Training and inference
학습과 추론
Training is when an AI learns from data. Inference is when the trained AI actually produces answers.
Overview
Training is the process of showing a model a huge amount of data and adjusting the numbers inside it bit by bit. Training one large model takes many high-performance chips, a long time and a lot of electricity.
Inference is putting a question into a trained model and getting an answer. It happens every time you use an AI service. Each one costs little, but as more people use it, the total cost becomes larger than training.
An everyday example
Training is studying for an exam, and inference is taking the exam. You study hard once, but you take the exam every day.
Why it matters
This is why AI companies spend so much on data centers and chips. When the news talks about "inference cost," it means the money needed to run the service.
Also calledtraining, inference