Static analysis#
Overview#
AIDGE propose a set of functions which compute and aggregate metrics, computed without running an inference, which allow the user to evaluate the model.
Get node KPI#
Computational footprint#
Give the number of Multiply-ACcumulate (MAC) operations per operator.
Memory footprint#
We can dinstiguish two memory footprints:
Compute buffers memory: Memory footprint of inputs/outputs per operator
Parameters memory: Memory footprint of parameters per operator
Aggregate KPI from a subgraph#
Agregate the KPI previously described for a subgraph. This allow to put them on a relative scale.
Example :