gigalens 1.2.1

This is a package for Gradient Informed, GPU Accelerated Lens modelling (GIGALens).

Gradient Informed, GPU Accelerated Lens modelling (GIGA-Lens) is a package for fast Bayesian inference on strong gravitational lenses, with support for multi-device and multi-node GPU acceleration for large-scale inference workloads. For details, please see our paper. See here for our documentation.

Usage

Note: Some .ipynb files cannot be run directly in the GitHub preview and must be downloaded to be opened in Jupyter Notebook or JupyterLab.

Installation

GIGA-Lens can be installed via pip:

pip install gigalens[cuda]

If pip notes an error after installation about conflicting dependencies, these can usually be safely ignored. 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, you can choose the kernel tensorflow-2.6.0, and include in the first cell: !pip install gigalens[cuda].

Requirements

Python Version >= 3.12

The following packages are requirements for GIGA-Lens. However, !pip install gigalens[cuda] is all you need to do. In fact, separately installing other packages can cause issues with subpackage dependencies. Some users may find it necessary to install PyYAML.

• 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==2.1.3

• tqdm>=4.67.1,<5.0.0

Authors

GIGALens was written by Andi Gu.