AI-300 Practice Question 77: Deployment per App Quota
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 8 · question 7 of 10
A chatbot and a nightly summarization job share one model deployment. When the summarizer runs, the chatbot receives 429 errors. The team must guarantee that the summarizer can never consume the chatbot's capacity. What should they do?
Show answer and explanation
Answer:
Rate limits are assigned per deployment from the subscription's quota for a model and region. Giving each workload its own deployment with its own tokens-per-minute allocation isolates them, because one deployment's usage cannot exhaust another's limit (option C).
- A is wrong because this changes the response size of each call. It does not stop the summarizer from using the shared limit. - B is wrong because the deployment would still be one shared pool, so the summarizer could still consume most of it. - D is wrong because retries help with short bursts, but the chatbot still competes for the same limit during a long summarizer run.
Reference: Manage Azure OpenAI quota