QML Observer

An open-source observability and diagnostic framework for variational quantum machine learning (QML) training. QML Observer watches training runs in real time, detects pathologies such as probable barren plateaus, stagnation, and noise-dominated optimization, and can log, warn, pause, or stop training before expensive quantum computation is wasted.

Status: v0.7.0 — public beta. Core schemas, monitoring engine, detectors, diagnosis engine, actions (including a real PauseAction), both the PennyLane and Qiskit adapters, JSONL logging, run summaries, compute-saved estimation, the CLI, the calibration benchmark suite, webhook alerting (including a Slack-compatible formatter, alert deduplication/cooldowns, evidence redaction, and a webhook-URL SSRF safeguard), an optional read-only dashboard (qml-observer[dashboard]: live loss/gradient charts, a diagnosis panel, compute-usage panel, run history, and data export), opt-in research-grade diagnostics (qml_observer.advanced: QFIM estimation/conditioning, parameter-redundancy detection, Hessian-vector products, loss-landscape sampling, and qubit/depth gradient-variance scaling analysis — see docs/research/geometry.md), an opt-in recovery engine (qml_observer.recovery: ranked recovery strategies — reinitialization, learning-rate/shot-budget adjustment, ansatz-depth reduction, optimizer switching, natural gradient — plus recovery evaluation and monitor resume after a pause; see docs/architecture/recovery.md), and broader ecosystem support (PyTorch/JAX hybrid-workflow adapters, a generic autograd adapter, MLflow/W&B experiment-tracker integrations, a third-party detector plugin API, and local run comparison/experiment management — see docs/development/plugin_api.md and docs/architecture/run_history.md) are all shipped (Milestones 0–14). See CHANGELOG.md for the full release notes and docs/roadmap.md for what's next. The 0.x API is not yet stable and may change without a major-version bump, per SemVer's 0.x convention.

Architecture

Training events flow one-way through the pipeline above; nothing here ever owns or drives the quantum computation itself (plan.md §2). See docs/architecture/overview.md for the full, module-by-module breakdown, including the diagnosis engine's weighted-evidence scoring and the opt-in telemetry layer.