensor,
+    w1: torch.Tensor,
+    w2: torch.Tensor,
+    topk_ids: torch.Tensor,
+    *,
+    use_fp8_w8a8: bool,
+    use_int8_w8a8: bool,
+    use_int8_w8a16: bool,
+    use_int4_w4a16: bool,
+    per_channel_quant: bool,
+    block_shape: Optional[List[int]],
+):
+    """Resolve config, down_config, TMA flag, and aligned expert routing ids.
+
+    Shared by ``fused_experts_impl`` and ``pre_permute_standard_to_triton`` so
+    both paths compute alignment from the same source.
+    """
+    padded_size = padding_size
+    if not (use_fp8_w8a8 or use_int8_w8a8) or block_shape is not None or _use_aiter:
+        padded_size = 0
+
+    num_tokens = hidden_states.shape[0]
+    E = w1.shape[0]
+    config_dtype = get_config_dtype_str(
+        use_fp8_w8a8=use_fp8_w8a8,
+        use_int8_w8a8=use_int8_w8a8,
+        use_int8_w8a16=use_int8_w8a16,
+        use_int4_w4a16=use_int4_w4a16,
+        dtype=hidden_states.dtype,
+    )
+
+    config, (down_config, _) = try_get_optimal_moe_config(
+        w1.shape,
+        (w2.shape[0], w2.shape[1], w2.shape[2] - padded_size),
+        topk_ids.shape[1],
+        config_dtype,
+        num_tokens,
+        block_shape=block_shape,
+        per_channel_quant=per_channel_quant,
+        return_down_config=True,
+    )
+    down_moe_use_tma = (
+        _down_moe_use_tma()
+        and down_config is not None
+        and down_config.pop("USE_TMA", False)
+    )
+
+    sorted_token_ids, expert_ids, num_tokens_post_padded = moe_align_block_size(
+        topk_ids, config["BLOCK_SIZE_M"], E
+    )
+
+    return (
+        config,
+        down_config,
+        down_moe_use_tma,
+        sorted_token_ids,
+        expert_ids,
+        num_tokens_post_padded,
+    )
+
+
+def _fused_moe_kernel_sequence(
+    hidden_states: torch.Tensor,
+    w1: torch.Tensor,
+    w2: torch.Tensor,
+    topk_weights: torch.Tensor,
+    topk_ids: torch.Tensor,
+    sorted_token_ids: torch.Tensor,
+    expert_ids: torch.Tensor,
+    num_tokens_post_padded: torch.Tensor,
+    config: Dict[str, Any],
+    down_config: Optional[Dict[str, Any]],
+    down_moe_use_tma: bool,
+    *,
+    b1: Optional[torch.Tensor],
+    b2: Optional[torch.Tensor],
+    use_fp8_w8a8: bool,
+    use_int8_w8a8: bool,
+    use_int8_w8a16: bool,
+    use_int4_w4a16: bool,
+    per_channel_quant: bool,
+    w1_scale: Optional[torch.Tensor],
+    w2_scale: Optional[torch.Tensor],
+    w1_zp: Optional[torch.Tensor],
+    w2_zp: Optional[torch.Tensor],
+    a1_scale: Optional[torch.Tensor],
+    a2_scale: Optional[torch.Tensor],
+    block_shape: Optional[List[int]],
+    activation: str,
+    is_gated: bool,
+    no_combine: bool,
+    inplace: bool,
+    apply_router_weight_on_input: bool,
+    routed_scaling_factor: Optional[float],
+    gemm1_alpha: Optional[float],
+    gemm1_limit: Optional[float],
+    filter_expert: bool,
+    hooks: Optional[Any] = None,
+) -> torch.Tensor:
+    """Run the MoE kernel/activation/kernel/combine sequence in a single shot.
+
+    Inputs are already aligned and the block-size config is already resolved.
+    Supports optional LoRA hooks that fire between the two kernels and before
+    combine. Returns ``out_hidden_states``.
+    """
+    num_tokens = hidden_states.shape[0]
+    E, N, _ = w1.shape
+    topk = topk_ids.shape[1]
+    compute_type = tl.bfloat16 if hidden_states.dtype == torch.bfloat16 else tl.float16
+
+    padded_tokens = (
+        min(num_tokens * topk, E + 1) * (config["BLOCK_SIZE_M"] - 1)
+        if down_moe_use_tma
+        else 0
+    )
+    total_tokens = num_tokens * topk + padded_tokens
+
+    if no_combine:
+        assert not inplace
+        out_hidden_states = torch.empty(
+            (num_tokens, topk, w2.shape[1]),
+            device=hidden_states.device,
+            dtype=hidden_states.dtype,
+        )
+    elif inplace:
+        out_hidden_states = hidden_states
+    else:
+        out_hidden_states = torch.empty_like(hidden_states)
+
+    use_fused_moe_sum_all_reduce = (
+        get_global_server_args().enable_fused_moe_sum_all_reduce
+        and (not no_combine)
+        and (topk > 2)
+        and (not use_int8_w8a16)
+        and (not use_int4_w4a16)
+    )
+
+    intermediate_cache1 = torch.empty(
+        (total_tokens, N),
+        device=hidden_states.device,
+        dtype=hidden_states.dtype,
+    )
+
+    invoke_fused_moe_kernel(
+        hidden_states,
+        w1,
+        b1,
+        intermediate_cache1,
+        a1_scale,
+        w1_scale,
+        w1_zp,
+        topk_weights,
+        topk_ids,
+        sorted_token_ids,
+        expert_ids,
+        num_tokens_post_padded,
+        apply_router_weight_on_input,
+        topk,
+        config,
+        compute_type=compute_type,
+        use_fp8_w8a8=use_fp8_w8a8,
+        use_int8_w8a8=use_int8_w8a8,
+        use_int8_w8a16=use_int8_w8a16,
+        use_int4_w4a16=use_int4_w4a16,
+        per_channel_quant=per_channel_quant,
+        block_shape=block_shape,
+        c_sorted=down_moe_use_tma,
+        filter_expert=filter_expert,
+    )
+
+    if hooks and hooks.after_gate_up:
+        # Hooks expect intermediate_cache1 shaped (num_tokens, topk, N); the
+        # underlying buffer is laid out as (total_tokens, N) where
+        # total_tokens = num_tokens * topk (+ TMA padding). Slice off any
+        # padding and reshape for the hook, which writes in-place on the view.
+        hooks.after_gate_up(
+            hidden_states,
+            intermediate_cache1[: num_tokens * topk].view(num_tokens, topk, N),
+            topk_weights,
+            topk_ids,
+        )
+
+    intermediate_cache2 = torch.empty(
+        (total_tokens, N // 2),
+        device=hidden_states.device,
+        dtype=hidden_states.dtype,
+    )
+
+    # Activation function with multiplication
+    if activation == "silu" and is_gated:
+        # - gemm1_alpha != None: GPT-OSS-style swiglu(alpha, limit)
+        # - gemm1_alpha == None and gemm1_limit != None: silu+clamp+mul(limit-only)
+        if gemm1_alpha is not None:
+            assert gemm1_limit is not None
+            intermediate_cache2 = _swiglu_gpt_oss_sigmoid_alpha(
+                intermediate_cache1.view(-1, N), gemm1_alpha, gemm1_limit
+            )
+        elif gemm1_limit is not None:
+            intermediate_cache2 = _swiglu_silu_clamp_mul(
+                intermediate_cache1.view(-1, N), gemm1_limit
+            )
+        elif _is_cuda or _is_hip or _is_xpu:
+            if not filter_expert:
+                silu_and_mul(intermediate_cache1.view(-1, N), intermediate_cache2)
+            else:
+                act_and_mul_triton(
+                    intermediate_cache1.view(-1, N),
+                    intermediate_cache2,
+                    config,
+                    topk_ids,
+                    expert_ids,
+                    down_moe_use_tma,
+                    activation,
+                )
+        else:
+            if _has_vllm_ops:
