Numba

Numba is an open-source just-in-time compiler for Python that focuses on accelerating numerical code. Built to work with NumPy and Python numerical constructs, Numba translates a subset of Python functions into optimized machine code at runtime using the LLVM compiler infrastructure. By compiling hot functions, it aims to deliver speeds close to compiled languages while remaining easy to integrate into existing Python code.

Numba provides decorators such as @jit and @njit to mark functions for compilation. In nopython mode, the function is compiled to native code with no Python object interactions; in object mode, Python objects may be used but performance may be reduced. Types can be inferred automatically, or a explicit signature can be supplied for better performance and error checking.

Numba supports parallel execution through the prange construct and multi-threading, enabling parallel loops to run on CPUs. It also includes GPU support via the numba.cuda module, allowing users to write CUDA kernels and launch them from Python to run on NVIDIA GPUs.

Numba targets a subset of Python and NumPy; features outside the supported subset may fall back to object mode or fail. It works best for numerical kernels and array-oriented code, but not all Python code or NumPy functions are supported. Compilation incurs a runtime overhead, so it is most beneficial for long-running calculations.

Developed as part of the NumPy ecosystem, Numba is maintained by the Numba project with contributions from the community and organizations; it is distributed as an open-source project under a permissive license (BSD 3-Clause). It is commonly used in scientific computing, data analysis, and performance-critical simulations.