問題文
A team has finished training a text classifier and hands the saved checkpoint to the platform group, which will serve it behind an internal endpoint. A junior engineer asks what actually differs between what the training group did and what the serving group will do. Which answer describes the difference correctly?
選択肢
- Training produces the labels for the dataset, and inference then uses those labels to measure how well the network has learned on the most recent pass.
- Inference updates the parameters slightly on every request so that the model keeps improving in production without a separate retraining job.
- Training adjusts the parameters using labeled examples and an error signal, while inference keeps those parameters fixed and computes an output for each new input.
- Training runs on the processor while inference always runs on specialized accelerators, and this hardware placement is the only difference between the two phases.