Introduction to Numba: Glossary

The central feature of Numba is the @jit decoration.

Function signatures provide Numba with additional information that can help improve performance.

Specialization can make the functions less flexible.

The nopython mode will generate the best performance, but has limitations.

Types that can’t be inferred by the compiler will generate an error.

Numba functions call other Numba functions efficiently.

The @vectorize decorator can be used to create fast ufuncs.

Numba provides an easy way to write CUDA programs.

Many CUDA features are provided by Numba.

Some of the low level CUDA API features are not supported by Numba.