README.md

---
license: apache-2.0
datasets:
- agentica-org/DeepScaleR-Preview-Dataset
base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
tags:
- reinforcement-learning
language:
- en
- zh
pipeline_tag: text-generation
library_name: transformers
---

<p align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/64ed568ccf6118a9379a61b8/BHITqJU33sXqf-Jbytrxg.png" width="100"/>
  <b><span style="font-size:28px">SIRI: Scaling Iterative Reinforcement Learning with Interleaved Compression</span></b>
</p>



<p align="center">
  📃 <a href="https://arxiv.org/abs/2509.25176" target="_blank">Paper</a> • 📝 <a href="https://api.wandb.ai/links/teamsiri/isge4elx" target="_blank">Wandb</a>
</p>

---

## 🔍 Overview

**SIRI (Scaling Iterative Reinforcement Learning with Interleaved Compression)** is a reinforcement-learning–based framework designed to improve the efficiency and accuracy of **Large Reasoning Models (LRMs)**.  

Traditional RL training often causes **overthinking** and long, redundant reasoning traces. Prior methods that compress outputs (length penalties, pruning, or skipping thought tokens) improve efficiency but hurt accuracy.  

SIRI solves this trade-off by **iteratively alternating between compression and expansion of the reasoning budget**, controlled by a cosine length scheduler. This approach dynamically balances concise reasoning with long-horizon exploration.

<p align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/64ed568ccf6118a9379a61b8/SXow6xntEgrwhvWtzvrkE.png" alt="pareto_front" width="500"/>
</p>

---

## 🚀 Key Features

- **Interleaved Compression–Expansion**:  
  - *Compression phase*: forces concise, high-density reasoning by limiting rollout length.  
  - *Expansion phase*: restores longer rollouts to encourage exploration and planning.  
- **Token Efficiency without Accuracy Loss**: Unlike previous methods, SIRI improves accuracy *while reducing average token usage*.  
- **Iterative RL Training**: Built on GRPO with modifications from DAPO (clip-high/low decoupling, KL removal).  
- **Generalization Across Model Sizes**: Validated on both **1.5B** and **7B** models.

---

## 📊 Benchmarks


![perf](https://cdn-uploads.huggingface.co/production/uploads/64ed568ccf6118a9379a61b8/0S2d9VZTiaoGI6_N9Vrh2.png)  


---

## 📝 Citation

```bibtex
@misc{wen2025siriscalingiterativereinforcement,
      title={SIRI: Scaling Iterative Reinforcement Learning with Interleaved Compression}, 
      author={Haoming Wen and Yushi Bai and Juanzi Li and Jie Tang},
      year={2025},
      eprint={2509.25176},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2509.25176}, 
}
```