Question 270
An MLflow training job logs loss and parameters but never logs the trained model artifact. The team wants to register it. What should happen first?
Explanation
Correct Answer
Registration needs an actual model artifact; scalar training metrics alone are not a deployable model package.
Incorrect Answers
Verify model-asset write permissions in the target workspace or registry. It addresses another case: An engineer has permission to read MLflow run metrics but receives an authorization error while registering the model. For that case, the distinguishing justification or acceptance check is: Run visibility and model registration are distinct operations governed by applicable Azure resource permissions. The present case instead states: An MLflow training job logs loss and parameters but never logs the trained model artifact. The team wants to register it. It calls for Log the model in a supported MLflow flavor and register the resulting artifact. Showing success for the other case would not establish: Registration needs an actual model artifact; scalar training metrics alone are not a deployable model package.
The alternative, Use the compatible MLflow flavor and model loading method for the artifact, belongs to a different technical condition: A model is logged using a framework-specific MLflow flavor, but a deployment script expects a different loading interface. There the source-specific mechanism or proof is: Model flavors describe serialization and inference interfaces; a mismatched loader cannot reliably reconstruct the model. The condition asked about here is: An MLflow training job logs loss and parameters but never logs the trained model artifact. The team wants to register it. A suitable correction is Log the model in a supported MLflow flavor and register the resulting artifact; the targeted justification or result is: Registration needs an actual model artifact; scalar training metrics alone are not a deployable model package.
Record the resolved registered-model version or artifact URI in deployment metadata. That operation would be relevant if: A team wants to validate the exact model loaded by a production deployment because both old and new versions are present. Its corresponding reason or verification is: Deployment provenance should identify the exact artifact, not merely a friendly model label shared by many versions. However the problem in this question is: An MLflow training job logs loss and parameters but never logs the trained model artifact. The team wants to register it. The requirement instead calls for Log the model in a supported MLflow flavor and register the resulting artifact; the question-specific check is: Registration needs an actual model artifact; scalar training metrics alone are not a deployable model package.