Class ov::pass::FakeConvertDecomposition

class FakeConvertDecomposition : public ov::pass::MatcherPass

FakeConvertDecomposition transformation decomposes FakeConvert layer. f8: f8e4m3, f8e5m2 downconvert: f32->f8, f16->f8, bf16->f8 upconvert: f8->f32, f8->f16, f8->bf16 output = (upconvert(downconvert(input * scale - shift)) + shift) / scale.