AWS Certified AI Practitioner AIF-C01 Practice Question

An organization needs its chatbot to answer questions by referencing 50,000 internal FAQ documents. The team has a limited budget and wants to avoid expensive model-training jobs. Which customization approach offers the lowest upfront compute cost while still allowing the bot to use those documents for accurate answers?

Apply parameter-efficient fine-tuning (such as LoRA) on the foundation model using the FAQ data.
Use Retrieval Augmented Generation (RAG) with an existing foundation model.
Fine-tune the entire parameters of an existing foundation model with the FAQ data.
Pre-train a new large language model on the FAQ data.

Report Issue

Answer Description

Retrieval Augmented Generation (RAG) lets a solution store document embeddings in a vector database and retrieve the most relevant passages at inference time, then pass them to an existing foundation model in the prompt. Because the model itself is not re-trained, the only additional cost is generating and storing embeddings, which is far lower than pre-training a new model or running any kind of fine-tuning, whether full or parameter-efficient. Pre-training from scratch and full fine-tuning both require large GPU clusters and many training epochs, while parameter-efficient fine-tuning (for example, LoRA) is cheaper than full fine-tuning but still entails running training jobs and storing new model weights. Therefore, RAG provides the smallest upfront compute expense for incorporating the company's documents.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.