Usage of CUDA Graphs is limited to models where-in all the model ops (graph nodes) can be partitioned