---
license: apache-2.0
datasets:
- DreamW1ngs/AdaR-orca-math-9k
language:
- en
metrics:
- accuracy
base_model:
- Qwen/Qwen2.5-Math-7B
pipeline_tag: question-answering
---

# Making Mathematical Reasoning Adaptive

<p align="center">
  <a href="https://arxiv.org/abs/2510.04617"> 📃 Paper</a> | 
  <a href="https://github.com/NJUNLP/AdaR"> ⚙️ Code</a> | 
  <a href="https://huggingface.co/collections/DreamW1ngs/adar-68e648e59b2c9aec1208b5ef"> 🤖 Project</a> | 
  <a href="https://resume.laizj.fun/"> 📭 Contact</a> 
</p>

---

## 🌱 Overview

Large Language Models (LLMs) have shown impressive reasoning capabilities, yet they often rely on **spurious reasoning** — producing answers from superficial features, leading to failure at robustness and generalization.

We propose **AdaR** framework to enable adaptive reasoning, wherein models rely on problem-solving logic to produce answers. **AdaR** synthesizes logically equivalent queries by varying variable values, and trains models with RLVR on these data to penalize spurious logic while encouraging adaptive logic.

The framework integrates *data synthesis* and *RLVR training* to enhance both **robustness (in-domain)** and **generalization (out-of-domain)**.

![AdaR Process Framework](./figs/process.png)

> **Figure 1.**  
> *Subfigure I:* Three reasoning modes — direct inference (black), spurious reasoning (red), adaptive reasoning (green).  
> *Subfigure II:* Logic-preserving variable perturbation and gold-answer generation via executable logic.  
> *Subfigure III:* RLVR optimization encouraging adaptive reasoning through comparative feedback.

## 📈 Highlights

- 🚀 **+8.5 Average Improvement** across in-domain robustness tasks and out-of-domain tasks.
- 🧮 **Only 9K synthetic data** needed for significant gains.
- ⚖️ **Enable algebraic thinking** and improved stability under scaling.
- 🔁 **Generalizable framework** applicable to instruct models.

---