Skip to content

feat: JAX default install, [jax] alias + install docs (#702) #703

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Jammy2211 merged 2 commits into from
Aug 19, 2026
Merged

feat: JAX default install, [jax] alias + install docs (#702) #703

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
59c8a50
feat: JAX default install, [jax] alias + install docs (#702)
Aug 19, 2026
94d8f54
test: mark solve_joint_optimization test requires_jax — caught by the…
Aug 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 15 additions & 12 deletions docs/installation/conda.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,20 +8,21 @@

This acceleration is achieved through \[**JAX**\](<https://docs.jax.dev/en/latest/notebooks/thinking_in_jax.html>), which provides GPU and TPU support.

**JAX is not installed by default.** To install **PyAutoLens** with JAX, use the `jax` extra:
**JAX is installed by default** — a plain `pip install autolens` includes it (the older
`pip install autolens[jax]` command still works and installs the same thing).

```bash
pip install autolens[jax] --no-cache-dir
```

A plain `pip install autolens` gives a fully working install that runs on NumPy, but without any of the JAX
acceleration described above.

The `[jax]` extra installs **CPU-only** JAX. To ensure GPU acceleration, it is recommended that you install JAX with
The default install includes **CPU-only** JAX. To ensure GPU acceleration, it is recommended that you install JAX with
GPU support **before** installing **PyAutoLens**, by following the official \[JAX installation guide\](<https://jax.readthedocs.io/en/latest/installation.html>).

If you install **PyAutoLens** without a proper GPU setup, a warning will be displayed.

:::{note}
**Intel Macs**: JAX no longer publishes wheels for Intel (x86_64) macOS, so on these machines
`pip install autolens` automatically installs without JAX and runs on the slower NumPy path — a
warning is printed at import to make this clear. Every other supported platform (Windows, Linux,
Apple-silicon Macs) gets JAX by default.
:::

## Install

Installation via a conda environment circumvents compatibility issues when installing certain libraries. This guide
Expand Down Expand Up @@ -57,11 +58,13 @@ The latest version of **PyAutoLens** is installed via pip as follows (the comman
caching issues impacting the installation):

```bash
pip install autolens[jax] --no-cache-dir
pip install autolens --no-cache-dir
```

The `[jax]` extra is recommended, as it enables the JAX acceleration described above. To install without JAX,
use `pip install autolens --no-cache-dir` instead.
This includes JAX by default, enabling the acceleration described above. If you need an install without
JAX on a platform where JAX wheels exist (e.g. a restricted environment), install normally and then run
`pip uninstall jax jaxlib` — **PyAutoLens** detects the absence at import and falls back to the fully
supported (but much slower) NumPy path.

If pip prints warnings about dependency version conflicts, these can usually be ignored — the instructions below
will identify clearly if the installation is a success.
Expand Down
33 changes: 18 additions & 15 deletions docs/installation/pip.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,20 +14,21 @@ distribution" error. Upgrade Python to 3.12+ before installing.

This acceleration is achieved through \[**JAX**\](<https://docs.jax.dev/en/latest/notebooks/thinking_in_jax.html>), which provides GPU and TPU support.

**JAX is not installed by default.** To install **PyAutoLens** with JAX, use the `jax` extra:
**JAX is installed by default** — a plain `pip install autolens` includes it (the older
`pip install autolens[jax]` command still works and installs the same thing).

```bash
pip install autolens[jax]
```

A plain `pip install autolens` gives a fully working install that runs on NumPy, but without any of the JAX
acceleration described above.

The `[jax]` extra installs **CPU-only** JAX. To ensure GPU acceleration, it is recommended that you install JAX with
The default install includes **CPU-only** JAX. To ensure GPU acceleration, it is recommended that you install JAX with
GPU support **before** installing **PyAutoLens**, by following the official \[JAX installation guide\](<https://jax.readthedocs.io/en/latest/installation.html>).

If you install **PyAutoLens** without a proper GPU setup, a warning will be displayed.

:::{note}
**Intel Macs**: JAX no longer publishes wheels for Intel (x86_64) macOS, so on these machines
`pip install autolens` automatically installs without JAX and runs on the slower NumPy path — a
warning is printed at import to make this clear. Every other supported platform (Windows, Linux,
Apple-silicon Macs) gets JAX by default.
:::

## Install

We strongly recommend that you install **PyAutoLens** in a
Expand All @@ -40,20 +41,22 @@ We upgrade pip to ensure certain libraries install:
pip install --upgrade pip
```

The latest version of **PyAutoLens** is installed via pip as follows (specifying the version as shown below ensures
the installation has clean dependencies):
The latest version of **PyAutoLens** is installed via pip as follows:

```bash
pip install autolens[jax]
pip install autolens
```

The `[jax]` extra is recommended, as it enables the JAX acceleration described above. To install without JAX,
omit the extra:
This includes JAX by default, enabling the acceleration described above. If you need an install without
JAX on a platform where JAX wheels exist (e.g. a restricted environment), install normally and then remove it:

```bash
pip install autolens
pip uninstall jax jaxlib
```

**PyAutoLens** detects the absence at import and falls back to the fully supported (but much slower)
NumPy path.

If pip prints warnings about dependency version conflicts, these can usually be ignored — the instructions below
will identify clearly if the installation is a success.

Expand Down
14 changes: 7 additions & 7 deletions pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,12 +24,10 @@ classifiers = [
]
keywords = ["cli"]
dependencies = [
# Floor, not a pin. Without one, pip backtracking the extras chain
# (autolens[jax] -> autogalaxy[jax] -> autofit[jax] -> autonerves[jax]) may
# walk the release history to 2022: "version X does not provide the extra
# 'jax'" is a pip *warning*, not an error, so a pre-extras release is a
# legal solution. That is how `autolens[optional]` came to install autofit
# 2026.4.30.582 and fail on `af.Latent` (#687).
# Floor, not a pin (#687 — a floorless backtrack once installed autofit
# 2026.4.30.582 and failed on `af.Latent`). Bump to the first release with
# JAX in the family's base dependencies once it exists (#702), so
# backtracking cannot pair this autolens with a jax-optional chain.
"autogalaxy>=2026.7.29.2",
"nautilus-sampler==1.0.5"
]
Expand All @@ -49,7 +47,9 @@ local_scheme = "no-local-version"


[project.optional-dependencies]
jax = ["autogalaxy[jax]>=2026.7.29.2"]
# JAX moved into the base dependencies (#702). Kept as a declared no-op so
# `pip install autolens[jax]` keeps resolving (#687).
jax = []
coolest = ["coolest"]
optional = [
"autolens[jax]",
Expand Down
1 change: 1 addition & 0 deletions test_autolens/potential_correction/test_iterative_interferometer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,7 @@ def test__solve_joint_optimization__finite_state_and_decreasing_cost(
assert np.isfinite(dpsi_opt).all()


@requires_jax
def test__solve_joint_optimization__identity_damping_finite(interferometer_7):
dataset = interferometer_7.apply_sparse_operator()
fit = iter_fit_from(dataset, damping="identity", max_consecutive_rejections=3)
Expand Down
Loading