+                vllm_ops.silu_and_mul(
+                    intermediate_cache2, intermediate_cache1.view(-1, N)
+                )
+            else:
+                # Fallback: native PyTorch silu_and_mul
+                x = intermediate_cache1.view(-1, N)
+                d = x.shape[-1] // 2
+                intermediate_cache2.copy_(F.silu(x[..., :d]) * x[..., d:])
+    elif activation == "gelu" and is_gated:
+        assert gemm1_alpha is None, "gemm1_alpha is not supported for gelu"
+        assert gemm1_limit is None, "gemm1_limit is not supported for gelu"
+        if _is_cuda or _is_hip:
+            if not filter_expert:
+                gelu_and_mul(intermediate_cache1.view(-1, N), intermediate_cache2)
+            else:
+                act_and_mul_triton(
+                    intermediate_cache1.view(-1, N),
+                    intermediate_cache2,
+                    config,
+                    topk_ids,
+                    expert_ids,
+                    down_moe_use_tma,
+                    activation,
+                )
+        else:
+            if _has_vllm_ops:
+                vllm_ops.gelu_and_mul(
+                    intermediate_cache2, intermediate_cache1.view(-1, N)
+                )
+            else:
+                # Fallback: native PyTorch gelu_and_mul
+                x = intermediate_cache1.view(-1, N)
+                d = x.shape[-1] // 2
+                intermediate_cache2.copy_(F.gelu(x[..., :d]) * x[..., d:])
+    # Activation function without multiplication
+    elif activation == "silu" and not is_gated:
+        intermediate_cache2 = F.silu(intermediate_cache1.view(-1, N))
+    elif activation == "gelu" and not is_gated:
+        intermediate_cache2 = F.gelu(intermediate_cache1.view(-1, N))
+    elif activation == "relu2" and not is_gated:
+        intermediate_cache2 = torch.square(F.relu(intermediate_cache1.view(-1, N)))
+    else:
+        raise ValueError(f"Unsupported activation: {activation=}, with {is_gated=}")
+
+    del intermediate_cache1
+
+    intermediate_cache3 = torch.empty(
+        (num_tokens, topk, w2.shape[1]),
+        device=hidden_states.device,
+        dtype=hidden_states.dtype,
+    )
+
+    # LoRA hooks force the second kernel to write to intermediate_cache3 so
+    # hooks.after_down can inspect/modify it before reduction.
+    _use_intermediate = not no_combine and (topk != 1 or hooks)
+
+    out_slice = None
+    if use_fused_moe_sum_all_reduce:
+        out_slice = out_hidden_states
+        out_slice.zero_()
+
+    invoke_fused_moe_kernel(
+        intermediate_cache2,
+        w2,
+        b2,
+        (
+            out_slice
+            if use_fused_moe_sum_all_reduce
+            else (
+                intermediate_cache3
+                if _use_intermediate
+                else out_hidden_states.unsqueeze(0)
+            )
+        ),
+        a2_scale,
+        w2_scale,
+        w2_zp,
+        topk_weights,
+        topk_ids,
+        sorted_token_ids,
+        expert_ids,
+        num_tokens_post_padded,
+        not apply_router_weight_on_input,
+        1,
+        down_config or config,
+        compute_type=compute_type,
+        use_fp8_w8a8=use_fp8_w8a8,
+        use_int8_w8a8=use_int8_w8a8,
+        use_int8_w8a16=use_int8_w8a16,
+        use_int4_w4a16=use_int4_w4a16,
+        per_channel_quant=per_channel_quant,
+        block_shape=block_shape,
+        a_use_tma=down_moe_use_tma,
+        b_use_tma=down_moe_use_tma,
+        filter_expert=filter_expert,
+        fuse_sum_all_reduce=use_fused_moe_sum_all_reduce,
+        router_topk=topk,
+    )
+
+    if hooks and hooks.after_down:
+        hooks.after_down(
+            intermediate_cache2, intermediate_cache3, topk_weights, topk_ids
+        )
+
+    del intermediate_cache2
+
+    if routed_scaling_factor is None:
+        routed_scaling_factor = 1.0
+
+    if no_combine:
+        pass
+    elif _is_cuda:
+        if use_fused_moe_sum_all_reduce:
+            if routed_scaling_factor != 1.0:
+                assert out_slice is not None
+                out_slice.mul_(routed_scaling_factor)
+        elif topk == 1 and routed_scaling_factor == 1.0 and not _use_intermediate:
+            pass  # we wrote directly into out_hidden_states
+        elif topk == 2 and routed_scaling_factor == 1.0:
+            torch.add(
+                intermediate_cache3[:, 0],
+                intermediate_cache3[:, 1],
+                out=out_hidden_states,
+            ).squeeze(dim=1)
+        else:
+            # According to micro benchmark results, torch.compile can get better performance for small token.
+            if num_tokens <= 32:
+                moe_sum_reduce_torch_compile(
+                    intermediate_cache3.view(*intermediate_cache3.shape),
+                    out_hidden_states,
+                    routed_scaling_factor,
+                )
+            else:
+                moe_sum_reduce(
+                    intermediate_cache3.view(*intermediate_cache3.shape),
+                    out_hidden_states,
+                    routed_scaling_factor,
+                )
+    elif _is_hip:
+        if _use_aiter:
+            moe_sum(
+                intermediate_cache3.view(*intermediate_cache3.shape),
+                out_hidden_states,
+            )
+        else:
+            # According to micro benchmark results, torch.compile can get better performance for small token.
+            if num_tokens <= 32:
+                moe_sum_reduce_torch_compile(
+                    intermediate_cache3.view(*intermediate_cache3.shape),
+                    out_hidden_states,
+                    routed_scaling_factor,
+                )
+            else:
+                moe_sum_reduce_triton(
+                    intermediate_cache3.view(*intermediate_cache3.shape),
+                    out_hidden_states,
+                    routed_scaling_factor,
+                )
+    elif _is_xpu:
+        moe_sum_reduce(
+            intermediate_cache3.view(*intermediate_cache3.shape),
+            out_hidden_states,
+            routed_scaling_factor,
+        )
+    else:
+        if _has_vllm_ops:
+            vllm_ops.moe_sum(
+                intermediate_cache3.view(*intermediate_cache3.shape),
+                out_hidden_states,
+            )
+        else:
+            # Fallback: use triton moe_sum_reduce when vllm is not available
+            moe_sum_reduce_triton(
+                intermediate_cache3.view(*intermediate_cache3.shape),
+                out_hidden_states,
+                routed_scaling_factor,
+            )
+
+    del intermediate_cache3
+
+    return out_hidden_states
+
+
 def fused_experts_impl(
     hidden_states: torch.Tensor,
     w1: torch.Tensor,
@@ -367,337 +766,63 @@ def fused_experts_impl(
     assert w2.is_contiguous(), "Expert weights2 must be contiguous"
     assert hidden_states.dtype in [torch.float32, torch.float16, torch.bfloat16]
 
