Installation¶

Pasteur ships as a Rust CLI and Python bindings. Choose the interface that matches your workflow; both execute locally.

Use case	Install	What you get
Simulation and evaluation	`cargo install pasteur-cli`	The complete command-line workflow
Python simulation	`pip install pypasteur`	Blackout and jitter simulators for Polars dataframes
Development	Build this workspace	CLI, libraries, bindings, and tests

Install the CLI¶

cargo install pasteur-cli
pasteur-cli --version
pasteur-cli --help

The workspace minimum supported Rust version is 1.95.

Install the Python bindings¶

python -m pip install pypasteur

The package is pypasteur, which requires CPython 3.12 or newer. The pasteur package on PyPI is an unrelated project; do not install it.

The simulators take polars DataFrames. To start from pandas, convert with pl.from_pandas(df). Always call fit before transform:

import polars as pl
import pypasteur

frame = pl.read_parquet("patients.parquet")
simulator = pypasteur.BlackoutSimulator(
    "glucose",
    rate=0.1,
    companions=["glucose_measured"],
    random_state=42,
)
simulator.fit(frame)
shifted = simulator.transform(frame)

Build from source¶

git clone https://github.com/Krv-Labs/pasteur.git
cd pasteur
cargo build -p pasteur-cli
cargo test --workspace --locked --exclude pypasteur-bindings

The executable is written to target/debug/pasteur-cli.

ONNX Runtime¶

The ort crate downloads ONNX Runtime at build time. Once built, pasteur-cli performs simulation and evaluation without opening network connections. For machines without internet access, see Offline installation.

Publishing is separate¶

Pasteur never treats a remote dataset name as a local path. Download artifacts with the official Hugging Face CLI first, then pass their local paths to Pasteur. Uploads are likewise explicit and external to pasteur-cli.