Installation

nirs4all-cluster needs Python ≥ 3.11. The server and client depend only on FastAPI / uvicorn / httpx / pydantic; nirs4all is not a dependency of the package.

install the server + client + worker transport

pip install nirs4all-cluster
# or, from a checkout:
uv pip install -e ".[dev]"

Server / client hosts

Nothing else is required. The coordinator and the client SDK/CLI run on the base package alone — you can submit and monitor jobs from a machine that has never seen nirs4all.

Worker hosts

A worker runs nirs4all.run() in a subprocess, so a worker host must already have a working nirs4all environment (plus whatever the pipelines need: scikit-learn, torch, tensorflow, …). Install nirs4all-cluster on top of that environment:

on a worker, inside the nirs4all environment

pip install nirs4all-cluster

The worker advertises the versions it found (nirs4all, numpy, torch, … and its own nirs4all-cluster version) so the server can route jobs to a compatible worker — see Version compatibility & pipeline fingerprints.

Documentation toolchain (optional)

pip install "nirs4all-cluster[docs]"   # sphinx + myst + furo, to build these docs