A historic-district intern scores an extractive QA helper with ordinary classification accuracy and reports the model broken. What smoke-check should the developer use?
Select an answer to reveal the explanation.
Short Explanation
Extractive QA is scored on whether the span matches, exact match or token-level F1. Ordinary classification accuracy will call a good span model broken. A multi-week experiment and a graph modularity number are not a developer smoke-check.
Full Explanation
A developer smoke-check must use a metric that matches the model head. Extractive QA is judged by whether the predicted span matches the gold span, typically with exact match or token-level F1. Ordinary classification accuracy belongs to a label head and will mis-score a span model. Full experiment design is Domain 3 work, and a graph modularity score is unrelated.