| name | aquilia-mlops-model-lifecycle |
| description | Build Aquilia MLOps model packaging, registry, serving, rollout, observability, plugins, lineage, experiments, export, and admin workflows. Use for aquilia.mlops APIs and aq pack/model/mlops-deploy/observe/export/plugin/lineage/experiment commands. |
Aquilia Mlops Model Lifecycle
Purpose
Use Aquilia's implemented MLOps stack for model packaging, registry, runtime serving, deployment rollout, observability, plugins, lineage, and experiments.
Trigger Conditions
Use for modelpack archives, registry push/inspect/verify, serving models, ONNX/edge export, drift detection, metrics, rollout plans, plugin management, lineage queries, and A/B experiments.
Inputs
- Model path, name, version, framework/runtime, env lock, signing key, registry URL, rollout versions, drift data, plugin package, or experiment arms.
Execution Flow
- Use CLI groups from
mlops_cmds.py: pack, model, mlops-deploy, observe, export, plugin, lineage, and experiment.
- Use
MLOpsManifest, runtime/base classes, registry service/storage, orchestrator, scheduler, and observe modules where programmatic integration is needed.
- Configure workspace with
Integration.mlops(...) or Workspace.mlops(...) when app-level MLOps is required.
- Use contract APIs for request/response schemas when exposing MLOps endpoints.
- Verify archives and signatures before deployment or registry push.
Constraints
- Optional dependencies are split by extras: core mlops, onnx, torch, s3, bento, explain.
- Do not fake support for a runtime if
select_runtime or exporter code does not implement it.
- Signing and registry credentials must not be committed.
Implementation Anchors
aquilia/mlops/, aquilia/cli/commands/mlops_cmds.py, aquilia/integrations/mlops.py, aquilia/admin/templates/mlops.html, tests/test_mlops_*.py, examples/mlops_model_registry_app/.
Examples
aq pack save model.pkl --name recommender --version 1.0.0 --framework sklearn.aq model serve modelpack.aq --runtime python --port 9000.aq mlops-deploy rollout fraud --from-version 1.0 --to-version 1.1 --strategy canary.
Failure Handling
Pack integrity/signature failures should stop deployment. Missing optional ML dependencies should produce actionable install guidance. Drift/metrics failures should map to MLOps observe faults.
