gigalens 2.0

Installation

GIGA-Lens can be installed via pip:

pip install gigalens[cuda]

the cuda extension is recommended to ensure If you wish to test the installation, tests can be run simply by running tox in the root directory.

If you don’t have access to institutional GPUs, one easy way is to use GPU on Google Colab. Please remember the very first cell should have !pip install gigalens[cuda].

If you do have access to institutional GPUs, you can set up a notebook to run on GPU. For example, at NESRC, for running GIGA-Lens on a single GPU node, you can choose the kernel tensorflow-2.6.0, and include in the first cell: !pip install gigalens[cuda].

Demo Notebooks

The notebooks below demonstrate some examples of lens modeling with JAX and Tensorflow respectively. For the quickstart notebooks, you can run them directly using Google Colab.

• JAX quickstart on Google Colab

• TensorFlow quickstart on Google Colab

• Shapelets demo to model the source galaxy (Birrer+ 2015) on JAX

Requirements

Python Version >= 3.12

The following packages are requirements for GIGA-Lens. However, we recommend using pip to avoid issues with subpackage dependencies.

• jax==0.6.2

• tensorflow-probability==0.25.0

• lenstronomy>=1.13.2,<2.0.0

• optax>=0.2.6,<0.3.0

• objax>=1.8.0,<2.0.0

• numpy>=2.0.2

• tqdm>=4.67.1,<5.0.0

Authors

GIGALens was written by Andi Gu in 2021, and is developed by:

• Chase Graham

• Xiaosheng Huang

• Joon Kim

• Harry Lu

• Evan Odell

• Ansel Parke

• Linus Upson

• Felipe Urcelay

• Sean Xu

• Elden Yap

Cite:

@misc{huang2026gigalens20stronglensmodeling,
    title={GIGA-Lens 2.0: Strong-Lens Modeling on Multiple GPU Nodes},
    author={Xiaosheng Huang and Linus Upson and Nicolas Ratier-Werbin and Harry Lu and Sean Xu and Elden Yap and Evan Odell and Ansel Parke and Harsh Ambardekar and Saul Baltasar and Nestor Demeure and Bradley Richardson and Andi Gu and Yuan-Ming Hsu and Junyi Liu},
    year={2026},
    eprint={2606.30633},
    archivePrefix={arXiv},
    primaryClass={astro-ph.CO},
    url={https://arxiv.org/abs/2606.30633},
}