Model Serving Canary Rollout Language Collocations
Practise the standard verbs for safely rolling out a new served model version.
0 / 5 completed
1 / 5
Fill in: 'We ___ a small percentage of live traffic to the new model version so a regression surfaces before it reaches every user.'
We 'route traffic' — the standard, established collocation for directing a portion of requests to a new deployment. The other options aren't the recognised term here.
2 / 5
Fill in: 'Rolling out a new model version to all traffic at once can ___ every user affected the moment a regression appears.'
We say a full rollout will 'leave' every user exposed at once — the standard, natural collocation for the resulting risk. The other options aren't idiomatic here.
3 / 5
Fill in: 'We ___ prediction quality between the canary and the current production model before increasing the canary's traffic share.'
We 'compare' quality — the standard, simple collocation for contrasting two measured outcomes. The other options are less idiomatic here.
4 / 5
Fill in: 'We ___ the canary's traffic percentage gradually once its error rate matches the baseline model over a sustained window.'
We 'widen' a percentage — the standard, simple collocation for gradually increasing a rollout's exposure. The other options are less idiomatic here.
5 / 5
Fill in: 'We ___ the rollout automatically the moment the canary's error rate crosses a predefined threshold, without waiting for a human page.'
We 'abort a rollout' — the standard, established collocation for automatically halting a deployment on a failed guardrail. The other options aren't the recognised term here.