Install#

The quickest route is the install picker, which gives the command for your platform and GPU. This page has the details.

Requirements#

Python

CPython 3.10, 3.11, 3.12, 3.13 and 3.14

macOS

Apple silicon (arm64). One wheel with Apple Metal GPU support and CPU support.

Linux

x86-64. The core wheel plus the mojolearn-nvidia and mojolearn-amd GPU packages, installed automatically.

pip#

pip install mojolearn

On Linux x86-64, pip also installs mojolearn-nvidia and mojolearn-amd, and mojolearn selects the backend for the GPU it finds. Both platforms also train and predict on the CPU (see Devices and the CPU).

Optional extras add NumPy without changing any numerical implementation:

pip install "mojolearn[verify]"   # NumPy for verification fixtures
pip install "mojolearn[numpy]"    # NumPy-backed conveniences (ANN filtering, TSNE, sequence APIs)

A plain pip install mojolearn is NumPy-free; APIs that need NumPy say so and name the install command.

GPU architectures in this release#

Device code is architecture specific. A GPU architecture the release does not carry needs a source build; mojolearn never substitutes the CPU for a GPU it cannot serve.

Vendor

Package

Architectures

Examples

Apple

mojolearn (macOS)

Apple silicon, built at the M1 instruction-set floor

M1 to M4 families

NVIDIA

mojolearn-nvidia

sm_89

L40S, L40, L4, RTX 4090, RTX 6000 Ada

AMD

mojolearn-amd

gfx942 (exact match required)

Instinct MI300X, MI325X

NVIDIA needs only the driver (libcuda); no CUDA toolkit is required. AMD needs the amdgpu driver and the ROCm HIP runtime (libamdhip64, ROCm 6 or 7).

Check your machine#

mojolearn doctor
python -m mojolearn verify --quick

mojolearn doctor reports what the installed wheel supports on this machine. verify checks the installed package against the reference bits recorded for this release; see Verify on your hardware.

From source#

A source build targets the GPU in the build machine unless MOJOLEARN_GPU_ARCHS says otherwise. Building needs the Mojo toolchain, which the repository’s pixi environment provides:

git clone https://github.com/mojolearn/mojolearn
cd mojolearn
pixi run probe        # reports the toolchain and the device it finds

The repository’s CONTRIBUTING guide describes the build tasks. Check a source build’s results with verify.