mojolearn documentation#
mojolearn is a GPU machine-learning library written in Mojo whose results are reproducible down to
the bits across GPU vendors. One codebase runs on Apple, NVIDIA and AMD GPUs and on the CPU, and covers
over 100 algorithms and model components, from gradient boosting and forests to transformers and
state space models. In the default identical mode, the same release, inputs and settings give the
same output bits on supported NVIDIA and AMD GPUs.
Install
Pick your platform and GPU, install with pip, and check your machine.
Quickstart
Train a model, predict, and see which device and numeric mode ran.
The reproducibility contract
What identical promises, what it does not, and what it costs.
API reference
Every public class and function, by family.