問題文
During fine-tuning, a team wants training to stop by itself when the evaluation loss stops improving. Which pair of hyperparameters controls that behavior?
選択肢
- The number of training epochs and the training batch size, because the two of them set how many steps the job runs before it stops.
- The LoRA rank and the LoRA dropout value, which as a pair decide the point at which a training run has converged.
- The early stopping threshold, which defines the minimum loss improvement that counts, and the early stopping patience, which defines how many evaluations may pass without that improvement.
- The learning rate and the log model metrics interval, because the interval controls how often the loss is checked and the rate controls whether it improves.