I want to evaluate the performance of a model. But I want to do a layer-wise evaluation for this I would need to split the DL model into layers/ sub-graphs. Do yall have any suggestions or resources where this has been done. I know ONNX lets you create a model sub-graph but that’t not specific to each layer. What if I want to split the models based on layers and not the nodes?

This is a little ahead but resnet had 515 nodes. Is is possible to create 10 subgraphs and evaluate each of their perfoemances with some profiling?

At the moment I am trying to explore creating sub-graphs of a given model (resnet.onnx). What are some ways of evaluating inference performance for specific layers/ sub-graphs