Class ov::pass::MergeDQ#
-
class MergeDQ : public ov::pass::MatcherPass#
Description: merges Dequantize (DQ) subgraphs used in MHA.
Detects repeated per-head dequantization patterns and merges them to enable packing of projections.
Site Navigation
Section Navigation
Description: merges Dequantize (DQ) subgraphs used in MHA.
Detects repeated per-head dequantization patterns and merges them to enable packing of projections.