vvas_xinfer supports models where the Vitis AI compiler assigns a subset of
operators to CPU execution (CPU subgraphs). These operators run on the CPU
while the remaining subgraphs run on the AI Engine. Execution order and data
handoff between subgraphs are managed automatically — no additional pipeline
elements or configuration keys are required.
When using the VART backend, set input-tensor-type and
output-tensor-type in vart-config according to the subgraph boundaries
of your compiled model.