base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
license: mit
pipeline_tag: text-generation
library_name: transformers
Mixture-Math-DeepSeek-R1-Distill-Qwen-1.5B
🌐 Project Page | 💻 Code | 📄 Paper
We introduce RecursiveMAS, a multi-agent framework that scales agent collaboration through latent-space recursion. RecursiveMAS treats a multi-agent system as a unified recursive computation, where heterogeneous agents iteratively exchange, refine, and evolve their latent states across recursion rounds.
In the Mixture-Style setting, this Math Specialist Agent focuses on mathematical reasoning tasks and collaborates with other domain-specialized agents through RecursiveLink modules for final response generation.
The model was presented in the paper Recursive Multi-Agent Systems.
Model Details
| Item | Description |
|---|---|
| Model | Mixture-Math-DeepSeek-R1-Distill-Qwen-1.5B |
| Collaboration Style | Mixture-Style |
| Agent Role | Math Specialist Agent |
| Base Model | DeepSeek-R1-Distill-Qwen-1.5B |
⚠️ Note: This checkpoint is a role-specific agent in RecursiveMAS, rather than a standalone model intended for plain-text generation.
Usage
To use this model as part of the RecursiveMAS system, you can load the entire multi-agent pipeline using the provided system loader from the official repository:
# Requires cloning the repository: git clone https://github.com/RecursiveMAS/RecursiveMAS.git
from system_loader import load_mas_system
mas = load_mas_system(
style="mixture",
device="cuda",
trust_remote_code=True,
)
# Access individual agents
math_agent = mas.agents["math"].model
Model Collections for RecursiveMAS
| Style | Model Collection |
|---|---|
| Sequential-Style | 🤗 HuggingFace |
| Mixture-Style | 🤗 HuggingFace |
| Distillation-Style | 🤗 HuggingFace |
| Deliberation-Style | 🤗 HuggingFace |
Experiment Results
Citation
@misc{recursivemas,
title={Recursive Multi-Agent Systems},
author={Xiyuan Yang and Jiaru Zou and Rui Pan and Ruizhong Qiu and Pan Lu and Shizhe Diao and Jindong Jiang and Hanghang Tong and Tong Zhang and Markus J. Buehler and Jingrui He and James Zou},
year={2026},
eprint={2604.25917},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2604.25917},
}