A company is trying to adapt a pretrained foundation model that can only give general-purpose responses, for its own specific task, by additionally training it with labeled data. Which lifecycle stage does this work correspond to?

1 / 1
Select an answer
CorrectA

Explanation

Question Overview

Select the stage that adapts a model to a specific task.

Requirements to satisfy
  • 1「for its own specific task」Fit a general-purpose model to a specific use
  • 2「additionally training it with labeled data」Update the weights to adapt = fine-tuning
Per-option explanation
ACorrect

Fine-tuning

This is correct. Fine-tuning is the stage that adapts a pretrained foundation model to a specific task or domain by additionally training it with labeled data. It can adjust behavior with a relatively small amount of data.

The difference from pre-training is that pre-training is the process of acquiring general knowledge from scratch with a large amount of unlabeled data (requiring vast data and compute), whereas fine-tuning adds a small amount of labeled data to a model that already has general capabilities to steer it toward a specific use.

BIncorrect

Pre-training

Pre-training is the stage that acquires general knowledge from scratch with a large amount of unlabeled data.

It is not the stage that adapts a pretrained model to a specific task, so it is incorrect.

CIncorrect

Data selection

Data selection is the first stage of deciding which data to use for training.

It is not the stage that additionally trains a model and adapts it to a specific task, so it is incorrect.

DIncorrect

Deployment

Deployment is the stage that rolls out a completed model into a production environment.

It is not the stage that additionally trains a model and adapts it to a specific task, so it is incorrect.

Key Takeaway

Remember the positioning of the correct answer, 'fine-tuning.'
- Adapts a pretrained foundation model to a specific task by additionally training it with labeled data.
- With less data and compute than pre-training, it can shape behavior, terminology, and output format to fit the task.
It is at a different stage from data selection (first), pre-training (general-purpose learning), and deployment (rollout).

Related Links