-    num_tokens, _ = hidden_states.shape
-    E, N, _ = w1.shape
-    # We execute the fused_moe kernel in chunks to circumvent this issue:
-    # https://github.com/vllm-project/vllm/issues/5938
-    CHUNK_SIZE = 64 * 1024
-    M = min(num_tokens, CHUNK_SIZE)
-    config_dtype = get_config_dtype_str(
+    (
+        config,
+        down_config,
+        down_moe_use_tma,
+        sorted_token_ids,
+        expert_ids,
+        num_tokens_post_padded,
+    ) = _prepare_fused_moe_run(
+        hidden_states,
+        w1,
+        w2,
+        topk_ids,
         use_fp8_w8a8=use_fp8_w8a8,
         use_int8_w8a8=use_int8_w8a8,
         use_int8_w8a16=use_int8_w8a16,
         use_int4_w4a16=use_int4_w4a16,
-        dtype=hidden_states.dtype,
-    )
-
-    get_config_func = functools.partial(
-        try_get_optimal_moe_config,
-        w1.shape,
-        (w2.shape[0], w2.shape[1], w2.shape[2] - padded_size),
-        topk_ids.shape[1],
-        config_dtype,
-        block_shape=block_shape,
         per_channel_quant=per_channel_quant,
-        return_down_config=True,
+        block_shape=block_shape,
     )
 
