AWS Certified AI Practitioner AIF-C01 Practice Question

Retrieval-Augmented Generation (RAG) leaves the foundation model weights unchanged, so no training or fine-tuning jobs are required. The company only incurs the cost of creating embeddings, storing them in a vector database, and calling the model at inference time. Pre-training or full fine-tuning would require extensive GPU training time and are far more expensive. Parameter-efficient fine-tuning (such as LoRA) reduces cost compared to full fine-tuning but still involves paid training runs, making it costlier than RAG for incorporating documents.