I’m trying a new model management set-up leveraging the model registry. My previous model management loop would go and check for any new aliases in a model which contained the word ‘challenger’, these models would then be pulled through to an evaluation job against the current ‘champion’ model to check whether any new challenger models should be staged for evaluation by a ML Eng to decide whether it should be promoted to production.
The previous code used to do this was leveraging api.artifact_versions
:
api = wandb.Api(overrides={"project": "test-project"})
artifacts = api.artifact_versions("model", "model-name")
This returns the different versions of that model which exist (with all the alias names) as a list of lists of aliases
When applying this same logic against a model-registry (collection) the api no longer works / isn’t able to resolve the items held in the model registry.
Is there a different way to access the items held in the model registry?