You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
hamiltonians: QDrift with n_steps > 1 now draws fresh terms at every Trotter step instead of replaying one draw, so its error keeps falling as steps are added. A seeded TimeEvolutionTrajectory gives each time point its own reproducible draw. Evolution circuits are lowered to the QASM basis, fixing failures on XZ/ZX terms, and QASM angles below 0.1 keep precision significant figures instead of rounding towards zero. Breaking: custom TrotterizationStrategy implementations must accept an n_steps keyword, TrotterizationResult.sampled_terms is a tuple with one SparsePauliOp per step, and sampled evolution rejects order != 1 (52b58e9).
🔄 Changed
ai: divi-ai defaults to Qwen 3.5 4B, with Qwen 3.5 9B as the higher-quality tier. Model downloads are pinned to fixed revisions, retrieval grounding is tighter, and inference runs on the available physical CPU cores. Breaking: the 1.5b, 3b and 7b models are removed; divi-ai asks you to choose again if one of them was saved (3767472).
deps: minimum versions raised to qiskit>=2.5, rich>=15.0 and numba>=0.68.0, and qiskit-ibm-runtime>=0.50 in the aer extra (56d2d3f).