aleph v0.2.0 — CPU parity release
Every cell of the competitive parity matrix is at or below 1.2× its reference, most cells faster (full report + protocol):
- State vector, 16 threads vs Qiskit Aer (n=25, default settings both sides): QFT 0.81×, Grover 0.60×, random brickwall 0.59× — aleph faster on 3/4; GHZ an allocation-bound tie (1.03×).
- MPS vs Aer matrix_product_state (sequential both sides): aleph 4.4–14× faster on all five workloads.
- Stabilizer vs Stim (surface-code syndrome cycle): 0.79× at d=11 — faster than Stim at every measured distance (was 1.64× behind in v0.1), via word-parallel row helpers + an AVX-512 in-register 64×64 bit-transpose.
Install
pip install aleph-sim # Python (module name: aleph)Wheels: Linux x86_64 (manylinux_2_28) + macOS arm64, Python ≥ 3.12 (abi3). CLI binaries attached below (static musl Linux x86_64 + macOS arm64).
Since v0.1
- Phase 4.5 (CPU parity): diagonal/k-qubit gate fusion already in the default pipeline now benchmarked default-vs-default against Aer; MPS lazy SWAP routing + faer hot path (P3-09); stabilizer word-parallel gates (P3-11) and rowops + AVX-512 transpose (P4.5-02).
- PyPI publication automated via trusted publishing (P4-09).
🤖 Generated with Claude Code