References

MLLM-SHAP is used for research purposes on behaviour of large language models (LLMs) and their explainability using SHAP values. Research paper and report is available at arXiv. All code is publicly available at official Github repository and can be used as further example on how to use package and how to interpret its results.

Package is fully tested, including analysis of correctness on different approximations algorithms of SHAP values, available here. Following image shows comparison of different SHAP approximations on same input and model - Liquid Audio LLM.

# TODO: add image