-    config, (down_config, max_block_m) = get_config_func(M)
-    down_moe_use_tma = (
-        _down_moe_use_tma()
-        and down_config is not None
-        and down_config.pop("USE_TMA", False)
+    return _fused_moe_kernel_sequence(
+        hidden_states,
+        w1,
+        w2,
+        topk_weights,
+        topk_ids,
+        sorted_token_ids,
+        expert_ids,
+        num_tokens_post_padded,
+        config,
+        down_config,
+        down_moe_use_tma,
+        b1=b1,
+        b2=b2,
+        use_fp8_w8a8=use_fp8_w8a8,
+        use_int8_w8a8=use_int8_w8a8,
+        use_int8_w8a16=use_int8_w8a16,
+        use_int4_w4a16=use_int4_w4a16,
+        per_channel_quant=per_channel_quant,
+        w1_scale=w1_scale,
+        w2_scale=w2_scale,
+        w1_zp=w1_zp,
+        w2_zp=w2_zp,
+        a1_scale=a1_scale,
+        a2_scale=a2_scale,
+        block_shape=block_shape,
+        activation=activation,
+        is_gated=is_gated,
+        no_combine=no_combine,
+        inplace=inplace,
+        apply_router_weight_on_input=apply_router_weight_on_input,
+        routed_scaling_factor=routed_scaling_factor,
+        gemm1_alpha=gemm1_alpha,
+        gemm1_limit=gemm1_limit,
+        filter_expert=filter_expert,
+        hooks=None,
     )
-    topk = topk_ids.shape[1]
-    max_padded_tokens = (
-        min(M * topk, E + 1) * (max_block_m - 1) if down_moe_use_tma else 0
-    )
-    total_tokens = M * topk + max_padded_tokens
-    cache = torch.empty(
-        total_tokens * max(N, w2.shape[1]),
-        device=hidden_states.device,
-        dtype=hidden_states.dtype,
-    )
-    intermediate_cache3 = cache[: M * topk * w2.shape[1]].view(
-        (M, topk, w2.shape[1]),
-    )
-
-    compute_type = tl.bfloat16 if hidden_states.dtype == torch.bfloat16 else tl.float16
-
-    if no_combine:
-        assert not inplace
-        out_hidden_states = torch.empty(
-            (num_tokens, topk, w2.shape[1]),
-            device=hidden_states.device,
-            dtype=hidden_states.dtype,
-        )
-    elif inplace:
-        out_hidden_states = hidden_states
-    else:
-        out_hidden_states = torch.empty_like(hidden_states)
-
-    for chunk in range((num_tokens // CHUNK_SIZE) + 1):
-        begin_chunk_idx, end_chunk_idx = (
-            chunk * CHUNK_SIZE,
-            min((chunk + 1) * CHUNK_SIZE, num_tokens),
-        )
-        curr_hidden_states = hidden_states[begin_chunk_idx:end_chunk_idx]
-        tokens_in_chunk, _ = curr_hidden_states.shape
-
-        if tokens_in_chunk == 0:
-            break
-
-        if tokens_in_chunk < CHUNK_SIZE and chunk > 0:
-            # Adjust the intermediate cache size and config for the last
-            # chunk. Note that in most cases we only have one chunk
-            # so the cache size and config are already set correctly and
-            # do not need to be adjusted.
-            config, (down_config, _) = get_config_func(tokens_in_chunk)
-            down_moe_use_tma = (
-                _down_moe_use_tma()
-                and down_config is not None
-                and down_config.pop("USE_TMA", False)
-            )
-            intermediate_cache3 = intermediate_cache3[:tokens_in_chunk]
-
-        padded_tokens = (
-            min(tokens_in_chunk * topk, E + 1) * (config["BLOCK_SIZE_M"] - 1)
-            if down_moe_use_tma
-            else 0
-        )
-        total_tokens = tokens_in_chunk * topk + padded_tokens
-        intermediate_cache1 = cache[: total_tokens * N].view(
-            (total_tokens, N),
-        )
-        intermediate_cache2 = torch.empty(
-            (total_tokens, N // 2),
-            device=hidden_states.device,
-            dtype=hidden_states.dtype,
-        )
-
-        curr_topk_ids = topk_ids[begin_chunk_idx:end_chunk_idx]
-        curr_topk_weights = topk_weights[begin_chunk_idx:end_chunk_idx]
-
-        use_fused_moe_sum_all_reduce = (
-            get_global_server_args().enable_fused_moe_sum_all_reduce
-            and (not no_combine)
-            and (curr_topk_ids.shape[1] > 2)
-            and (not use_int8_w8a16)
-            and (not use_int4_w4a16)
-        )
-
-        sorted_token_ids, expert_ids, num_tokens_post_padded = moe_align_block_size(
-            curr_topk_ids, config["BLOCK_SIZE_M"], E
-        )
-
-        invoke_fused_moe_kernel(
-            curr_hidden_states,
-            w1,
-            b1,
-            intermediate_cache1,
-            a1_scale,
-            w1_scale,
-            w1_zp,
-            curr_topk_weights,
-            curr_topk_ids,
-            sorted_token_ids,
-            expert_ids,
-            num_tokens_post_padded,
-            apply_router_weight_on_input,
-            topk_ids.shape[1],
-            config,
-            compute_type=compute_type,
-            use_fp8_w8a8=use_fp8_w8a8,
-            use_int8_w8a8=use_int8_w8a8,
-            use_int8_w8a16=use_int8_w8a16,
-            use_int4_w4a16=use_int4_w4a16,
-            per_channel_quant=per_channel_quant,
-            block_shape=block_shape,
-            c_sorted=down_moe_use_tma,
-            filter_expert=filter_expert,
-        )
-
-        # Activation function with multiplication
-        if activation == "silu" and is_gated:
-            # - gemm1_alpha != None: GPT-OSS-style swiglu(alpha, limit)
-            # - gemm1_alpha == None and gemm1_limit != None: silu+clamp+mul(limit-only)
-            if gemm1_alpha is not None:
-                assert gemm1_limit is not None
-                intermediate_cache2 = _swiglu_gpt_oss_sigmoid_alpha(
-                    intermediate_cache1.view(-1, N), gemm1_alpha, gemm1_limit
-                )
-            elif gemm1_limit is not None:
-                intermediate_cache2 = _swiglu_silu_clamp_mul(
-                    intermediate_cache1.view(-1, N), gemm1_limit
-                )
-            elif _is_cuda or _is_hip or _is_xpu:
-                if not filter_expert:
-                    silu_and_mul(intermediate_cache1.view(-1, N), intermediate_cache2)
-                else:
-                    act_and_mul_triton(
-                        intermediate_cache1.view(-1, N),
-                        intermediate_cache2,
-                        config,
-                        curr_topk_ids,
-                        expert_ids,
-                        down_moe_use_tma,
-                        activation,
-                    )
-            else:
-                if _has_vllm_ops:
-                    vllm_ops.silu_and_mul(
-                        intermediate_cache2, intermediate_cache1.view(-1, N)
-                    )
-                else:
-                    # Fallback: native PyTorch silu_and_mul
-                    x = intermediate_cache1.view(-1, N)
-                    d = x.shape[-1] // 2
-                    intermediate_cache2.copy_(F.silu(x[..., :d]) * x[..., d:])
