Exam-style
A team has fine-tuned an open-weight model on SageMaker. Every training run must be kept alongside its evaluation metrics so an auditor can later see which artefact served which quarter, and a human has to approve a candidate before any traffic moves to it. Which mechanism records the runs and gates the promotion?
Reveal the answer
C. Register a model package version in a SageMaker Model Registry model package group, carrying the metrics and an approval status a pipeline reads
A model package version records the artefact URI, the inference container, the evaluation metrics and the model card. Group those versions and the history of a fine-tuning line reads as an ordered set of candidates. Its approval status is the gate: a pipeline only deploys a version marked Approved, and a person moves it there. Bedrock prompt versions freeze prompt text. Agent aliases freeze an agent’s configuration. A custom model id names imported weights. None of them carries training metrics or an approval state. S3 tags hold the same strings with nothing enforcing them.
Q. Two versioning vocabularies sit next to each other on this track, and they version different things. What does SageMaker Model Registry version, and what does Bedrock version?
A. SageMaker Model Registry for versioning covers model package versions inside a model package group: artefact URI, inference container, metrics, model card. Each carries an approval status, which is what approval workflows read before a pipeline may promote it. Bedrock versions something else entirely: model ids, prompt versions, guardrail versions and agent aliases. None of those describes a training artefact. See importing custom weights for where the two meet, and versioning the Bedrock side for the other half.
Why? Because rollback strategies for failed deployments differ with the vocabulary: a SageMaker endpoint update rolls back through deployment guardrails with a canary and a CloudWatch alarm, while a Bedrock custom model rolls back by repointing the model id or the Provisioned Throughput commitment behind it.