Find your next starting point.

Serialization of Python objects and interpreter state for research workflows.

Data and pipeline versioning tools for reproducible computational projects.

Common filesystem interfaces and utilities for Python data-access libraries.

Attention-kernel implementations for comparing memory-efficient computation with a directly specified attention reference.

Neural-network tooling built around JAX for investigating explicit state, transformations and model composition.

Configuration injection for Python functions and classes, useful for making experiment parameter binding explicit.

Neural-network transformations for JAX, supporting studies of parameter initialization and functional apply behavior.

Configuration composition tooling for making experimental variants and overrides inspectable.

Array transformations and automatic differentiation for examining compilation, vectorization and functional numerical programs.

Function-result caching, persistence, and parallel execution helpers used in scientific Python workflows.

Text representations and synchronization for Jupyter notebooks.

Neural-network APIs for examining model behavior across declared computation backends.