Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2403.14459 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909700278190080 |
|---|---|
| author | Paes, Lucas Monteiro Wei, Dennis Do, Hyo Jin Strobelt, Hendrik Luss, Ronny Dhurandhar, Amit Nagireddy, Manish Ramamurthy, Karthikeyan Natesan Sattigeri, Prasanna Geyer, Werner Ghosh, Soumya |
| author_facet | Paes, Lucas Monteiro Wei, Dennis Do, Hyo Jin Strobelt, Hendrik Luss, Ronny Dhurandhar, Amit Nagireddy, Manish Ramamurthy, Karthikeyan Natesan Sattigeri, Prasanna Geyer, Werner Ghosh, Soumya |
| contents | Despite the increasing use of large language models (LLMs) for context-grounded tasks like summarization and question-answering, understanding what makes an LLM produce a certain response is challenging. We propose Multi-Level Explanations for Generative Language Models (MExGen), a technique to provide explanations for context-grounded text generation. MExGen assigns scores to parts of the context to quantify their influence on the model's output. It extends attribution methods like LIME and SHAP to LLMs used in context-grounded tasks where (1) inference cost is high, (2) input text is long, and (3) the output is text. We conduct a systematic evaluation, both automated and human, of perturbation-based attribution methods for summarization and question answering. The results show that our framework can provide more faithful explanations of generated output than available alternatives, including LLM self-explanations. We open-source code for MExGen as part of the ICX360 toolkit: https://github$.$com/IBM/ICX360. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_14459 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Multi-Level Explanations for Generative Language Models Paes, Lucas Monteiro Wei, Dennis Do, Hyo Jin Strobelt, Hendrik Luss, Ronny Dhurandhar, Amit Nagireddy, Manish Ramamurthy, Karthikeyan Natesan Sattigeri, Prasanna Geyer, Werner Ghosh, Soumya Computation and Language Artificial Intelligence Despite the increasing use of large language models (LLMs) for context-grounded tasks like summarization and question-answering, understanding what makes an LLM produce a certain response is challenging. We propose Multi-Level Explanations for Generative Language Models (MExGen), a technique to provide explanations for context-grounded text generation. MExGen assigns scores to parts of the context to quantify their influence on the model's output. It extends attribution methods like LIME and SHAP to LLMs used in context-grounded tasks where (1) inference cost is high, (2) input text is long, and (3) the output is text. We conduct a systematic evaluation, both automated and human, of perturbation-based attribution methods for summarization and question answering. The results show that our framework can provide more faithful explanations of generated output than available alternatives, including LLM self-explanations. We open-source code for MExGen as part of the ICX360 toolkit: https://github$.$com/IBM/ICX360. |
| title | Multi-Level Explanations for Generative Language Models |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2403.14459 |