Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2403.18680 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866907928218304512 |
|---|---|
| author | Hoscilowicz, Jakub Wiacek, Adam Chojnacki, Jan Cieslak, Adam Michon, Leszek Urbanevych, Vitalii Janicki, Artur |
| author_facet | Hoscilowicz, Jakub Wiacek, Adam Chojnacki, Jan Cieslak, Adam Michon, Leszek Urbanevych, Vitalii Janicki, Artur |
| contents | In this work, we explore LLM's internal representation space to identify attention heads that contain the most truthful and accurate information. We further developed the Inference Time Intervention (ITI) framework, which lets bias LLM without the need for fine-tuning. The improvement manifests in introducing a non-linear multi-token probing and multi-token intervention: Non-Linear ITI (NL-ITI), which significantly enhances performance on evaluation benchmarks. NL-ITI is tested on diverse multiple-choice datasets, including TruthfulQA, on which we report over 16% relative MC1 (accuracy of model pointing to the correct answer) improvement with respect to the baseline ITI results. Moreover, we achieved a 10% relative improvement over the recently released Truth Forest (TrFf) method that also focused on ITI improvement. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_18680 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Non-Linear Inference Time Intervention: Improving LLM Truthfulness Hoscilowicz, Jakub Wiacek, Adam Chojnacki, Jan Cieslak, Adam Michon, Leszek Urbanevych, Vitalii Janicki, Artur Computation and Language Machine Learning In this work, we explore LLM's internal representation space to identify attention heads that contain the most truthful and accurate information. We further developed the Inference Time Intervention (ITI) framework, which lets bias LLM without the need for fine-tuning. The improvement manifests in introducing a non-linear multi-token probing and multi-token intervention: Non-Linear ITI (NL-ITI), which significantly enhances performance on evaluation benchmarks. NL-ITI is tested on diverse multiple-choice datasets, including TruthfulQA, on which we report over 16% relative MC1 (accuracy of model pointing to the correct answer) improvement with respect to the baseline ITI results. Moreover, we achieved a 10% relative improvement over the recently released Truth Forest (TrFf) method that also focused on ITI improvement. |
| title | Non-Linear Inference Time Intervention: Improving LLM Truthfulness |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2403.18680 |