フリー問題

問題文

Which end-to-end design keeps a fine-tuned model manageable in production?

選択肢

1. The fine-tuned model is deployed without further evaluation, since the training already optimized it
2. The training data is discarded after the job completes to reduce storage cost, and the model's own evaluation results are kept instead so that the recorded scores stand in for the data when the model's behavior is questioned later
3. The fine-tuned model replaces the base deployment so there is only one artifact to manage
4. The training data and its provenance recorded, the model evaluated against the same suite as the base model, deployed alongside rather than replacing it, and monitored with the successor model's retirement schedule tracked