GiLOT: Interpreting Generative Language Models via Optimal Transport

Xuhong Li, Jiamin Chen, Yekun Chai, Haoyi Xiong
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:27515-27530, 2024.

Abstract

While large language models (LLMs) surge with the rise of generative AI, algorithms to explain LLMs highly desire. Existing feature attribution methods adequate for discriminative language models like BERT often fail to deliver faithful explanations for LLMs, primarily due to two issues: (1) For every specific prediction, the LLM outputs a probability distribution over the vocabulary–a large number of tokens with unequal semantic distance; (2) As an autoregressive language model, the LLM handles input tokens while generating a sequence of probability distributions of various tokens. To address above two challenges, this work proposes GiLOT that leverages Optimal Transport to measure the distributional change of all possible generated sequences upon the absence of every input token, while taking into account the tokens’ similarity, so as to faithfully estimate feature attribution for LLMs. We have carried out extensive experiments on top of Llama families and their fine-tuned derivatives across various scales to validate the effectiveness of GiLOT for estimating the input attributions. The results show that GiLOT outperforms existing solutions on a number of faithfulness metrics under fair comparison settings. Source code is publicly available at https://github.com/holyseven/GiLOT.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-li24i, title = {{G}i{LOT}: Interpreting Generative Language Models via Optimal Transport}, author = {Li, Xuhong and Chen, Jiamin and Chai, Yekun and Xiong, Haoyi}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {27515--27530}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/li24i/li24i.pdf}, url = {https://proceedings.mlr.press/v235/li24i.html}, abstract = {While large language models (LLMs) surge with the rise of generative AI, algorithms to explain LLMs highly desire. Existing feature attribution methods adequate for discriminative language models like BERT often fail to deliver faithful explanations for LLMs, primarily due to two issues: (1) For every specific prediction, the LLM outputs a probability distribution over the vocabulary–a large number of tokens with unequal semantic distance; (2) As an autoregressive language model, the LLM handles input tokens while generating a sequence of probability distributions of various tokens. To address above two challenges, this work proposes GiLOT that leverages Optimal Transport to measure the distributional change of all possible generated sequences upon the absence of every input token, while taking into account the tokens’ similarity, so as to faithfully estimate feature attribution for LLMs. We have carried out extensive experiments on top of Llama families and their fine-tuned derivatives across various scales to validate the effectiveness of GiLOT for estimating the input attributions. The results show that GiLOT outperforms existing solutions on a number of faithfulness metrics under fair comparison settings. Source code is publicly available at https://github.com/holyseven/GiLOT.} }
Endnote
%0 Conference Paper %T GiLOT: Interpreting Generative Language Models via Optimal Transport %A Xuhong Li %A Jiamin Chen %A Yekun Chai %A Haoyi Xiong %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-li24i %I PMLR %P 27515--27530 %U https://proceedings.mlr.press/v235/li24i.html %V 235 %X While large language models (LLMs) surge with the rise of generative AI, algorithms to explain LLMs highly desire. Existing feature attribution methods adequate for discriminative language models like BERT often fail to deliver faithful explanations for LLMs, primarily due to two issues: (1) For every specific prediction, the LLM outputs a probability distribution over the vocabulary–a large number of tokens with unequal semantic distance; (2) As an autoregressive language model, the LLM handles input tokens while generating a sequence of probability distributions of various tokens. To address above two challenges, this work proposes GiLOT that leverages Optimal Transport to measure the distributional change of all possible generated sequences upon the absence of every input token, while taking into account the tokens’ similarity, so as to faithfully estimate feature attribution for LLMs. We have carried out extensive experiments on top of Llama families and their fine-tuned derivatives across various scales to validate the effectiveness of GiLOT for estimating the input attributions. The results show that GiLOT outperforms existing solutions on a number of faithfulness metrics under fair comparison settings. Source code is publicly available at https://github.com/holyseven/GiLOT.
APA
Li, X., Chen, J., Chai, Y. & Xiong, H.. (2024). GiLOT: Interpreting Generative Language Models via Optimal Transport. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:27515-27530 Available from https://proceedings.mlr.press/v235/li24i.html.

Related Material