Thought calibration: Efficient and confident test-time scaling
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909621970534400 |
|---|---|
| author | Wu, Menghua Zhou, Cai Bates, Stephen Jaakkola, Tommi |
| author_facet | Wu, Menghua Zhou, Cai Bates, Stephen Jaakkola, Tommi |
| contents | Reasoning large language models achieve impressive test-time scaling by thinking for longer, but this performance gain comes at significant compute cost. Directly limiting test-time budget hurts overall performance, but not all problems are equally difficult. We propose thought calibration to decide dynamically when thinking can be terminated. To calibrate our decision rule, we view a language model's growing body of thoughts as a nested sequence of reasoning trees, where the goal is to identify the point at which novel reasoning plateaus. We realize this framework through lightweight probes that operate on top of the language model's hidden representations, which are informative of both the reasoning structure and overall consistency of response. Based on three reasoning language models and four datasets, thought calibration preserves model performance with up to a 60% reduction in thinking tokens on in-distribution data, and up to 20% in out-of-distribution data. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_18404 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Thought calibration: Efficient and confident test-time scaling Wu, Menghua Zhou, Cai Bates, Stephen Jaakkola, Tommi Machine Learning Artificial Intelligence Reasoning large language models achieve impressive test-time scaling by thinking for longer, but this performance gain comes at significant compute cost. Directly limiting test-time budget hurts overall performance, but not all problems are equally difficult. We propose thought calibration to decide dynamically when thinking can be terminated. To calibrate our decision rule, we view a language model's growing body of thoughts as a nested sequence of reasoning trees, where the goal is to identify the point at which novel reasoning plateaus. We realize this framework through lightweight probes that operate on top of the language model's hidden representations, which are informative of both the reasoning structure and overall consistency of response. Based on three reasoning language models and four datasets, thought calibration preserves model performance with up to a 60% reduction in thinking tokens on in-distribution data, and up to 20% in out-of-distribution data. |
| title | Thought calibration: Efficient and confident test-time scaling |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2505.18404 |