A pharmaceutical company's Foundry-hosted assistant is used to draft regulatory submission summaries, and outputs must remain reproducible across a documented validation cycle that references a specific software configuration. Which deployment practice best supports this requirement?
Select an answer to reveal the explanation.
Short Explanation
You need to pin the deployment to a specific, fixed model version because a validation cycle has to reproduce the exact behavior that was signed off. An alias that always points to the latest release can swap the model out later, and autoscaling or temperature 0 won't lock the model version down.
Full Explanation
The validation cycle must reference a documented software configuration, so the exact model behavior has to stay fixed until the company deliberately revalidates it. The deployment should therefore point to a specific, immutable model version rather than an alias that can silently swap in a newer model. Deploying the model using a version alias that always points to the latest release is incorrect because that alias can change the underlying model at any time as the provider releases updates, breaking reproducibility for a submission that was validated against a prior version. Enabling autoscaling on the deployment to keep response times consistent is incorrect because autoscaling adjusts compute resources to maintain response times under varying load; it has no effect on which model version answers the request. Setting the temperature parameter to 0 to eliminate randomness is incorrect because it reduces randomness in a single model's output but does not prevent the underlying model version itself from being changed or retired by the provider, so it does not satisfy a reproducibility requirement tied to a documented configuration.