-        elif activation == "gelu" and is_gated:
-            assert gemm1_alpha is None, "gemm1_alpha is not supported for gelu"
-            assert gemm1_limit is None, "gemm1_limit is not supported for gelu"
-            if _is_cuda or _is_hip:
-                if not filter_expert:
-                    gelu_and_mul(intermediate_cache1.view(-1, N), intermediate_cache2)
-                else:
-                    act_and_mul_triton(
-                        intermediate_cache1.view(-1, N),
-                        intermediate_cache2,
-                        config,
-                        curr_topk_ids,
-                        expert_ids,
-                        down_moe_use_tma,
-                        activation,
-                    )
-            else:
-                if _has_vllm_ops:
-                    vllm_ops.gelu_and_mul(
-                        intermediate_cache2, intermediate_cache1.view(-1, N)
-                    )
-                else:
-                    # Fallback: native PyTorch gelu_and_mul
-                    x = intermediate_cache1.view(-1, N)
-                    d = x.shape[-1] // 2
-                    intermediate_cache2.copy_(F.gelu(x[..., :d]) * x[..., d:])
-        # Activation function without multiplication
-        elif activation == "silu" and not is_gated:
-            intermediate_cache2 = F.silu(intermediate_cache1.view(-1, N))
-        elif activation == "gelu" and not is_gated:
-            intermediate_cache2 = F.gelu(intermediate_cache1.view(-1, N))
-        elif activation == "relu2" and not is_gated:
-            intermediate_cache2 = torch.square(F.relu(intermediate_cache1.view(-1, N)))
-        else:
-            raise ValueError(f"Unsupported activation: {activation=}, with {is_gated=}")
-
-        out_slice = None
-        if use_fused_moe_sum_all_reduce:
-            out_slice = out_hidden_states[begin_chunk_idx:end_chunk_idx]
-            out_slice.zero_()
-
-        invoke_fused_moe_kernel(
-            intermediate_cache2,
-            w2,
-            b2,
-            (
-                out_slice
-                if use_fused_moe_sum_all_reduce
-                else (
-                    intermediate_cache3
-                    if not no_combine and topk_ids.shape[1] != 1
-                    else out_hidden_states[begin_chunk_idx:end_chunk_idx].unsqueeze(0)
-                )
-            ),
-            a2_scale,
-            w2_scale,
-            w2_zp,
-            curr_topk_weights,
-            curr_topk_ids,
-            sorted_token_ids,
-            expert_ids,
-            num_tokens_post_padded,
-            not apply_router_weight_on_input,
-            1,
-            down_config or config,
-            compute_type=compute_type,
-            use_fp8_w8a8=use_fp8_w8a8,
-            use_int8_w8a8=use_int8_w8a8,
-            use_int8_w8a16=use_int8_w8a16,
-            use_int4_w4a16=use_int4_w4a16,
-            per_channel_quant=per_channel_quant,
-            block_shape=block_shape,
-            a_use_tma=down_moe_use_tma,
-            b_use_tma=down_moe_use_tma,
-            filter_expert=filter_expert,
-            fuse_sum_all_reduce=use_fused_moe_sum_all_reduce,
-            router_topk=curr_topk_ids.shape[1],
-        )
-
-        if routed_scaling_factor is None:
-            routed_scaling_factor = 1.0
-
-        if no_combine:
-            pass
-        elif _is_cuda:
-            if use_fused_moe_sum_all_reduce:
-                if routed_scaling_factor is None:
-                    routed_scaling_factor = 1.0
-                if routed_scaling_factor != 1.0:
-                    assert out_slice is not None
-                    out_slice.mul_(routed_scaling_factor)
-            elif topk_ids.shape[1] == 1 and routed_scaling_factor == 1.0:
-                pass  # we write directly into out_hidden_states
-            elif topk_ids.shape[1] == 2 and routed_scaling_factor == 1.0:
-                torch.add(
-                    intermediate_cache3[:, 0],
-                    intermediate_cache3[:, 1],
-                    out=out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                ).squeeze(dim=1)
-            else:
-                # According to micro benchmark results, torch.compile can get better performance for small token.
-                if tokens_in_chunk <= 32:
-                    moe_sum_reduce_torch_compile(
-                        intermediate_cache3.view(*intermediate_cache3.shape),
-                        out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                        routed_scaling_factor,
-                    )
-                else:
-                    moe_sum_reduce(
-                        intermediate_cache3.view(*intermediate_cache3.shape),
-                        out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                        routed_scaling_factor,
-                    )
-
-        elif _is_hip:
-            if _use_aiter:
-                moe_sum(
-                    intermediate_cache3.view(*intermediate_cache3.shape),
-                    out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                )
-            else:
-                # According to micro benchmark results, torch.compile can get better performance for small token.
-                if tokens_in_chunk <= 32:
-                    moe_sum_reduce_torch_compile(
-                        intermediate_cache3.view(*intermediate_cache3.shape),
-                        out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                        routed_scaling_factor,
-                    )
-                else:
-                    moe_sum_reduce_triton(
-                        intermediate_cache3.view(*intermediate_cache3.shape),
-                        out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                        routed_scaling_factor,
-                    )
-        elif _is_xpu:
-            moe_sum_reduce(
-                intermediate_cache3.view(*intermediate_cache3.shape),
-                out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                routed_scaling_factor,
-            )
-        else:
-            if _has_vllm_ops:
-                vllm_ops.moe_sum(
-                    intermediate_cache3.view(*intermediate_cache3.shape),
-                    out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                )
-            else:
-                # Fallback: use triton moe_sum_reduce when vllm is not available
-                moe_sum_reduce_triton(
-                    intermediate_cache3.view(*intermediate_cache3.shape),
-                    out_hidden_states[begin_chunk_idx:end_chunk_idx],
-                    routed_scaling_factor,
-                )
-
-    return out_hidden_states
 
