Chromobius: color code decoder

Chromobius is an implementation of a "Möbius decoder", which approximates the color code decoding problem as a minimum weight matching problem. Chromobius uses PyMatching to solve the minimum weight matching problem.

See the paper "New circuits and an open source decoder for the color code" for more details on how Chromobius works.

How to use Chromobius

Install the Python package from PyPI:

pip install chromobius

Once you've installed it, then:

• Work through the getting started notebook for an introductory tutorial on how to use Chromobius.
• Consult the Python API reference for detailed information about the classes and methods available.

Programmers who want to edit and build Chromobius can check the developer documentation. The pip package only installs the Python module; the standalone chromobius command-line tool used in some of the examples below must be built from source as described in the developer documentation.

Example Snippets

Decoding shots with Chromobius

From Python:

import stim
import chromobius
import numpy as np

def count_mistakes(circuit: stim.Circuit, shots: int) -> int:
    # Sample the circuit.
    dets, actual_obs_flips = circuit.compile_detector_sampler().sample(
        shots=shots,
        separate_observables=True,
        bit_packed=True,
    )

    # Decode with Chromobius.
    decoder = chromobius.compile_decoder_for_dem(circuit.detector_error_model())
    predicted_obs_flips = decoder.predict_obs_flips_from_dets_bit_packed(dets)

    # Count mistakes.
    return np.count_nonzero(np.any(predicted_obs_flips != actual_obs_flips, axis=1))

From the command line:

# Sample shots of detectors and observable flips.
stim detect \
    --shots 100000 \
    --in "example_circuit.stim" \
    --out "dets.b8" \
    --out_format "b8" \
    --obs_out "obs_actual.txt" \
    --obs_out_format "01"

# Extract a detector error model used to configure Chromobius.
stim analyze_errors \
    --in "example_circuit.stim" \
    --fold_loops \
    --out "dem.dem"

# Decode the shots.
chromobius predict \
    --dem "dem.dem" \
    --in "dets.b8" \
    --in_format "b8" \
    --out "obs_predicted.txt" \
    --out_format "01"

# Count the number of shots with a prediction mistake.
# (Appending "" forces awk to compare the fields as strings, not numbers.)
paste obs_actual.txt obs_predicted.txt \
    | awk '$1 "" != $2 ""' \
    | wc -l

From Python using sinter:

import sinter
import chromobius
import os

tasks: list[sinter.Task] = ...
stats: list[sinter.TaskStats] = sinter.collect(
    decoders=["chromobius"],
    custom_decoders=chromobius.sinter_decoders(),
    tasks=tasks,
    num_workers=os.cpu_count(),
    max_shots=100_000,
    max_errors=100,
)

From the command line using sinter:

sinter collect \
    --circuits "example_circuit.stim" \
    --decoders chromobius \
    --custom_decoders_module_function "chromobius:sinter_decoders" \
    --max_shots 100_000 \
    --max_errors 100 \
    --processes auto \
    --save_resume_filepath "stats.csv"

Disclaimer

This is not an officially supported Google product. This project is not eligible for the Google Open Source Software Vulnerability Rewards Program.

Copyright 2025 Google LLC.