Efficient Reasoning via Thought-Training and Thought-Free Inference
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_ | 1866911291468152832 |
|---|---|
| author | Wu, Canhui Cao, Qiong Xue, Chao Xi, Wei He, Xiaodong |
| author_facet | Wu, Canhui Cao, Qiong Xue, Chao Xi, Wei He, Xiaodong |
| contents | Recent advances in large language models (LLMs) have leveraged explicit Chain-of-Thought (CoT) prompting to improve reasoning accuracy. However, most existing methods primarily focus on compressing verbose reasoning outputs. These Long-to-Short transformations aim to improve efficiency, but require a large amount of short CoT data. In this work, we introduce \textbf{3TF} (\textbf{T}hought-\textbf{T}raining and \textbf{T}hought-\textbf{F}ree inference), a framework for efficient reasoning that takes a Short-to-Long perspective. We first train a hybrid model that can operate in both reasoning and non-reasoning modes, and then further train it on CoT-annotated data to internalize structured reasoning, while enforcing concise, thought-free outputs at inference time using the no-reasoning mode. Unlike compression-based approaches, 3TF improves the reasoning quality of non-reasoning outputs, enabling models to perform rich internal reasoning implicitly while keeping external outputs short. Empirically, 3TF-trained models obtain large improvements on reasoning benchmarks under thought-free inference, demonstrating that high quality reasoning can be learned and executed implicitly without explicit step-by-step generation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_03408 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Efficient Reasoning via Thought-Training and Thought-Free Inference Wu, Canhui Cao, Qiong Xue, Chao Xi, Wei He, Xiaodong Computation and Language I.2.7 Recent advances in large language models (LLMs) have leveraged explicit Chain-of-Thought (CoT) prompting to improve reasoning accuracy. However, most existing methods primarily focus on compressing verbose reasoning outputs. These Long-to-Short transformations aim to improve efficiency, but require a large amount of short CoT data. In this work, we introduce \textbf{3TF} (\textbf{T}hought-\textbf{T}raining and \textbf{T}hought-\textbf{F}ree inference), a framework for efficient reasoning that takes a Short-to-Long perspective. We first train a hybrid model that can operate in both reasoning and non-reasoning modes, and then further train it on CoT-annotated data to internalize structured reasoning, while enforcing concise, thought-free outputs at inference time using the no-reasoning mode. Unlike compression-based approaches, 3TF improves the reasoning quality of non-reasoning outputs, enabling models to perform rich internal reasoning implicitly while keeping external outputs short. Empirically, 3TF-trained models obtain large improvements on reasoning benchmarks under thought-free inference, demonstrating that high quality reasoning can be learned and executed implicitly without explicit step-by-step generation. |
| title | Efficient Reasoning via Thought-Training and Thought-Free Inference |
| topic | Computation and Language I.2.7 |
| url | https://arxiv.org/abs/2511.03408 |