Installation¶

Requirements¶

Python 3.10 or newer
The only runtime dependencies are NumPy and Numba — both are installed automatically.

Install¶

pip install numba-utils

That is the whole story: no compilers to configure, no optional native extensions. Numba brings its own LLVM-based compiler.

Verify¶

import numba_utils as nu
print(nu.__version__)

The cache kill-switch¶

Numba caches compiled kernels to disk. On multi-process farms, network filesystems, and ephemeral containers that cache can be dangerous — see Cache for the full story. If you hit crashes that look like they come from a stale or shared cache, disable it globally:

NUMBA_UTILS_CACHE=0        # environment

import numba_utils as nu
nu.configure(cache=False)  # or from code

This override wins even over per-call cache=True, by design: it exists as an environment policy tool, and a policy any call site can defeat is not a policy.

Development install¶

git clone https://github.com/nicoseijas/numba-utils.git
cd numba-utils
python -m venv .venv
.venv/Scripts/pip install -e .[dev]     # POSIX: .venv/bin/pip
.venv/Scripts/python -m pytest

See Contributing for the rules — benchmarks are mandatory, honesty is policy.