Twelve PennyLane datasets from two Xanadu studies that score quantum classifiers and generators against classical models on synthetic and real bit-string tasks.
unassessed
| Category | domain |
|---|---|
| Subcategory | quantum machine-learning classification and generative datasets |
| Page status | active |
| Metric | test accuracy (classification); MMD (generative) |
| Direction | higher_is_better |
| Dataset size | 12 |
| Dataset licence | CC-BY-SA-4.0 |
| Publisher | Xanadu / PennyLane |
The collection is a shared testbed for near-term quantum machine learning, not a language-model exam. Six families come from a 2024 classification study that runs 12 quantum models on 160 binary datasets of rising dimension. Six more come from a 2025 generative study that trains IQP-style circuits on bit strings and scores samples with maximum mean discrepancy. Inputs are vectors or binary strings. Labels, when present, are class ids. The point is whether a quantum model beats a tuned classical baseline on the same split, not whether it chats.
Supervised binary classification (labels -1/1 or 0/7) or unlabelled generative sampling. Load via pennylane.data.load("other", name=<slug>) or generate from qml_benchmarks.data. sklearn-style fit/predict or sample().
No model card in ModelSpec reports this benchmark yet.