contextually appropriate responses. Which of the following steps are essential when preparing the dataset for prompt -tuning in this context? (Select two)
similarity calculation. B. Use different transformer models for documents and queries, and normalize their embeddings to align them in the same latent space. C. Fine-tune a transformer model on a document -query similarity task, so that both queries and documents are encoded into the same vector space for retrieval. D. Use a pre-trained BERT model to encode the documents and a pre-trained GPT model to encode the queries, ensuring diversity in embeddings. You are designing a Retrieval -Augmented Generation (RAG) system that will handle real -time queries from users, using a combination of a retriever and a transformer -based generator. Which of the following implementation details is the most critical to ensure that the system delivers responses in a timely manner while maintaining accuracy?
You are optimizing a large language model (LLM) for deployment on edge devices with limited computational resources. To reduce the model size and improve efficiency without significantly compromising performance, which of the following quantization techniques is most appropriate for this scenario?
computation time for each request. B. Prompt variables reduce redundancy by allowing dynamic inputs to be injected into a single prompt template, improving scalability. C. Using prompt variables allows the model to dynamically adjust its output based on context, without requiring multiple task -specific prompts. D. Prompt variables eliminate the need for fine-tuning the model on specific tasks since they allow on- the-fly customization of responses. E. Prompt variables require a complete re-training of the model whenever a new variable is introduced, which can be time -consuming. You are tasked with designing an AI prompt to extract specific data from unstructured text. You decide to use either a zero -shot or a few -shot prompting technique with an IBM Watsonx model. Which of the following statements best describes the key difference between zero -shot and few -shot prompting?
Which of the following stopping criteria can help in generating coherent and well-structured text without cutting off mid -sentence or continuing unnecessarily?
In the context of Tuning Studio in IBM watsonx, what is one of the key benefits of using Compute Unit Hours (CUHs) during the fine -tuning process?
