Bivariate Bicycle Neural Decoder

Neural network decoder for IBM bivariate bicycle (BB) qLDPC codes — reproducing and extending Blue et al. (arXiv:2504.13043) and benchmarking against BP-OSD (Roffe et al. 2020) and Relay-BP (Müller et al., arXiv:2506.01779) on the [[72,12,6]] and [[144,12,12]] gross codes.

See docs/PLAN.md for the full project plan, success tiers, and phase structure.

Quickstart

python -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -e ".[dev]"

# Sanity-check core deps:
python -c "import stim, sinter, ldpc, torch; print('ok')"

# Run tests:
pytest -q

relay-bp requires a separate install (Rust toolchain). See https://github.com/trmue/relay and Phase 3 of the plan.

Layout

src/bb_decoder/   # library
configs/          # Hydra configs
scripts/          # entrypoints
notebooks/        # exploration + paper figures
tests/            # pytest suite
docs/PLAN.md      # full plan

References

• Bravyi et al., Nature 627, 778 (2024) — BB codes. arXiv:2308.07915
• Blue et al., arXiv:2504.13043 (2025) — ML decoder for BB codes. Primary reference.
• Müller et al., arXiv:2506.01779 (2025) — Relay-BP.
• Bausch et al., Nature (2024) and arXiv:2512.07737 — AlphaQubit / AlphaQubit 2.
• Roffe et al., Phys. Rev. Research 2, 043423 (2020) — BP-OSD.