Skip to content

vllm.model_executor.layers.fused_moe.router.fused_topk_router

FusedTopKRouter

Bases: BaseRouter

Default router using standard fused top-k routing.

Source code in vllm/model_executor/layers/fused_moe/router/fused_topk_router.py 
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
class FusedTopKRouter(BaseRouter):
    """Default router using standard fused top-k routing."""

    def __init__(
        self,
        top_k: int,
        global_num_experts: int,
        eplb_state: EplbLayerState,
        scoring_func: str = "softmax",
        renormalize: bool = True,
        enable_eplb: bool = False,
        indices_type_getter: Callable[[], torch.dtype | None] | None = None,
    ):
        super().__init__(
            top_k=top_k,
            global_num_experts=global_num_experts,
            eplb_state=eplb_state,
            enable_eplb=enable_eplb,
            indices_type_getter=indices_type_getter,
        )
        self.renormalize = renormalize
        self.scoring_func = scoring_func

    @property
    def routing_method_type(self) -> RoutingMethodType:
        return get_routing_method_type(
            scoring_func=self.scoring_func,
            top_k=self.top_k,
            renormalize=self.renormalize,
            num_expert_group=None,
            has_e_score_bias=False,
        )

    def _compute_routing(
        self,
        hidden_states: torch.Tensor,
        router_logits: torch.Tensor,
        indices_type: torch.dtype | None,
    ) -> tuple[torch.Tensor, torch.Tensor]:
        """Compute routing using standard fused top-k."""
        topk_weights, topk_ids, token_expert_indices = fused_topk(
            hidden_states=hidden_states,
            gating_output=router_logits,
            topk=self.top_k,
            renormalize=self.renormalize,
            indices_type=indices_type,
            scoring_func=self.scoring_func,
        )

        return topk_weights, topk_ids

_compute_routing 

_compute_routing(
    hidden_states: Tensor,
    router_logits: Tensor,
    indices_type: dtype | None,
) -> tuple[Tensor, Tensor]

Compute routing using standard fused top-k.

Source code in vllm/model_executor/layers/fused_moe/router/fused_topk_router.py 
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
def _compute_routing(
    self,
    hidden_states: torch.Tensor,
    router_logits: torch.Tensor,
    indices_type: torch.dtype | None,
) -> tuple[torch.Tensor, torch.Tensor]:
    """Compute routing using standard fused top-k."""
    topk_weights, topk_ids, token_expert_indices = fused_topk(
        hidden_states=hidden_states,
        gating_output=router_logits,
        topk=self.top_k,
        renormalize=self.renormalize,
        indices_type=indices_type,
        scoring_func=self.scoring_func,
    )

    return topk_weights, topk_ids