Nabla is a scientific computing and machine learning framework combining imperative and functional APIs. Seamlessly drop custom Mojo kernels into the autodiff engine and automatically shard distributed workloads.

Quick Install:

pip install --pre --extra-index-url https://whl.modular.com/nightly/simple/ modular nabla-ml

Getting Started Setup, tensors, autodiff, and your first training loop.

API Reference Core tensors, transforms, ops, and neural-network primitives.

Examples Guided notebooks across MLPs, transformers, and fine-tuning.

Nabla is currently in Alpha. It is an experimental framework designed to explore new ideas in ML infrastructure on top of Modular MAX. APIs are subject to change, and we welcome early adopters to join us in building the next generation of ML tools.