Parcae-Large-770M

Parcae is a novel stable, looped architecture that addresses instability issues in prior looped language models by recasting looping as a nonlinear time-variant dynamical system over the residual stream. This specific checkpoint is the 770M parameter version, trained on the FineWeb-Edu dataset.

• Paper: Parcae: Scaling Laws For Stable Looped Language Models
• Repository: https://github.com/sandyresearch/parcae
• Project Page: https://sandyresearch.github.io/parcae/

Installation

To use this model, you can install the official package:

pip install parcae-lm

Usage

You can load the pretrained weights and instantiate the model using the following code:

import parcae_lm

# Load the pretrained model from HuggingFace
model = parcae_lm.from_pretrained("SandyResearch/parcae-large-770m")

Model Details

This model variant has the following configuration:

Citation

@misc{prairie2026parcaescalinglawsstable,
      title={Parcae: Scaling Laws For Stable Looped Language Models}, 
      author={Hayden Prairie and Zachary Novack and Taylor Berg-Kirkpatrick and Daniel Y. Fu},
      year={2026},
      eprint={2604.12946},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2604.12946}, 
}