A fire-inspection loft wants to experiment by putting each candidate checkpoint behind a live server after every epoch. Where does the comparison belong?
Select an answer to reveal the explanation.
Short Explanation
The comparison belongs in an offline eval loop on pinned checkpoints. Standing up Triton or NIM after every epoch is a later serving choice, not the experiment. A rack queue and a safety-policy rewrite answer different questions.
Full Explanation
The experiment is the eval loop that scores pinned checkpoints on a gold set. Standing up Triton or NIM is a serving decision for a later domain, not the comparison itself. A datacenter queue or a safety-policy rewrite answers a different question. Associate practice keeps scoring offline until a candidate earns a deploy look.