A company introduces a large language model (LLM) in an application in order to monitor...

Basic Concept: LLMs process and generate text as discrete units called tokens, which represent words or word pieces. The number of tokens consumed is the primary metric for LLM resource utilization and billing. Understanding token consumption is essential for both performance monitoring and cost management. CompTIA SecAI+ Study Guide covers token-based measurement under AI system management.

Why A is Correct: Tokens are the fundamental unit of measurement for LLM utilization. Both input tokens sent in prompts and output tokens generated in responses are counted. Token consumption directly correlates with computational resource usage, response time, and API cost. Monitoring token counts enables the company to establish usage baselines and detect the abnormal token consumption spikes associated with DoS attacks or denial-of-wallet attacks against the LLM.

Why B is Wrong: A transformer is the neural network architecture that underlies modern LLMs. It is an architectural description, not a measurable utilization metric. You cannot quantify LLM resource consumption in " transformers. "

Why C is Wrong: Chain of thought is a prompting technique that encourages models to reason step-by-step before providing answers. It is an inference strategy for improving response quality, not a unit of measurement for monitoring LLM utilization or resource consumption.

Why D is Wrong: A prompt is the input text submitted to the LLM. While monitoring prompt volumes provides request count information, tokens provide a more granular and meaningful utilization metric because they capture both the complexity of inputs and the length of generated outputs.