Use Azure AI Foundry evaluation with a representative evaluation dataset and quality metrics to compare the fine-tuned model against the base model before production promotion. Among the available options, this corresponds to Prompt flow bulk testing / evaluation . Foundry evaluation supports systematic testing of model or application outputs across many test records rather than relying on manual inspection of isolated prompts. Microsoft provides built-in evaluators including Groundedness, Coherence, Relevance, Fluency, and Response Completeness , allowing teams to quantify quality consistently across candidate implementations.
For a valid comparison, run the same representative dataset against both the base and fine-tuned deployments and compare the resulting evaluation metrics. This creates objective evidence about whether fine-tuning improves the targeted behavior without materially degrading relevance, coherence, grounding, or other required quality dimensions. Prompt flow evaluation workflows also support batch execution, per-record scores, and aggregated metrics for comparing variants.
Application Insights is essential for production telemetry such as latency, failures, and usage, but it is not by itself a model-quality comparison framework. Content Safety configuration evaluates or controls harmful content rather than comparative model quality. Azure AI Search indexer scheduling manages search-index ingestion and is unrelated to model evaluation.
Therefore, A is the correct capability.
Study Guide references: Microsoft Foundry → evaluations; evaluation datasets; bulk/batch testing; Groundedness; Coherence; Relevance; comparative model assessment.
