Andrii Staikov
Andrii Staikov
[Transformations] Make ov::ModelPass transformations execute recursively ### Details: Some ov::ModelPass transformations lack recursive execution for subgraphs leaving it not processed. Add the required recursive call for MultiSubGraphOp operations. ### Tickets:...
wip: added the setup part and half of the first matcher
Deduce the number of KV heads and head_size from the model without relying on HF config, and set the deduced values as KV cache input dimension. Applied HW specific layout...
### Details: - Add a ConcatToTile transformation to replace Concat having inputs from the same output with a Tile or Broadcast - Add a test for the ConcatToTile transformation -...
Add unique models to precommit PA tests Change the list of models covering all the big real models. ### Tickets: - CVS-145243 Signed-off-by: Andrii Staikov
[TRANSFORMATIONS] Make TotalSequenceLengthPattern pattern stricter Make TotalSequenceLengthPattern pattern stricter to match one of the cases when 'scale' is calculated from shape. ### Tickets: - CVS-138933 Signed-off-by: Andrii Staikov
Fix Coverity issues Fix the following issues reported by Coverity: * Overflowed constant * Explicit null dereferenced * Dereference null return value * Using invalid iterator ### Tickets: - CVS-145094...
Add MakeSDPA attention Add MakeSDPA attention transformation that inserts a ScaledDotProductAttention node for the future processing in SDPAToPagedAttention. Previously this was done via Optimum-Intel Currently supported models: * katuni4ka/tiny-random-jais ###...
Remove usage of deprecated get_ie_output_name() function Fix OPENVINO_DEBUG usages Signed-off-by: Andrii Staikov