RAGEN-2: Reasoning Collapse in Agentic RL
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866908945362190336 |
|---|---|
| author | Wang, Zihan Gui, Chi Jin, Xing Wang, Qineng Liu, Licheng Wang, Kangrui Chen, Shiqi Li, Linjie Yang, Zhengyuan Zhang, Pingyue Lu, Yiping Wu, Jiajun Fei-Fei, Li Wang, Lijuan Choi, Yejin Li, Manling |
| author_facet | Wang, Zihan Gui, Chi Jin, Xing Wang, Qineng Liu, Licheng Wang, Kangrui Chen, Shiqi Li, Linjie Yang, Zhengyuan Zhang, Pingyue Lu, Yiping Wu, Jiajun Fei-Fei, Li Wang, Lijuan Choi, Yejin Li, Manling |
| contents | RL training of multi-turn LLM agents is inherently unstable, and reasoning quality directly determines task performance. Entropy is widely used to track reasoning stability. However, entropy only measures diversity within the same input, and cannot tell whether reasoning actually responds to different inputs. In RAGEN-2, we find that even with stable entropy, models can rely on fixed templates that look diverse but are input-agnostic. We call this template collapse, a failure mode invisible to entropy and all existing metrics. To diagnose this failure, we decompose reasoning quality into within-input diversity (Entropy) and cross-input distinguishability (Mutual Information, MI), and introduce a family of mutual information proxies for online diagnosis. Across diverse tasks, mutual information correlates with final performance much more strongly than entropy, making it a more reliable proxy for reasoning quality. We further explain template collapse with a signal-to-noise ratio (SNR) mechanism. Low reward variance weakens task gradients, letting regularization terms dominate and erase cross-input reasoning differences. To address this, we propose SNR-Aware Filtering to select high-signal prompts per iteration using reward variance as a lightweight proxy. Across planning, math reasoning, web navigation, and code execution, the method consistently improves both input dependence and task performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_06268 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | RAGEN-2: Reasoning Collapse in Agentic RL Wang, Zihan Gui, Chi Jin, Xing Wang, Qineng Liu, Licheng Wang, Kangrui Chen, Shiqi Li, Linjie Yang, Zhengyuan Zhang, Pingyue Lu, Yiping Wu, Jiajun Fei-Fei, Li Wang, Lijuan Choi, Yejin Li, Manling Machine Learning RL training of multi-turn LLM agents is inherently unstable, and reasoning quality directly determines task performance. Entropy is widely used to track reasoning stability. However, entropy only measures diversity within the same input, and cannot tell whether reasoning actually responds to different inputs. In RAGEN-2, we find that even with stable entropy, models can rely on fixed templates that look diverse but are input-agnostic. We call this template collapse, a failure mode invisible to entropy and all existing metrics. To diagnose this failure, we decompose reasoning quality into within-input diversity (Entropy) and cross-input distinguishability (Mutual Information, MI), and introduce a family of mutual information proxies for online diagnosis. Across diverse tasks, mutual information correlates with final performance much more strongly than entropy, making it a more reliable proxy for reasoning quality. We further explain template collapse with a signal-to-noise ratio (SNR) mechanism. Low reward variance weakens task gradients, letting regularization terms dominate and erase cross-input reasoning differences. To address this, we propose SNR-Aware Filtering to select high-signal prompts per iteration using reward variance as a lightweight proxy. Across planning, math reasoning, web navigation, and code execution, the method consistently improves both input dependence and task performance. |
| title | RAGEN-2: Reasoning Collapse in Agentic RL |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2604.06268 |