TensorFI2 Non Linear Model Support for layer outputs

[sabujlaskar]

Here we provide two features.

1. Non-linear model support for layer_outputs with single fault injection mode
2. Batch fault injection support

More details can be found in this link

Testing command:

python test_non_linear_fi.py ResNet50 experiments/layer-outputs/confFiles/sample.yaml 20 224

Here 20 is the total fault number and 224 is the required dimension for the ResNet50 model.

Model graphs and supernodes are model-related. So, if multiple fault injection on the same model is required, then we can compute them priorly and pass them as parameters to this function to make the process faster(Line 41-43 in tensorfi2.py).