pennylane
Use PennyLane for quantum circuits, differentiable QNodes, devices, transforms, domain modules, data/io, and source-repo maintenance.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- pennylane
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use PennyLane for quantum circuits, differentiable QNodes, devices, transforms, domain modules, data/io, and source-repo maintenance.
- disable-model-invocation
- true
02 资源文件 8
- 参考文档references/development-conventions.md4.1 KB
- 参考文档references/install-and-inspection.md3.5 KB
- 参考文档references/package-map.md3.8 KB
- 参考文档references/repo-provenance.md2.9 KB
- 参考文档references/repo-routing-metadata.json328 B
- 参考文档references/troubleshooting.md4.5 KB
- 脚本scripts/inspect_pennylane_api.py1.5 KB
- 脚本scripts/pennylane_smoke.py1.0 KB
03 SKILL.md 正文
PennyLane repo skill
Use this skill when a task involves the PennyLane Python package, a PennyLane source checkout, or quantum-computing workflows built with PennyLane. It covers the verified CPU core of PennyLane 0.46.0-dev73 and explicitly marks optional framework, plugin, and accelerator paths.
First checks
- Import the package as
qpin examples and tests unless the surrounding project requires another alias:import pennylane as qp print(qp.version()) - For an installed-package smoke test, run
scripts/pennylane_smoke.py. It imports PennyLane, creates adefault.qubitQNode, executes it, and checks an Autograd gradient. - For API signatures, run
scripts/inspect_pennylane_api.pywith names such asQNode,qnode,device,grad,compile,qchem.Molecule, orestimator.estimate. - Read
references/install-and-inspection.mdbefore changing dependencies, optional extras, interface packages, or backend claims. - Read
references/troubleshooting.mdwhen imports, devices, measurements, gradients, optional backends, datasets, or source-checkout tests fail.
Route by task
- Build or debug circuits, QNodes, devices, measurements, shots, drawing, or
pl-device-test: readsub-skills/circuits-devices/SKILL.md. - Choose operations/templates, compose observables, inspect matrices, decompose/compile/transform circuits, or create a custom operator: read
sub-skills/operators-transforms/SKILL.md. - Work with gradients, differentiable interfaces, training loops, optimizers,
TorchLayer, JAX/Torch/Autograd behavior, or Catalyst/qjit caveats: readsub-skills/gradients-interfaces/SKILL.md. - Use application modules such as qchem, fermi/bose/spin mappings, QAOA, kernels, qcut, shadows, pulse,
specs, resources, or the estimator: readsub-skills/applications-qchem-resource/SKILL.md. - Import/export circuits, load datasets, use debugging snapshots, configure logging, or reason about pytrees/concurrency support: read
sub-skills/io-data-logging/SKILL.md. - Modify a PennyLane source checkout, add operators/devices/plugins, choose focused tests, run lint/format/tach, update changelog, or respect repo AI/GitHub policy: read
sub-skills/repo-development/SKILL.mdandreferences/development-conventions.md.
What is verified here
- Required scope: CPU package import,
default.qubit, QNode execution, Autograd gradients, installed package metadata, and thepl-device-testconsole script help. - Optional/unverified unless the user or environment proves them: CUDA/ROCm/MPS devices,
lightning.gpu, external plugins, Catalyst runtime, Torch/JAX/TensorFlow packages, qchem external solver extras, remote datasets, and network downloads. - TensorFlow support is documented as no longer maintained in this repository version; prefer JAX or Torch for new ML workflows unless maintaining existing TensorFlow code.
Core mental model
PennyLane quantum programs are Python functions containing quantum operations and ending in measurement processes. A QNode binds such a function to a device and exposes it as a callable object. Interfaces and differentiation methods determine how trainable arrays flow through the QNode and how gradients are computed. Operations, templates, transforms, and domain modules are composable layers around that core.
Public package map
Read references/package-map.md for the high-level module map. Use source-level facts only after checking live API signatures or the relevant reference in this skill; avoid relying on memory because PennyLane changes signatures and module boundaries frequently.
Development safety
If the task changes a PennyLane checkout, keep edits minimal, run only relevant tests, then run configured lint/format tools on changed files. Do not open or comment on GitHub issues/PRs, do not commit, and do not silence lint/coverage warnings without explicit human approval. Details are in references/development-conventions.md.
Provenance and routing metadata
- Source snapshot and evidence paths:
references/repo-provenance.md - Router import metadata:
references/repo-routing-metadata.json
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).