61 lines
2.9 KiB
Markdown
61 lines
2.9 KiB
Markdown
|
---
|
|||
|
|
license: apache-2.0
|
|||
|
|
language:
|
|||
|
|
- en
|
|||
|
|
pipeline_tag: text-generation
|
|||
|
|
library_name: transformers
|
|||
|
|
tags:
|
|||
|
|
- agent
|
|||
|
|
- reasoning
|
|||
|
|
- tool-use
|
|||
|
|
- simulative-planning
|
|||
|
|
base_model: Qwen/Qwen3-8B
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
# SR²AM-v0.1-8B
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
We argue that efficient agentic reasoning benefits from decomposing deliberation into three interacting systems: **reactive execution** (System I) for fine-grained reasoning and direct action; **simulative reasoning** (System II) that predicts consequences of proposed actions through a world model; and **self-regulation** (System III) that decides *when* and *how deeply* to plan through a learned **configurator**.
|
|||
|
|
|
|||
|
|
**SR²AM** (Self-Regulated Simulative Reasoning Agentic LLM) is our instantiation: the configurator and simulative planner are realized as distinct stages within an LLM's chain-of-thought reasoning, with the LLM itself serving as the world model in language space.
|
|||
|
|
|
|||
|
|
SR²AM-v0.1-8B achieves an overall Pass@1 of **57.0** across 11 benchmarks spanning math, science, tabular analysis, and web information seeking — competitive with systems at 120–355B parameters.
|
|||
|
|
|
|||
|
|
More details: [project website](https://sailing-lab.github.io/sr2am-self-regulated-planning) | [paper](https://arxiv.org/abs/2605.22138) | [GitHub](https://github.com/sailing-lab/sr2am).
|
|||
|
|
|
|||
|
|
|
|||
|
|
## Key Features
|
|||
|
|
|
|||
|
|
- **System I + II + III decomposition**: a configurator (System III) decides per-turn whether to plan, continue an existing plan, or act directly; a simulative planner (System II) constructs plans grounded in predicted future states; reactive execution (System I) handles fine-grained reasoning and tool use.
|
|||
|
|
- **SFT + RL training**: supervised learning on data encoding the self-regulated planning structure, followed by reinforcement learning (GRPO) for task success.
|
|||
|
|
- **Agentic tool use**: web search (SerpAPI), web browsing with LLM summarization, and stateless Python code execution (SandboxFusion).
|
|||
|
|
- **Compact and efficient**: 3,698 reasoning tokens per trajectory on average — fewer or comparable to other systems at the same scale while outperforming them in Pass@1.
|
|||
|
|
|
|||
|
|
## Quick Start
|
|||
|
|
|
|||
|
|
See the [GitHub repository](https://github.com/sailing-lab/sr2am) for setup and inference instructions.
|
|||
|
|
|
|||
|
|
## Main Results
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
SR²AM-v0.1-8B sits above the size-vs-accuracy trendline in (a). The full benchmark breakdown is in the [paper](https://arxiv.org/abs/2605.22138).
|
|||
|
|
|
|||
|
|
## Citation
|
|||
|
|
|
|||
|
|
```bibtex
|
|||
|
|
@article{deng2026sr2am,
|
|||
|
|
title={Efficient Agentic Reasoning Through Self-Regulated Simulative Planning},
|
|||
|
|
author={Deng, Mingkai and Hou, Jinyu and Neves, Lara Sá and
|
|||
|
|
Pimpalkhute, Varad and Killian, Taylor W. and
|
|||
|
|
Liu, Zhengzhong and Xing, Eric P.},
|
|||
|
|
journal={arXiv preprint arXiv:2605.22138},
|
|||
|
|
year={2026}
|
|||
|
|
}
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
## License
|
|||
|
|
|
|||
|
|
Released under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0).
|