polymathic-ai/aion-base

AION-1: Astronomical Omnimodal Network

AION-base is a 300M parameter large omnimodal model specifically designed for astronomical surveys. It integrates 39 distinct astronomical data types and enables adaptation to a wide range of astronomical tasks through multimodal masked modeling.

Model Details

• Architecture: Encoder-Decoder Transformer (12 blocks each, 768 dim, 12 heads)
• Parameters: 300M
• Training: Multimodal Masked Modeling (4M) on astronomical survey data
• Modalities: 39 data types including imaging, spectra, catalogs, and photometry

Quick Start

from aion import AION

# Load the pretrained model
model = AION.from_pretrained('polymathic-ai/aion-base')

# Your astronomical analysis begins here!

Supported Data Types

AION-Base processes data from major astronomical surveys:

• Imaging: Legacy Survey, HSC Wide
• Spectra: SDSS, DESI
• Catalog: Legacy Survey entries
• Gaia: BP/RP spectra, parallax, coordinates, photometry
• Photometry: Legacy Survey (g,r,i,z + WISE), HSC (g,r,i,z,y)
• Shape: Ellipticity and morphological parameters

Installation

pip install aion

For development:

pip install aion[torch,dev]

Resources

• GitHub: https://github.com/PolymathicAI/AION
• Documentation: https://polymathic-ai.github.io/AION/
• Tutorial: https://colab.research.google.com/github/PolymathicAI/AION/blob/main/notebooks/Tutorial.ipynb

Model Variants

Variant	Parameters	Encoder/Decoder	Model Dim
Base	300M	12/12 blocks	768
Large	800M	24/24 blocks	1024
XLarge	3B	24/24 blocks	2048

License

MIT License - see https://github.com/PolymathicAI/AION/blob/main/LICENSE for details.

Citation

@article{aion2025, title={AION: Astronomical Omnimodal Network}, author={...}, journal={arXiv preprint arXiv:XXXX.XXXXX}, year={2025} }

---

Built with ❤ for the astronomical community by https://polymathic-ai.org/