Explain Wrapper for torch-influence

[dilyabareeva]

https://github.com/alstonlo/torch-influence

Advice from Galip:

giving depth and repeat parameters very low (1-2 ) should ensure low computation time
possibly override model.named_params()