Saturn-1.5B & Saturn-7B

Model Overview

Saturn-1.5B and Saturn-7B are part of the SATURN framework, which utilizes Boolean Satisfiability (SAT) problems to continuously improve language model reasoning through a curriculum learning pipeline.

Model Details

• Saturn-1.5B: A 1.5 billion parameter model.
• Saturn-7B: A 7 billion parameter model.

Usage

To use these models, simply load them from Hugging Face’s transformers library, as shown below:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "gtxygyzb/Saturn-7B"  # or "gtxygyzb/Saturn-1.5B"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

Further Information

For more detailed information, please refer to:

• Our GitHub repository
• The research paper

Citation

If you use Saturn-1.5B or Saturn-7B in your research, please cite our work:

@article{saturn2025,
  author       = {Huanyu Liu and Jia Li and Hao Zhu and Kechi Zhang and Yihong Dong and Ge Li},
  title        = {SATURN: SAT-based Reinforcement Learning to Unleash Language Model Reasoning},
  journal      = {CoRR},
  volume       = {abs/2505.16368},
  year         = {2025},
}