quiz bu lëkkale ak tegtal · Diggante Niveau
Deployée bu baxa-wert ngir quiz model ML
Xam dagg trafik bi ak dellu ginaaw ci génne yu baxa-wert, boole ci kàttan, deggoo schema ak àppu dellu ginaaw ci yoon wi.
Laaj 1 de 8
naka lañuy joxe environmaa bu baxa ak bu wert ci diiru génne gi?
Xoolaatal laaj bu nekk8 laaj
Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.
- naka lañuy joxe environmaa bu baxa ak bu wert ci diiru génne gi?
- One serves current traffic while the other stages a candidate release
- They store separate training and test labels
- They run different model classes permanently
- They split the GPU into two hardware drivers
- Lan mooy tax ñu mëna delloosi yoon wi gaaw ginaaw dagg bu baaxul?
- The new model is retrained during every request
- The previous environment remains available and healthy
- The database schema is deleted
- The old artifact was overwritten
- Lan moo tax ñuy waajal kàttan ngir baxa ak wert benn yoon?
- The model runs on both environments after cleanup forever
- Traffic is always doubled by the router
- Both production-like environments may need resources during rollout
- Blue-green requires two datasets
- Lan moo waral coppite yi ci base de done yi wara wéy di méngoo ak ñaari version yi ci diiru cutover?
- Green must rewrite blue's weights
- The router changes database columns automatically
- ML models do not read data schemas
- A rollback may return traffic to the older application or model
- Luy trafik lëndëm?
- An automatic full production switch
- Candidate predictions for comparison without serving them to users
- A guarantee of zero latency
- A replacement for model evaluation data
- What does routing rollback not necessarily undo?
- The router's traffic target
- Which environment receives future requests
- External side effects or writes already made
- The old environment's availability
- How does a canary relate to the basic blue-green cutover?
- It only validates container image signatures
- It is another name for two environments with no traffic shift
- It eliminates the need for a green environment
- It can expose a small traffic fraction before the full switch
- Why warm a large model in the green environment before switching?
- It can reduce initialization delay and accelerator allocation surprises
- It guarantees output quality
- It eliminates the old environment's costs
- It validates every subgroup metric