Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.26219 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915793585831936 |
|---|---|
| author | Kanai, Sekitoshi Yoshida, Tsukasa Takahashi, Hiroshi Kuroki, Haru Hashimoto, Kazumune |
| author_facet | Kanai, Sekitoshi Yoshida, Tsukasa Takahashi, Hiroshi Kuroki, Haru Hashimoto, Kazumune |
| contents | Test-time alignment of large language models (LLMs) attracts attention because fine-tuning LLMs requires high computational costs. In this paper, we propose a new test-time alignment method called adaptive importance sampling on pre-logits (AISP) on the basis of the sampling-based model predictive control with the stochastic control input. AISP applies the Gaussian perturbation into pre-logits, which are outputs of the penultimate layer, so as to maximize expected rewards with respect to the mean of the perturbation. We demonstrate that the optimal mean is obtained by importance sampling with sampled rewards. AISP outperforms best-of-n sampling in terms of rewards over the number of used samples and achieves higher rewards than other reward-based test-time alignment methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_26219 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Test-Time Alignment of LLMs via Sampling-Based Optimal Control in pre-logit space Kanai, Sekitoshi Yoshida, Tsukasa Takahashi, Hiroshi Kuroki, Haru Hashimoto, Kazumune Machine Learning Artificial Intelligence Test-time alignment of large language models (LLMs) attracts attention because fine-tuning LLMs requires high computational costs. In this paper, we propose a new test-time alignment method called adaptive importance sampling on pre-logits (AISP) on the basis of the sampling-based model predictive control with the stochastic control input. AISP applies the Gaussian perturbation into pre-logits, which are outputs of the penultimate layer, so as to maximize expected rewards with respect to the mean of the perturbation. We demonstrate that the optimal mean is obtained by importance sampling with sampled rewards. AISP outperforms best-of-n sampling in terms of rewards over the number of used samples and achieves higher rewards than other reward-based test-time alignment methods. |
| title | Test-Time Alignment of LLMs via Sampling-Based Optimal Control in pre-logit space |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2510.26219 |