Filip Naudot

llmSHAP

Maintainer. Developed and currently maintaining the LLM explainability library llmSHAP, which provides Shapley value-based feature attributions for LLMs.

View on GitHub

Quantitative-Bipolar-Argumentation

Contributor. Modification to handle sets of contributors, enabling the quantification of multiple arguments collectively contributing to a topic argument.

View on GitHub

lightbench

Maintainer. Developed and currently maintaining the LLM benchmarking framework lightbench, which enables automatic evaluation for code and text generation tasks.

View on GitHub

fantastic-arg-tools

Contributor. Contributed to visuals and filtering improvements in analysis charts.

View on GitHub

llmSHAP: A Principled Approach to LLM Explainability

Filip Naudot, Tobias Sundqvist, Timotheus Kampik. (2025). (Preprint, arXiv).

View on arXiv

Set Contribution Functions for Quantitative Bipolar Argumentation and their Principles

Filip Naudot, Andreas Brännström, Vicenç Torra, Timotheus Kampik. (2025). (Preprint, arXiv).

View on arXiv

Performance and Computational Demands of LLMs: Impact of Model Size and Quantization

Naudot, Filip. (2025). In Proceedings of Umeå's 28th Student Conference in Computing Science (USCCS 2025), edited by Thomas Hellström, Umeå University, Sweden.

View on DiVA

Scalable Feature Attribution in LLM-Based Recommender Systems

Naudot, Filip. (2025). Scalable Feature Attribution in LLM-Based Recommender Systems, M.Sc. thesis, Department of Computing Science, Umeå University, Sweden.

View on DiVA

Winner - Reinforcement Learning Tournament

Artificial Intelligence - Methods and Applications (Ht24), Umeå University, 2024.