If you consider them lossy compression, then yes.

The goal would be to find the minimum model that, with a fixed seed, would exactly reproduce your text.