nielsr's picture
nielsr HF Staff
Add pipeline_tag, library_name and sample usage
841feac verified
|
raw
history blame
2.99 kB
metadata
base_model:
  - meta-llama/Llama-3.2-1B-Instruct
license: mit
pipeline_tag: text-generation
library_name: transformers

Sequential-Light-Critic-Llama3.2-1B

๐ŸŒ Project Page | ๐Ÿ’ป Code | ๐Ÿ“„ 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 Sequential-Light setting, the Critic Agent is responsible for reviewing and refining the initial plan produced by the Planner Agent, which is then passed to the Solver Agent for final solution generation.

Model Details

Item Description
Model Sequential-Light-Critic-Llama3.2-1B
Collaboration Style Sequential-Light
Agent Role Critic Agent
Base Model Llama3.2-1B

โš ๏ธ Note: This checkpoint is a role-specific agent in RecursiveMAS, rather than a standalone model intended for plain-text generation.

Sample Usage

You can load the entire MAS pipeline using the code provided in the GitHub repository:

from system_loader import load_mas_system

mas = load_mas_system(
    style="sequential_light",
    device="cuda",
    trust_remote_code=True,
)

planner = mas.agents["planner"].model
critic = mas.agents["critic"].model
solver = mas.agents["solver"].model

For detailed usage instructions, please refer to the GitHub repository.

Model Collections for RecursiveMAS

Style Model Collection
Sequential-Style ๐Ÿค— HuggingFace
Mixture-Style ๐Ÿค— HuggingFace
Distillation-Style ๐Ÿค— HuggingFace
Deliberation-Style ๐Ÿค— HuggingFace

Experiment Results

RecursiveMAS 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 front 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}, 
}