The same storing dK into a fused projection's gradient (skProjectionEpilogue).
3137def streamingAttentionKeyHeadsFirstProjectionSM86 =
3138 (lambda unrestricted seq : Nat . (lambda unrestricted heads : Nat .
3139 (lambda unrestricted keyValueHeads : Nat . (lambda unrestricted width : Nat .
3140 (lambda erased admitted : (equal Nat (streamingAttentionKeyProjectionAdmitted seq heads keyValueHeads width) 1) .
3141 (sm86GuardLateReads
3142 (sm86TensorProductsFixed 12
3143 (streamingAttentionKeyGroupedOrderedWithSM86 (skProjectionEpilogue width) 1 seq heads keyValueHeads 0 1))))))))The compiler supplied declaration spans and resolved links from this source snapshot. This page does not assert that this file belongs to a checked closure.