Disclaimer :
This implementation is for educational and research purposes only. The SIPIT algorithm is the intellectual property of its authors, and this repository does not claim any rights over the algorithm itself.
This repository provides an implementation of the SIPIT algorithm, as described in the paper:
"Language Models Are Injective and Hence Invertible" (Giorgos Nikolaou, Tommaso Mencattini, Donato Crisostomi, Andrea Santilli, Yannis Panagakis, Emanuele Rodolà) [1].
The SIPIT algorithm leverages the injective property of large language models (LLMs): given a distinct input, an LLM produces a distinct output.
By treating LLMs as 1-1 functions, SIPIT efficiently reverses the output back to the input, in such an efficient way -compared to raw brute-force methods- that makes it doable with commodity hardware.
This enables opportunities such as:
- Input recovery from model outputs.
- Privacy and security analysis of LLM-generated content.
- Model interpretability and inversion studies.
make install
make run