Mila::Dnn::Quant::Weight::AttentionPrecisionRoles Concept Reference

Fused QKV projection and the attention output projection. More...

Concept definition

        template<typename TPlan>
        concept AttentionPrecisionRoles =
        requires { typename TPlan::QkvProjection; typename TPlan::OutputProjection; }
        && WeightQuantPolicy<typename TPlan::QkvProjection>
        && WeightQuantPolicy<typename TPlan::OutputProjection>

Detailed Description

Fused QKV projection and the attention output projection.