Kubeflow container component
Opened this issue · 0 comments
vstirbu commented
The model card validation functionality can be integrated into Kubeflow as a container component.
The container should receive the MLflow run_id
, tracking_uri
and modelcard.yml
template from the Kubeflow pipeline, validate the model card and append the artefact.json
as a MLflow experiment artefact.
%%{init: {'theme':'forest'}}%%
sequenceDiagram
actor Data Scientist
participant kf as Kubeflow
participant mc as ModelCard Validation
participant mlflow as MLflow
Data Scientist ->> Data Scientist: create pipeline
Data Scientist ->> kf: deploy
activate kf
Note over kf: Training steps where run_id is created
kf ->> mc: trigger
activate mc
mc ->> mlflow: get run_id experiment details
mc ->> mc: validate
mc ->> mlflow: add modelcard.json artefact <br /> to run_id experiment
mc -->> kf: output modelcard <br /> as markdown
deactivate mc
Note over kf: More steps
kf -->> Data Scientist:
deactivate kf