GMM custom operator optimization in small batch scenarios (#7100)

### What this PR does / why we need it?
GMM custom operator optimization in small batch scenarios

### How was this patch tested?

Qwen3-30B input: 4k, output: 1k

batch 1：
TPOT 7.9 ms -> 7.0 ms
Output Token Throughput 125.4651 token/s -> 140.6278 token/s

batch 2：
TPOT 9.4 ms -> 8.8 ms
Output Token Throughput 211.8187 token/s -> 225.2254 token/s

batch 16：
TPOT 13.6 ms -> 13.5 ms
Output Token Throughput 1159.8213 token/s -> 1165.0982 token/s

- vLLM version: v0.16.0
- vLLM main:
4034c3d32e

---------

Signed-off-by: chenxi-hh <chen464822955@163.com>

csrc/torch_binding.cpp

@@ -697,7 +697,7 @@ std::vector<at::Tensor> moe_grouped_matmul(
     y.emplace_back(y_0);
     at::TensorList result = at::TensorList(y);
 
-    EXEC_NPU_CMD(aclnnMoeGroupedMatmulWeightNz,
+    EXEC_NPU_CMD(aclnnMoeGroupedMatmul,
                 x_list, weight_list, group_list, transpose_weight, result);
 
     return y;

vllm_ascend/device/device_op.py

@@ -17,6 +17,7 @@
 #
 import torch
 import torch_npu
+from vllm.forward_context import get_forward_context
 
 from vllm_ascend.device.mxfp_compat import (
     FLOAT4_E2M1FN_X2_DTYPE,
@@ -27,6 +28,8 @@ from vllm_ascend.utils import AscendDeviceType, get_ascend_device_type
 
 
 class BaseDeviceAdaptor:
+    small_batch_gmm_batch_num = 16
+
     @classmethod
     def reshape_and_cache(cls, key, value, key_cache, value_cache, slot_mapping):
         torch_npu._npu_reshape_and_cache(
@@ -46,6 +49,21 @@ class BaseDeviceAdaptor:
         active_expert_range=None,
         quant_mode: int = -1,
     ):
+        # In small batch and non-quantization scenarios, npu_moe_init_routing_v2 is more efficient.
+        # It is expected that further improvements will be made after it is incorporated into CANN on June 30th.
+        if quant_mode == -1 and get_forward_context().num_tokens <= DeviceOperator.small_batch_gmm_batch_num:
+            return torch_npu.npu_moe_init_routing_v2(
+                hidden_states,
+                topk_ids,
+                scale=scale,
+                active_num=active_num,
+                expert_num=expert_num,
+                expert_tokens_num_type=2,
+                expert_tokens_num_flag=expert_tokens_num_flag,
+                active_expert_range=active_expert_range,
+                quant_mode=quant_mode,
+            )
+        else:
             return torch.ops._C_ascend.npu_moe_init_routing_custom(
                 hidden_states,
                 topk_ids,

vllm_ascend/ops/fused_moe/moe_mlp.py

@@ -18,6 +18,7 @@
 import torch
 import torch_npu
 from torch.nn.functional import pad
+from vllm.forward_context import get_forward_context
 from vllm.triton_utils import HAS_TRITON
 
 from vllm_ascend.ascend_forward_context import _EXTRA_CTX, MoECommType
@@ -339,6 +340,17 @@ def unquant_apply_mlp(
         w1 = w1.transpose(1, 2)
         w2 = w2.transpose(1, 2)
 
+    # In the small batch scenario, use _C_ascend.moe_grouped_matmul
+    if group_list.dim() == 2 and get_forward_context().num_tokens <= DeviceOperator.small_batch_gmm_batch_num:
+        gate_up_out = torch.ops._C_ascend.moe_grouped_matmul(
+            x=hidden_states,
+            weight=w1,
+            split_item=2,
+            group_list_type=group_list_type,
+            group_type=0,
+            group_list=group_list,
+        )[0]
+    else:
         gate_up_out = torch_npu.npu_grouped_matmul(
             x=[hidden_states],
             weight=[w1],
@@ -358,6 +370,17 @@ def unquant_apply_mlp(
     if topk_scales is not None:
         gate_up_out *= topk_scales
 
+    # In the small batch scenario, use _C_ascend.moe_grouped_matmul
+    if group_list.dim() == 2 and get_forward_context().num_tokens <= DeviceOperator.small_batch_gmm_batch_num:
+        hidden_states = torch.ops._C_ascend.moe_grouped_matmul(
+            x=gate_up_out,
+            weight=w2,
+            split_item=2,
+            group_list_type=group_list_type,
+            group_type=0,
+            group_list=group_list,
+        )[0]
+    else:
         hidden_states = torch_npu.npu_grouped_matmul(
             x=[gate_up_out],
             weight=[w2],