C_AIG_2412 - SAP Generative AI Developer Part 2

Why does SAP AI Core use Kubernetes infrastructure to manage the containerized services? There are three correct answers. The applications can be broken down into smaller parts. The applications circulate between the different containers to offload them. The applications can handle more load when needed. The applications are more widely available.

How can you leverage the power of GPUs for training a Machine Learning Model on SAP AI Core? Choose the correct answer. Specify an NVIDIA GPU in your python code. Chose one of the resource plans that include GPU, for example, "train.l" in the training template. GPUs are a scarce resource and may not be always available. Kubernetes resources must be reserved via API.

In which of the following ways can the costs for serving models be reduced? Choose the correct answer. When processing inference requests, Kubernetes allows to scale Model Servers on demand. By applying the same binary string to data used for training. By sending new data to the model every quarter. When using the Autoscale to Zero feature, inference servers are "stopped" until the next request is received. Container.

What is NOT a benefit of using SDKs for generative AI hub mentioned in the lesson? Choose the correct answer. Increased efficiency and ease of integration. Customization opportunities. Streamlined development processes. In-house training of new AI models.

In which scenario are orchestration services particularly useful in the context of the problem stated? There are two correct answers. To avoid complex and redundant code and workflows. To enhance UI/UX design. For seamless integration and management of diverse components like data pipelines, AI models, and prebuilt modules. For video rendering.

Which key component is necessary for configuring models in the generative AI hub SDK? There are three correct answers. Client ID. Authentication URL. AI Core Base URL. SAP ERP Instances.

Which evaluation function aspect is NOT implemented in the provided solution? Choose the correct answer. RateLimitedIterator for controlling iterations. Evaluation function to validate JSON output. A final function to evaluate large datasets. Financial impact analysis.

Why is it important to evaluate prompts using generative AI hub SDK? There are two correct answers. To ensure automated and consistent evaluation across various scenarios. To automate all accounting processes. To use objective metrics such as relevance, coherence, and fluency. To enhance company annual revenue directly.

What are the benefits of using LLMs in generative AI hub? There are three correct answers. They can be used to enhance business user experience. They are appropriate for a wide range of industries. They can be accessed only through SDKs. They increase flexibility by handling multimodal inputs and outputs.

When utilizing SDK in generative AI hub, which function is primarily used to generate completions? Choose the correct answer. Chat(). Embeddings(). Completions.create(). Prompt().

What is a key advantage of using orchestration services in generative AI hub? There are two correct answers. Coordinating and managing deployment and integration of AI components. Automated accounting and payroll services. Streamlining and automating the end-to-end lifecycle of AI applications. Manual model tuning.

Which of the following can be improved by advanced prompting refinements? There are three correct answers. Specificity of AI outputs. Overall user experience. Hardware performance. Accuracy and relevance of responses.

Which method was specifically implemented to improve prompt responses in the Facility solutions scenario? There are two correct answers. Few-shot prompting. Metaprompting. Inserting code snippets manually. Updating the SAP ERP system.

What does the few-shot prompting technique involve? Choose the correct answer. Providing a single example to the model. Utilizing multiple examples to guide the model. Using a model without examples. Manual fine-tuning without context.

Which models are specifically managed under the global AI scenario foundation-models? There are four correct answers. Azure OpenAI Service models. SAP AI Core models. GCP Vertex AI models. Amazon Bedrock models. WuDao models.

Which model is exemplified as the best proprietary OpenAI model during evaluation? Choose the correct answer. meta--llama3-70b. gpt-4o. SAP HANA SQL. spacy-lm.

Which benefits did generative AI hub offer for selecting different models? There are two correct answers. Flexible access to various models. Hardware-specific integration. Ability to switch seamlessly between models. Managing financial transactions.

When evaluating different models, which factors must you consider? There are three correct answers. Cost Efficiency. Scalability. UI design principles. Flexibility.

Which the following best describes the capabilities of the orchestration service? Choose the correct answer. It orchestrates the activities of AI developer teams. It can streamline AI workflows. It ensures that non SAP data is prevented from accessing the AI model. It recommends Generative AI use cases.

Which advantages does advanced prompt engineering bring? There are three correct answers. Improved specificity. Improved adherence to ethical standards. Improved accuracy. Improved user experience. Improved number of hallucinations.

How does SAP make SAP AI Core capabilities assessable to non technical audiences? Choose the correct answer. By using a low-code approach. By offering parallel training in Python, TensorFlow, and R. Non technical audiences must avoid working on AI models due to the damage that amateurs could cause. By offering AI hackathons for business users.

Which technique must be applied to avoid the risk of exposing sensitive information during AI processing? Choose the correct answer. Document grounding. Data masking and content filtering. Prompt templating. Connect to LLMs.

Which technique must be applied in the case of inaccurate or irrelevant AI-generated responses? Choose the correct answer. Document grounding. Data masking and content filtering. Prompt templating. Connect to LLMs.

Which of the following statements are true of the AI Foundation on SAP BTP? There are two correct answers. This foundation provides tools and services to tackle business challenges. This foundation is only accessible to internal SAP developers to add new BTP capabilities. This foundation contains SAP hosted models and non SAP hosted models.

Which of the following are the advantages of using the orchestration service? There are three correct answers. Provider Agnostic: Easily access LLMs from various providers through a unified interface. Simplified Integration: Streamline the integration of LLMs into applications, promoting efficient and cost-effective development. Connection to SAP S/4HANA systems: Interface and APIs to connect to SAP S/4HANA systems. Expandability: Begin with essential features and add modules as needed, ensuring a smooth learning curve. Using a service architecture is the modern corporate way of rolling out such capabilities and this is consistent with that practice.

