| --- |
| license: mit |
| datasets: |
| - agentica-org/DeepScaleR-Preview-Dataset |
| base_model: |
| - deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B |
| tags: |
| - LRM |
| - hybrid_reasoning |
| - efficient_reasoning |
| --- |
| |
| # AdaptThink: LLM Can Learn When to Think |
|
|
| <p align="center"> |
| 🤗 <a href="https://huggingface.co/collections/THU-KEG/adaptthink-682a1059aa9f5102c4fa0470" target="_blank">HF Collections</a> • 💻 <a href="" target="_blank">Github Repo</a> • 📃 <a href="https://arxiv.org/abs/2505.13417" target="_blank">Paper</a> |
| </p> |
| |
| ## 🔍 Table of Contents |
| - [🤖️ AdaptThink](#adapt_think) |
| - [⚙️ Released Models](#model) |
| - [📊 Evaluation](#evaluation) |
| - [📝 Citation](#citation) |
|
|
| <a name="adapt_think"></a> |
| ## 🤖️ AdaptThink |
| We present **AdapThink**, a novel reinforcement learning (RL) algorithm that enables reasoning models to adaptively choose between **Thinking** and **NoThinking** modes according to the difficulty of each input problem, thereby achieving automatic hybrid reasoning. Specifically, the model engages in thinking only when the problem is determined to be challenging; for other simple question, it will bypass the thinking process and directly produce a concise final solution. This approach substantially reduces inference costs while further improving overall performance. |
|
|
|  |
|
|
|
|
|
|
| <a name="model"></a> |
| ## ⚙️ Released Models |
|
|
| ### All Available Datasets and Models |
| We apply the AdaptThink algorithm on DeepSeek-R1-Distill-Qwen-1.5B with $\delta$ from 0 to 0.1, and DeepSeek-R1-Distill-Qwen-7B with $\delta=0.05$. A larger $\large$ results in a higher proportion of NoThinking responses, which reduces more inference costs but also diminish the resultant improvement in accuracy. |
|
|
| All the trained models are available on HuggingFace. |
|
|
|
|
| | Name | HF Repo | |
| |---|---| |
| | AdaptThink-1.5B-delta0 | [🤗 HF Repo](https://huggingface.co/THU-KEG/AdaptThink-1.5B-delta0) | |
| | AdaptThink-1.5B-delta0.01 | [🤗 HF Repo](https://huggingface.co/THU-KEG/AdaptThink-1.5B-delta0.01) | |
| | AdaptThink-1.5B-delta0.02 | [🤗 HF Repo](https://huggingface.co/THU-KEG/AdaptThink-1.5B-delta0.02) | |
| | AdaptThink-1.5B-delta0.05 | [🤗 HF Repo](https://huggingface.co/THU-KEG/AdaptThink-1.5B-delta0.05) | |
| | AdaptThink-1.5B-delta0.075 | [🤗 HF Repo](https://huggingface.co/THU-KEG/AdaptThink-1.5B-delta0.075) | |
| | AdaptThink-1.5B-delta0.1 | [🤗 HF Repo](https://huggingface.co/THU-KEG/AdaptThink-1.5B-delta0.1) | |
| | AdaptThink-7B-delta0.05 | [🤗 HF Repo](https://huggingface.co/THU-KEG/AdaptThink-7B-delta0.05) | |
|
|
| <a name="training"></a> |
|
|
| ## 📊 Evaluation Results |
|
|
| We list our evaluation results as follows: |
| ##### 1. Comparison with existing methods for efficient reasoning on mathematics datasets |
|
|
|  |
|
|
| ##### 2. Nothinking responses ratio and accuracy across different difficulty levels on MATH500 |
|
|
|  |
|
|
| ##### 3. Comparison of different $\delta$ values |
|
|
|  |
|
|
| ##### 4. Evaluation results on MMLU |
|
|
| <img width="1000" alt="image" src="https://cdn-uploads.huggingface.co/production/uploads/66cdd285c51a915bd5f2d017/19K2u6PNmYz3gx3JnHgn4.png"> |
|
|
| <a name="citation"></a> |
| ## 📝 Citation |
|
|
| If you find our work useful, please consider citing LongReward: |
|
|
| ``` |
| @article{zhang2025adapt_think, |
| title = {AdaptThink: LLM Can Learn When to Think} |
| author={Jiajie Zhang and Nianyi Lin and Lei Hou and Ling Feng and Juanzi Li}, |
| journal={arXiv preprint arXiv: 2505.13417}, |
| url={https://arxiv.org/abs/2505.13417} |
| year={2025} |
| } |
| ``` |
|
|
|
|
