rm router logits Improve TTOP 3ms (#1407)

### What this PR does / why we need it?

The previous code is
router_logits, _ = self.gate(hidden_states)
hidden_states = get_dp_group().all_gather(hidden_states, 0)
router_logits = get_dp_group().all_gather(router_logits, 0)
I want to change the two all_gathers to one, reduce one all_gather
communication, and make it
hidden_states = get_dp_group().all_gather(hidden_states, 0)
router_logits, _ = self.gate(hidden_states)

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?
bash examples/run_dp_attention_etp16.sh
bash examples/run_dp_attention_etp16_benmark.sh

gsm8k accuracy verification
<img width="1809" alt="截屏2025-06-24 21 53 24"
src="https://github.com/user-attachments/assets/47eace3b-a86b-41b4-9de8-773f57fea33b"
/>



- vLLM version: v0.9.2
- vLLM main:
vllm-project/vllm@77f77a9

---------

Signed-off-by: ttanzhiqiang <389825161@qq.com>

Author: ttanzhiqiang

vllm_ascend/models/deepseek_v2.py

@@ -367,6 +367,7 @@ def __init__(
         self.ep_group = get_ep_group()
 
         self.params_dtype = torch.get_default_dtype()
+        self.rm_router_logits = self.experts.rm_router_logits
 
     def forward(self,
                 hidden_states: torch.Tensor,
@@ -389,7 +390,9 @@ def forward(self,
                 is_prefill = is_prefill or attn_metadata.with_prefill_across_dp
 
         # router_logits: (num_tokens, n_experts)
-        router_logits, _ = self.gate(hidden_states)
+        router_logits = None
+        if not self.rm_router_logits:
+            router_logits, _ = self.gate(hidden_states)
 
         experts_hidden_states = self.experts(
             hidden_states=hidden_states,
@@ -398,6 +401,7 @@ def forward(self,
             top_k=CustomDeepseekV2MoE.top_k,
             enable_force_load_balance=enable_force_load_balance,
             shared_experts=self.shared_experts,
+            gate=self.gate,
             replace_allreduce=replace_allreduce)
 
         hidden_states = (

vllm_ascend/ops/fused_moe.py

@@ -45,7 +45,8 @@
 from vllm_ascend.ops.expert_load_balancer import ExpertLoadBalancer
 from vllm_ascend.utils import (FusedMoEState, dispose_tensor,
                                get_all_reduce_merge_state, get_fused_moe_state,
-                               is_310p, npu_stream_switch, npu_wait_tensor)
+                               get_rm_router_logits_state, is_310p,
+                               npu_stream_switch, npu_wait_tensor)
 
 MOE_ALL2ALL_BUFFER: bool = envs_ascend.MOE_ALL2ALL_BUFFER
 
@@ -1148,6 +1149,8 @@ def __init__(
         self.global_redundant_expert_num = 0
 
         is_deepseek_v3_r1 = self.global_num_experts == 256
+        self.rm_router_logits = get_rm_router_logits_state(
+            self.moe_parallel_config.ep_size, self.dp_size, is_deepseek_v3_r1)
         self.all_reduce_merge = get_all_reduce_merge_state(
             self.moe_parallel_config.ep_size, is_deepseek_v3_r1)
 
@@ -1240,7 +1243,9 @@ def forward(self,
                 enable_force_load_balance: bool = False,
                 top_k: Optional[int] = None,
                 shared_experts: Optional[Any] = None,
+                gate=None,
                 replace_allreduce: bool = False):
+
         assert self.quant_method is not None
 
         if top_k:
@@ -1277,6 +1282,7 @@ def forward(self,
             tp_rank = get_tensor_model_parallel_rank()
             hidden_states = chunk_hidden_states[tp_rank]
             router_logits = chunk_router_logits[tp_rank]
+
         if self.dp_size > 1:
             if fused_moe_state == FusedMoEState.AllGather:
                 # NOTE: When in torchair graph, it has been padded in model_runner_v1
@@ -1289,19 +1295,27 @@ def forward(self,
                                 hidden_states,
                                 (0, 0, 0,
                                  max_num_tokens_across_dp - num_tokens))
-                            router_logits = nn.functional.pad(
-                                router_logits,
-                                (0, 0, 0,
-                                 max_num_tokens_across_dp - num_tokens))
+                            if not self.rm_router_logits:
+                                router_logits = nn.functional.pad(
+                                    router_logits,
+                                    (0, 0, 0,
+                                     max_num_tokens_across_dp - num_tokens))
                 hidden_states = get_dp_group().all_gather(hidden_states, 0)
-                router_logits = get_dp_group().all_gather(router_logits, 0)
+                if self.rm_router_logits:
+                    router_logits, _ = gate(hidden_states)
+                else:
+                    router_logits = get_dp_group().all_gather(router_logits, 0)
+
             elif fused_moe_state == FusedMoEState.NaiveMulticast:
                 cu_tokens_across_dp_cpu = get_forward_context(
                 ).dp_metadata.cu_tokens_across_dp_cpu
                 hidden_states = self.naive_multicast(hidden_states,
                                                      cu_tokens_across_dp_cpu)
-                router_logits = self.naive_multicast(router_logits,
-                                                     cu_tokens_across_dp_cpu)
+                if self.rm_router_logits:
+                    router_logits, _ = gate(hidden_states)
+                else:
+                    router_logits = self.naive_multicast(
+                        router_logits, cu_tokens_across_dp_cpu)
 
         # Matrix multiply.
         e_hidden_states = self.quant_method.apply(

vllm_ascend/utils.py

@@ -439,6 +439,22 @@ class FusedMoEState(Enum):
     NaiveMulticast = 4
 
 
+# TODO(ttanzhiqiang): rm_router_logits
+# dp>1 will trigger
+# In theory, this solution is only applicable to AllGather and AllGatherEP, because in the dp scenario, the previous operation was gate + two communications, and now it is changed to one communication + gate operation, which can save some communication time. In theory, all moe AllGather and AllGatherEP solutions can follow this logic, but now other moe models (qwen3-235b) dp solutions are not adjusted, so use the switch to control it to prevent code errors.
+def get_rm_router_logits_state(ep_size: int, dp_size: int,
+                               is_deepseek_v3_r1: bool):
+    # the fusion operator torch_npu.npu_grouped_matmul_finalize_routing called by allgather ep
+    # only supports deepseek v3/r1
+    if dp_size > 1:
+        if (envs.VLLM_ENABLE_FUSED_EXPERTS_ALLGATHER_EP and ep_size > 1
+                and is_deepseek_v3_r1):
+            return True
+        elif ep_size == 1 and is_deepseek_v3_r1:
+            return True
+    return False
+
+
 # TODO(ttanzhiqiang): all_reduce merge
 # When all_reduce_merge is in progress, shared_experts does not do all_reduce in mlp, but waits until shared_experts+router_experts are completed before doing all_reduce
 # Currently, all_reduce_merge is enabled by default in the AllGather, AllGatherEP and NaiveMulticast scenarios of the deepseek model.