A financial services firm is using InstructLab to customize an IBM Granite model for internal policy compliance checks. The goal is to teach the model two things: first, to recognize and classify new, firm-specific financial product names, and second, to follow a specific three-step reasoning process for evaluating compliance. How should the engineering team structure their taxonomy files in InstructLab to achieve this?
An engineer is implementing Low-Rank Adaptation (LoRA) to fine-tune a foundation model on a domain-specific dataset. To ensure training is both effective and resource-efficient, which TWO of the following are the most critical LoRA-specific hyperparameters to configure? (Select TWO)
A team is building a RAG system to answer questions from a large corpus of lengthy legal documents. The documents have a clear hierarchical structure (chapters, sections, clauses). The initial prototype using a fixed-size chunking strategy is performing poorly, often missing context that spans across chunk boundaries. Which chunking strategy should the team implement to best preserve the semantic context within these structured documents?
When using the watsonx.ai Prompt Lab, an engineer wants to increase the creativity and diversity of the model's generated responses, even at the risk of occasional non-factual statements. Which model parameter should be increased to achieve this effect?
A global logistics company plans to develop a generative AI-powered assistant for its supply chain managers. The assistant must provide real-time shipment tracking summaries, predict potential delays by analyzing weather and traffic data from external APIs, and answer queries about internal shipping protocols stored in a document repository. The solution needs to be highly responsive, secure, and capable of grounding its answers in the company's proprietary protocol documents to avoid hallucinations. The development team has expertise in Python and is using the watsonx.ai platform. The internal protocol documents are updated weekly. Which architectural design provides the most effective, secure, and maintainable solution for this use case?
A development team is managing multiple versions of a prompt template for a customer service chatbot within the watsonx.ai Prompt Lab. They need to test a new prompt version (v2) against the current production version (v1) without impacting all users. What is the most appropriate industry-standard strategy for deploying and evaluating the new prompt version?
True or False: Applying INT8 quantization to a fine-tuned foundation model will always reduce its inference latency and memory footprint without any impact on its accuracy.
A RAG-based chatbot designed to answer questions about internal HR policies is frequently providing answers that are factually correct but irrelevant to the user's specific question. For example, when asked "What is the policy for paternity leave?", it returns a detailed paragraph about the company's general holiday policy. The system uses an appropriate embedding model and a vector database. What is the most likely cause of this issue?
A developer is building a Python application to interact with a deployed model on watsonx.ai. They need to send a prompt and receive a generated response. Which class from the `ibm_watson_machine_learning.foundation_models` library is primarily used for this purpose?
A team is using the synthetic data generation feature within watsonx.ai to augment their dataset for fine-tuning. They provide a few high-quality examples of instruction-response pairs. What is the primary purpose of this feature?