A toolbox of low-level utilities for working with numba.

Documentation is available at numbox.

• Any: A lightweight structure that wraps any value into the same type leveraging a variant of the type erasure technique.
• Bindings: JIT-wrappers of functions imported from dynamically linked libraries.
• Node: A lightweight JIT-compatible graph node with type-erased dependencies in a uniform dynamically-sized vector (numba List).
• Proxy: Create a proxy for a decorated function with specified signatures, enabling efficient JIT compilation and caching.
• Variable: Pure Python framework of a graph calculation. JIT-compiled dispatchers can be invoked by it.
• Work: JIT-compatible unit of calculation work with dependencies, inner states, and custom calculation.

pip install numbox