openvino
openvino copied to clipboard
[CPU] only allow per-oc or per-tensor FQ fusing into FC
Details:
- Add a check to reject non-supported FakeQuantize from fusing into FC node, so they can run in standalone mode w/o causing exceptions when composing oneDNN postOps.
- port from https://github.com/openvinotoolkit/openvino/pull/23009
- add test case
Tickets:
- CVS-131890