 
 def fused_moe(
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py b/python/sglang/srt/layers/moe/moe_runner/triton_utils/fused_moe_triton_config.py
similarity index 99%
rename from python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/fused_moe_triton_config.py
index 90a6a31c7..fb1387808 100644
--- a/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py
+++ b/python/sglang/srt/layers/moe/moe_runner/triton_utils/fused_moe_triton_config.py
@@ -211,7 +211,7 @@ def try_get_optimal_moe_config(
     per_channel_quant: bool = False,
     return_down_config: bool = False,
 ):
-    from sglang.srt.layers.moe.fused_moe_triton import get_config
+    from sglang.srt.layers.moe.moe_runner.triton_utils import get_config
 
     down_config = None
     max_block_m = None
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_kernels.py b/python/sglang/srt/layers/moe/moe_runner/triton_utils/fused_moe_triton_kernels.py
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_kernels.py
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/fused_moe_triton_kernels.py
diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/moe_align_block_size.py b/python/sglang/srt/layers/moe/moe_runner/triton_utils/moe_align_block_size.py
similarity index 100%
rename from python/sglang/srt/layers/moe/fused_moe_triton/moe_align_block_size.py
rename to python/sglang/srt/layers/moe/moe_runner/triton_utils/moe_align_block_size.py
diff --git a/python/sglang/srt/layers/moe/topk.py b/python/sglang/srt/layers/moe/topk.py
index c65942147..77660ed55 100644
--- a/python/sglang/srt/layers/moe/topk.py
+++ b/python/sglang/srt/layers/moe/topk.py
@@ -1042,7 +1042,7 @@ def _post_process_topk_ids(
         )
 
