After training, what is commonly done to adapt a model to new tasks or data?

After training, what is commonly done to adapt a model to new tasks or data?

• A. Increase model size
• B. Fine-tuning after training
• C. Deploy without updates
• D. Increase training data

Answer: (B)

Fine-tuning after training is how a model is adapted to a new task or domain. After a model has been pre-trained on broad data to learn general representations, you tailor it to a specific job by continuing to train it on task-specific data, usually with a smaller learning rate. This updates the model’s weights so it can produce outputs that align with the new task while preserving the useful knowledge it learned during pretraining. Often you’ll freeze some earlier layers or add small adapters to avoid overhauling the whole network, making the adaptation efficient and targeted. Increasing the model size isn’t the standard way to adapt to a new task; it changes capacity rather than directly aligning the model to the new objective. Deploying without updates won’t equip the model to handle the new task. Simply more training data can help, but without the fine-tuning process that adjusts the model’s parameters to the specifics of the task, the adaptation isn’t achieved. Fine-tuning is the direct, effective approach to specialize a pre-trained model for a new task.

Fine-tuning after training is how a model is adapted to a new task or domain. After a model has been pre-trained on broad data to learn general representations, you tailor it to a specific job by continuing to train it on task-specific data, usually with a smaller learning rate. This updates the model’s weights so it can produce outputs that align with the new task while preserving the useful knowledge it learned during pretraining. Often you’ll freeze some earlier layers or add small adapters to avoid overhauling the whole network, making the adaptation efficient and targeted.

Increasing the model size isn’t the standard way to adapt to a new task; it changes capacity rather than directly aligning the model to the new objective. Deploying without updates won’t equip the model to handle the new task. Simply more training data can help, but without the fine-tuning process that adjusts the model’s parameters to the specifics of the task, the adaptation isn’t achieved. Fine-tuning is the direct, effective approach to specialize a pre-trained model for a new task.