Implement Custom Language Models AI-102 Practice Questions (Page 13)

Questions 61–65

1. 61foundation · easy

   You are building a language model for a restaurant reservation bot. Which of the following should be defined as an entity?

   • AMakeReservation
   • BCancelReservation
   • CReservationTime
   • DFindRestaurant
   Reveal Solution Discussion Select an answer first
2. 62foundation · easy

   Which of the following is required to consume a deployed custom language model from a client application?

   • AThe endpoint URL and an authentication key
   • BThe model's training data
   • CA backup file of the model
   • DAccess to the Azure portal to view the model
   Reveal Solution Discussion Select an answer first
4. 63foundation · easy

   When should you train a knowledge base in the question answering lifecycle?

   • ABefore creating the project
   • BAfter adding or editing question-answer pairs and before publishing
   • COnly after publishing the knowledge base
   • DOnly when the knowledge base is deleted
   Reveal Solution Discussion Select an answer first
5. 64application · medium

   A company wants to build a custom question answering solution for their internal HR portal. Employees will ask questions about company policies. The team has a set of frequently asked questions in a Word document. What is the first step to create the solution?

   • ACreate a new custom question answering project in the Language service and import the Word document as a source.
   • BUse Azure Cognitive Search to index the Word document and create a search service.
   • CPublish the Word document to a public website and use it as a URL source.
   • DTrain a custom language model with intents and entities to answer the questions.
   Reveal Solution Discussion Select an answer first
6. 65application · medium

   A team has trained a custom language model for a banking chatbot. They want to measure how well the model performs on unseen data before going live. They have a labeled test set that was not used during training. What should they do to evaluate the model?

   • AUse the training data to compute accuracy, since the model already knows these examples.
   • BCompare the model's predictions on the test set to the predictions of a random baseline.
   • CDeploy the model and rely on user feedback to evaluate performance.
   • DRun the test set through the trained model and compare the predicted intents and entities against the ground truth labels.
   Reveal Solution Discussion Select an answer first