         # Lazy import to avoid circular-import issues
-        from sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_kernels import (
+        from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_kernels import (
             fused_append_shared_experts,
         )
 
diff --git a/python/sglang/srt/layers/quantization/bitsandbytes.py b/python/sglang/srt/layers/quantization/bitsandbytes.py
index 51b92624f..6e20992d9 100644
--- a/python/sglang/srt/layers/quantization/bitsandbytes.py
+++ b/python/sglang/srt/layers/quantization/bitsandbytes.py
@@ -475,7 +475,7 @@ class BitsAndBytesMoEMethod(FusedMoEMethodBase):
         layer: torch.nn.Module,
         dispatch_output: StandardDispatchOutput,
     ) -> CombineInput:
-        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.token_dispatcher import StandardCombineInput
 
         x = dispatch_output.hidden_states
diff --git a/python/sglang/srt/lora/lora_moe_runner_marlin.py b/python/sglang/srt/lora/lora_moe_runner_marlin.py
index 58ee9ed01..89b451fe6 100644
--- a/python/sglang/srt/lora/lora_moe_runner_marlin.py
+++ b/python/sglang/srt/lora/lora_moe_runner_marlin.py
@@ -29,10 +29,10 @@ if _is_cuda:
     from sglang.srt.layers.moe.fused_moe_triton.fused_marlin_moe import (
         get_scalar_type,
     )
-    from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
+    from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import (
         moe_align_block_size,
     )
-    from sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_kernels import (
+    from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_kernels import (
         moe_sum_reduce_triton,
     )
     from sglang.srt.layers.quantization.marlin_utils import marlin_make_workspace
diff --git a/python/sglang/srt/lora/triton_ops/virtual_experts.py b/python/sglang/srt/lora/triton_ops/virtual_experts.py
index 22b5b55ae..3da01ea58 100644
--- a/python/sglang/srt/lora/triton_ops/virtual_experts.py
+++ b/python/sglang/srt/lora/triton_ops/virtual_experts.py
@@ -417,7 +417,7 @@ def _merged_experts_fused_moe_lora_add_impl(
         weight: torch.Tensor,
         stage_top_k: int,
     ) -> dict[str, Any]:
-        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_dtype_str,
             try_get_optimal_moe_config,
         )
@@ -459,7 +459,7 @@ def _merged_experts_fused_moe_lora_add_impl(
         # The native align kernel consumes num_experts + 1 internally for its
         # sentinel bucket, so the 1024-expert boundary must use the fallback path.
         if num_experts < 1024:
-            from sglang.srt.layers.moe.fused_moe_triton.moe_align_block_size import (
+            from sglang.srt.layers.moe.moe_runner.triton_utils.moe_align_block_size import (
                 moe_align_block_size as native_moe_align_block_size,
             )
 
@@ -513,7 +513,7 @@ def _merged_experts_fused_moe_lora_add_impl(
 
         return result
 
-    from sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_kernels import (
+    from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_kernels import (
         invoke_fused_moe_kernel,
     )
 