How can orchestration workflows be edited? Choose the correct answer. By showing or hiding modules through the Edit button and toggling the switch for the modules required. By showing or hiding modules through the Edit button and dragging the active modules into the ‘active’ pane. This is an administration function, for which the necessary permissions must be requested by the role. Once orchestration workflows have been designed and implemented, they cannot be edited but must be rebuilt from the start.

Which of the following are capabilities of the orchestration service? There are three correct answers. Coordinating and managing the deployment, integration, and interaction of multiple AI models and components within a system. Streamlining and automating processes such as: data flow management, model execution, and resource utilization. Grounding capabilities that enrich AI requests with relevant business context, reducing the need for custom data connections and using existing business data. Designing on-the-fly AI user interfaces that are appropriate to the business environment and context, and based on best practice design principles.

Which of the following is a valid, simple orchestration workflow? Choose the correct answer. A basic orchestration scenario allows you to combine different modules into a pipeline but the response must be kept separate by module. A basic orchestration scenario allows you to combine different modules into a pipeline from which the response from one module can be used as the input for the next module. A basic orchestration scenario only allows you to use one module in one pipeline.

Which of the following are options for creating vector embeddings for the Grounding module? There are two correct answers. Upload Documents to Supported Data Repository and Run Data Pipeline. Ensure that all documents are of the same type, before copying into a data repository. Provide Chunks of Documents via Vector API Directly.

Which sequence of steps are used in the Document Grounding module as part of the orchestration service to generate content with the RAG approach? Choose the correct answer. Configure the Document Grounding module, create the knowledge base, and finally generate content using the RAG approach based on the knowledge base. Generate content using the RAG approach based on the knowledge base and Configure the Document Grounding module. Note that the knowledge base is automatically created. Create the knowledge base, then Configure the Document Grounding module and finally generate content using the RAG approach based on the knowledge base.

The generative AI hub supports document grounding through which key features and processes? There are three correct answers. Access to LLMs: The generative AI hub provides instant access to a range of LLMs from different providers, such as GPT-4 by Azure OpenAI and OpenSource Meta LLama. Document Indexing: Unstructured and semistructured data from documents are preprocessed, split into chunks, and converted into embeddings using LLMs. Improved Context and Accuracy: By grounding AI responses in customers' specific documents (like HR policies), the generative AI hub enhances the context and accuracy of generated content. SQL-like document search capabilities, allowing for text-based embeddings in human readable format, for those experienced in SQL.

Which of the following accurately describes Document Grounding? Choose the correct answer. Document grounding describes the regular cadence when LLMs are updated ‘from the ground up’ to ensure they contain up-to-date data. Document grounding is the process by which multiple versions of corporate policy documents are reviewed for accuracy before being released for consumption by LLMs. Document grounding is a process that combines generative LLMs with advanced benchmarking techniques to filter out inaccurate information and hallucinations. Document grounding is a process that combines generative LLMs with advanced information retrieval techniques to improve the quality and accuracy of responses.

Vector embeddings in the context of the SAP HANA vector engine refer to which of the following? Choose the correct answer. A numerical representation of objects such as text, images, or audio. A text representation of objects such as text, images, or audio. A hash chain representation of objects such as text, images, or audio.

In Generative AI, what does a vector represent? Choose the correct answer. It brings together a discrete set of objects with their common characteristics into a set of normalized relational tables to allow AI models to quickly identify patterns. It represents an object such as a book, car, or customer record by describing the object itself, its attributes, or its characteristics for comparison. It represents the encoded description of the most interesting and unusual characteristics of objects such as a book, car, or customer record in string format.

The L2DISTANCE() function in SAP HANA Cloud vector engine allows you to: Choose the correct answer. Calculates the Euclidean (straight-line) distance between two vectors in a high-dimensional space, which is useful for various applications such as clustering and nearest neighbor search. Calculates the R-squared of a series of variables to identify potential patterns of relationship and correlation. Calculates the cosine similarity between two vectors. Cosine similarity measures the cosine of the angle between two vectors, which indicates how similar the vectors are in terms of their direction.

Which of the following can you do after training a machine learning model in SAP AI Core? Note: There are 2 correct answers to this question. Validate the accuracy of the model against business requirements. Analyze the model's performance metrics. Assess the resource allocation costs during training. Review the Kubernetes infrastructure scalability.

How can Kubernetes infrastructure enhance the functionality of SAP AI Core? Note: There are 3 correct answers to this question. By enabling the allocation of GPU resources for Al models. By eliminating the requirement for Docker containers. By supporting autoscaling of containers based on demand. By automatically developing Al models without human intervention. By ensuring high availability and fault tolerance for applications.

What are agents?. Software programs that act like intermediaries between humans and LLMs. Small units of code that perform specific tasks such as access to external data sources. Software programs that can be used to interact with LLMs.

What is the reason for dividing documents into smaller overlapping chunks in a RAG system?. To enable the matching of different relevant passages to user queries. To simplify the process of training the embedding model. To improve the efficiency of encoding queries into vector representations. To reduce the storage space required for the vector database.

Which of the following ways can Joule enhance workforce productivity? Note: There are 2 correct answers to this question. By providing context-based role-specific task assistance. By offering generic task recommendations unrelated to specific roles. By resolving hardware malfunctions. By maintaining strict adherence to data privacy regulations.