xaig
Light, framework-agnostic tooling for E3SM AI campaigns: what an emulator holds inside, sparse autoencoders trained on it, figures, and a local app over them.
Research tool.
xaigis early. What is described here works; expect it to change.
Install
The base install pulls only Click. Anything heavier sits behind an extra named after the subpackage that needs it:
$ uv pip install 'xaig[latents]' # or: pip install 'xaig[latents]'
| Extra | Pulls | Gets you |
|---|---|---|
latents |
numpy, xarray, netCDF4 | xaig.latents: latent diagnostics on a grid |
figures |
matplotlib, cartopy | xaig.figures: maps and series figures (brings latents) |
app |
streamlit | xaig app: a local web app (brings figures) |
nn |
torch | xaig.nn and xaig nn: sparse autoencoders (brings latents) |
hf |
huggingface_hub | archives read from a Hugging Face repository, hf://datasets/... (brings latents) |
A missing extra says so, with the command that fits how xaig was installed.
Try it, with no model and no data
A toy emulator writes a latent archive with nothing but numpy; everything else reads it like any other:
$ xaig latents toy scratch/toy/control
$ xaig latents info scratch/toy/control --mask-variable sst
$ xaig latents region scratch/toy/control --lat 10 --lon -114 --time 2 --centered --pcs 2
$ xaig latents fields scratch/toy/control --field precipitation --top 3
$ xaig nn sae scratch/toy/control --features 64 --k 4 --out scratch/toy/sae.npz # needs xaig[nn]
$ xaig app --latents scratch/toy # needs xaig[app]
The CLI is a thin client of the Python API; anything it can do, a notebook can:
from xaig.latents import Region, analyze_region, open_source
source = open_source("scratch/toy/control", mask_variable="sst")
result = analyze_region(
source, time=2, layer=3, region=Region(lat=10, lon=-114, radius_km=1500), centered=True
)
result.ranking.channels # the channels that respond most strongly there
More
- Guides: https://e3sm-project.github.io/aigroup/package/
- Source and issues: https://github.com/E3SM-Project/aigroup
BSD-3-Clause. xaig.latents, xaig.figures and xaig.app grew out of
the latent space visualiser for weather models
(Tempest, Beylich & Craig 2026, arXiv:2604.20467, doi:10.1007/978-3-032-29915-4_10); see
NOTICE. The sparse autoencoders follow MacMillan & Ouellette (2025, arXiv:2512.24440);
the B-spline autoencoder of Cheon (2026, arXiv:2605.17493) is what xaig.nn is
heading for and does not implement yet.
1 maintainer
Metadata
1 maintainer
Release files for xaig 0.5.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| xaig-0.5.0.tar.gz | 198.5 kB | With provenance | Details for xaig-0.5.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| xaig-0.5.0-py3-none-any.whl | Python 3 | none | any | Details for xaig-0.5.0-py3-none-any.whl |
Total release size: 344.0 kB
Release files / xaig-0.5.0.tar.gz
| Download URL | xaig-0.5.0.tar.gz |
|---|---|
| Size | 198.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
027783d0dab5a12fcd1f833ad3ecc36948434f042c768badd7210636110dce33
|
|
BLAKE2b-256 checksum How to use checksums |
659ce4f1f895121042062439b4f5f5113a43cf0aa3c939cbd0065d2f933da91b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 3, 2026.
Transparency logRelease files / xaig-0.5.0-py3-none-any.whl
| Download URL | xaig-0.5.0-py3-none-any.whl |
|---|---|
| Size | 145.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
91513e6284ab09c45407db7ffc2c2f1578bf9c9cc20b693f359457910f98e3e7
|
|
BLAKE2b-256 checksum How to use checksums |
3a78af386d420849b676d739664b8b699087b6df2441fb61b95139f6af9b48e4
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 3, 2026.
Transparency logSecurity for xaig 0.5.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for xaig-0.5.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 3, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
