Class ov::pass::SDPAToPagedAttention

class SDPAToPagedAttention : public ov::pass::ModelPass

The transformation replaces KV-cache processing part in LLMs by PagedAttention operation. NOTE: The transformation may throw an exception when some configuration of the model failed: i.e. the SDPA node is absent in the model. This means the graph cannot be processed for the PA scenario, so the GenAI pipeline (the only pipeline the transformation is used in so far) will fallback to the SDPA implementaion and run inference using it.