Skip to main content
v1.0.1 (Source-Available) · Free Workbench GUI (CustomTkinter + MCP tools)

QECTOR Decoder v3

Rust-core Python library implementing 17 decoder configurations from exact MWPM to GPU batch. v1.0.1 is the current stable release: API stability tiers, Relay-BP and CS-OSD, Sinter/qiskit entry points, and the qector CLI, with build provenance for every wheel. A reproducible benchmark harness (qector bench) ships with the package so you can measure on your own hardware. Stim-native. PyPI binary wheels. All artifacts published on GitHub.

Decoder Configs
17

From exact MWPM to GPU batch, across documented stability tiers

Binary Wheels
18

Python 3.9-3.14 on Windows amd64, Linux x86_64, macOS arm64

Stable API
v1.0.1

Stable release with documented API stability tiers

CI Test Suite
Automated

Continuous validation and Stim comparison test suite on GitHub

Decoding Algorithms

Stable decoders with validation artifacts on GitHub. Experimental decoders are research-stage. Since v1.0.0 the public API is frozen under documented stability tiers. No universal benchmark figures are published; the qector bench harness ships in the package for measuring on your own hardware.

Stable Decoders

Union-Find

Near-Linear

Near-linear time approximate decoder. High-throughput option for graph-like codes: trades some LER accuracy for speed.

Fast Union-Find

Hot Path

Optimized Union-Find hot path for low-latency offline simulation pipelines.

Blossom MWPM

Exact Reference

Exact minimum-weight perfect matching for graph-like codes. The reference decoder for surface codes.

CPU & GPU Batch Decoder

Parallel

Native CUDA / OpenCL batch decoding for supported workloads. Throughput depends on hardware and batch size; CUDA accepts edge_weights and precision="f64" in v1.0.0.

Experimental & Research Decoders

BP-OSD

qLDPC Research

Belief propagation + ordered statistics decoding. v1.0.0 adds Relay-BP schedules, CS-OSD(lambda, w) and LLR damping.

Belief-Matching

Correlated Noise

BP pre-processing + reweighted exact MWPM. Research path for correlated noise scenarios.

Sparse Blossom

Near-Optimal

Region-growing blossom variant for detector graphs. Zero-allocation hot path in v1.0.0.

Hybrid & AutoDecoder

Adaptive Routing

AutoDecoder and HybridDecoder route between Union-Find and Blossom based on code properties.

Lookup-Table Decoder

Small Codes

Precomputed correction table for small codes or syndrome subspaces.

Colour Code / Two-Stage / Ambiguity-Cluster

Research

New v1.0.0 families: ColourCodeDecoder (opt-in cluster_bposd), TwoStageDecoder, AmbiguityClusterDecoder.

Technical Specifications

LanguagesRust core (PyO3) / Python 3.9-3.14 API
PlatformsLinux x86_64, macOS arm64, Windows amd64
GPUCUDA / OpenCL batch paths (supported configurations per package docs)
QEC LibraryStim (quantum-circuit noise simulation)
PackagingPyPI binary wheels (manylinux, macOS, Windows)
LicensePolyForm Noncommercial 1.0.0 (community) / Commercial
ValidationGitHub: github.com/GuillaumeLessard/qector-decoder
Validation Status

v1.0.0 (2026-08-06) was the first stable release; v1.0.1 (2026-11-06) is current. Decode runs are syndrome-validated (H·c = s) through the self-debugging harness, and artifact manifests are SHA-256 sealed on GitHub. No universal benchmark figures are published; the qector bench harness ships in the package for measuring on your own hardware.

GitHub Artifacts and Harness →