diff --git a/python/sglang/srt/models/afmoe.py b/python/sglang/srt/models/afmoe.py
index b5ce2afbb..50548c1cc 100644
--- a/python/sglang/srt/models/afmoe.py
+++ b/python/sglang/srt/models/afmoe.py
@@ -46,8 +46,8 @@ from sglang.srt.layers.linear import (
     RowParallelLinear,
 )
 from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.moe.fused_moe_triton import fused_moe
 from sglang.srt.layers.moe.moe_runner import MoeRunnerConfig
+from sglang.srt.layers.moe.moe_runner.triton_utils import fused_moe
 from sglang.srt.layers.moe.topk import TopK
 from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
diff --git a/python/sglang/srt/models/dbrx.py b/python/sglang/srt/models/dbrx.py
index 0c7e16052..5ed65a304 100644
--- a/python/sglang/srt/models/dbrx.py
+++ b/python/sglang/srt/models/dbrx.py
@@ -35,8 +35,8 @@ from sglang.srt.layers.linear import (
     RowParallelLinear,
 )
 from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
 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.topk import TopK
 from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
diff --git a/python/sglang/srt/models/deepseek.py b/python/sglang/srt/models/deepseek.py
index 6baa1fe0f..21423aca5 100644
--- a/python/sglang/srt/models/deepseek.py
+++ b/python/sglang/srt/models/deepseek.py
@@ -36,8 +36,8 @@ from sglang.srt.layers.linear import (
     RowParallelLinear,
 )
 from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.moe.fused_moe_triton import fused_moe
 from sglang.srt.layers.moe.moe_runner import MoeRunnerConfig
+from sglang.srt.layers.moe.moe_runner.triton_utils import fused_moe
 from sglang.srt.layers.moe.topk import TopK
 from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
diff --git a/python/sglang/srt/models/qwen2_moe.py b/python/sglang/srt/models/qwen2_moe.py
index de026c560..4cafaab6e 100644
--- a/python/sglang/srt/models/qwen2_moe.py
+++ b/python/sglang/srt/models/qwen2_moe.py
@@ -340,7 +340,7 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
         if shared_weights is None:
             return topk_output
 
-        from sglang.srt.layers.moe.fused_moe_triton.fused_moe_triton_kernels import (
+        from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe_triton_kernels import (
             fused_append_shared_experts_with_weights,
         )
 
diff --git a/python/sglang/srt/models/xverse_moe.py b/python/sglang/srt/models/xverse_moe.py
index 9b88060d5..8418c83fc 100644
--- a/python/sglang/srt/models/xverse_moe.py
+++ b/python/sglang/srt/models/xverse_moe.py
@@ -36,8 +36,8 @@ from sglang.srt.layers.linear import (
     RowParallelLinear,
 )
 from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
 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.topk import TopK
 from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
diff --git a/python/sglang/test/test_block_fp8.py b/python/sglang/test/test_block_fp8.py
index 6b73a736a..a4f26d500 100644
--- a/python/sglang/test/test_block_fp8.py
+++ b/python/sglang/test/test_block_fp8.py
@@ -5,7 +5,7 @@ from functools import lru_cache
 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 (
     per_tensor_quant_mla_fp8,
diff --git a/python/sglang/test/test_cutlass_moe.py b/python/sglang/test/test_cutlass_moe.py
index 4e4eee376..d57d9d63c 100755
--- a/python/sglang/test/test_cutlass_moe.py
+++ b/python/sglang/test/test_cutlass_moe.py
@@ -6,8 +6,8 @@ import triton.testing  # Added import
 from transformers import AutoConfig
 
 from sglang.srt.layers.moe.cutlass_moe import cutlass_fused_experts_fp8
-from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_experts
 from sglang.srt.layers.moe.moe_runner.base import MoeRunnerConfig
+from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import fused_experts
 from sglang.srt.layers.moe.topk import StandardTopKOutput
 
 
diff --git a/test/manual/test_triton_moe_wna16.py b/test/manual/test_triton_moe_wna16.py
index 1c3928a9d..ac198e4cb 100644
--- a/test/manual/test_triton_moe_wna16.py
+++ b/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
diff --git a/test/registered/lora/test_lora_moe_runner.py b/test/registered/lora/test_lora_moe_runner.py
index eab94a4e6..437f6b1c0 100644
--- a/test/registered/lora/test_lora_moe_runner.py
+++ b/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)
diff --git a/test/registered/lora/test_marlin_lora_correctness.py b/test/registered/lora/test_marlin_lora_correctness.py
index 696fb5f8d..97a3d71e4 100644
--- a/test/registered/lora/test_marlin_lora_correctness.py
+++ b/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)
diff --git a/test/registered/moe/test_fused_moe.py b/test/registered/moe/test_fused_moe.py
index 2b267fa9e..b67d1b4ed 100644
--- a/test/registered/moe/test_fused_moe.py
+++ b/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
diff --git a/test/registered/moe/test_triton_moe_channel_fp8_kernel.py b/test/registered/moe/test_triton_moe_channel_fp8_kernel.py
index 4acbc9ab2..49bdff402 100644
--- a/test/registered/moe/test_triton_moe_channel_fp8_kernel.py
+++ b/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
diff --git a/test/registered/quant/test_block_int8.py b/test/registered/quant/test_block_int8.py
index 68829f90f..aa0c8bae5 100644
--- a/test/registered/quant/test_block_int8.py
+++ b/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
diff --git a/test/registered/quant/test_int8_kernel.py b/test/registered/quant/test_int8_kernel.py
index 84591e3b4..4114887dd 100644
--- a/test/registered/quant/test_int8_kernel.py
+++ b/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