-
laibao authored
移除 forward 中对 experts.use_overlapped/_shared_experts 的状态改写,避免 torch.compile 启动期 shared/non-shared 路径不一致 FusedMoE.forward_impl 仅在 shared_output 为空时计算 shared experts,防止透传值被本地重算覆盖
eb933fe1
移除 forward 中对 experts.use_overlapped/_shared_experts 的状态改写,避免 torch.compile 启动期 shared/non-shared 路径不一致 FusedMoE.forward_impl 仅在 shared_output 为空时计算 shared experts,防止透传值被本地重算覆盖