Question 1

As a Generative AI engineer, you are responsible for developing and deploying a machine learning model that provides real-time predictions for a financial services application. The model must be monitored for performance, and you need to track its training metrics, serve it in production, and evaluate its accuracy over time. Which combination of Databricks products would best address these requirements?

Options :

A : Use MLflow Tracking to record and manage the model-s training metrics, Model Serving for deploying the model in production, and Lakehouse Monitoring to monitor the model-s performance and operational metrics.

B : Use Lakehouse Monitoring to evaluate the model-s accuracy, MLflow Tracking to deploy the model, and Model Serving to monitor the model-s performance in production.

C : Use MLflow Evaluation to track the model-s training process, MLflow Tracking for evaluating the model-s accuracy, and Model Serving to deploy the model.

D : Use MLflow Tracking to monitor the model-s real-time performance, MLflow Evaluation for tracking training metrics, and Lakehouse Monitoring to serve the model in production.

Answer: A

Question 2

What is a key challenge when implementing reranking in a Retrieval-Augmented Generation (RAG) pipeline for a Gen AI application?

Options :

A : Reranking simplifies the RAG pipeline by reducing the overall complexity and streamlining the process.

B : Reranking requires minimal computational resources, making it an efficient process for large datasets.

C : Reranking eliminates the need for any further processing or adjustments to the retrieved documents before generating a response.

D : Reranking involves multiple calls to the LLM, which can significantly increase both the cost and latency of the RAG chain.

Answer: D

Question 3

When choosing between "Deploy Model" and "Deploy Code" patterns in a Databricks environment, which approach is generally recommended for more effective management of ML workflows?

Options :

A : Deploy Model: Move the model artifact from staging to production after training in the dev environment, and handle other code (inference, monitoring) separately.

B : Deploy Code: Develop and test code in separate environments, then deploy it to production, ensuring the training pipeline is also run in each environment while deploying the model only in production.

C : Deploy Model: Train the model in the development environment and deploy it directly to production without separate processes for inference and monitoring.

D : Deploy Code: Train the model only in the production environment and use the same environment for testing and deployment.

Answer: B

Question 4

In which of the following scenarios model learns using RAG?

Options :

A : Model Fine Tuning

B : None of these options

C : Model Pre-Training

D : Passing Contextual Information

Answer: D

Question 5

As a Gen AI engineer managing the lifecycle of models in MLflow, which of the following URI patterns are used for referencing models in the Unity Catalog (MLflow) Model Registry? Choose two correct answers.

Options :

A : models:/{model_name}/latest

B : models:/{model_name}@{alias}

C : models:/{model_name}@{version}

D : models:/{model_name}/{model_version}

Answer: B,D