Release Notes#

cuStabilizer v0.3.0#

Added GPU-accelerated Detector Error Model (DEM) sampling for generating detection events from quantum error correction circuits without full circuit simulation.
New C API functions for Bernoulli sampling and GF(2) matrix multiplication:
- custabilizerSampleProbArray() – dense bit-packed Bernoulli sampling.
- custabilizerSampleProbArraySparsePrepare() / custabilizerSampleProbArraySparseCompute() – sparse Bernoulli sampling into CSR format with a prepare/compute workflow.
- custabilizerGF2SparseDenseMatrixMultiply() – GF(2) sparse × dense matrix multiplication.
- custabilizerGF2SparseSparseMatrixMultiply() – GF(2) sparse × sparse matrix multiplication.
New status code CUSTABILIZER_STATUS_INSUFFICIENT_SPARSE_STORAGE returned when sparse output capacity is exceeded.
FrameSimulator now randomizes the initial Z frame with Bernoulli(0.5) samples when randomize_measurements=True and no input tables are provided.

cupy is now required to import cuquantum.stabilizer. In v0.2.0, the module could be used with numpy-only workflows. This will be addressed in a future release.
DEM sampling classes (DEMSampler, BitMatrixSparseSampler) accept only numpy and cupy arrays. torch.Tensor is not supported.

Added shot masking instructions MASK_SET/MASK_XOR (aliases: SHOT_MASK_SET/SHOT_MASK_XOR) to enable or disable ranges of shots for subsequent circuit instructions. See Shot mask instructions for usage examples.

The initial release of cuStabilizer. Supported features: Circuit conversion and Pauli frame simulation.