#scale

Question 1: Select all that apply

Prefix tuning and prompt tuning both freeze the pretrained language model and learn a few continuous "virtual token" vectors per task. Which statements are correct? Select all that apply.

Both pick real tokens from the vocabulary, so the learned prompts stay readable1
At the T5-XXL scale (11B parameters), prompt tuning matches full model tuning2
Prefix tuning prepends trainable vectors to the activations of every layer3
Prefix tuning also updates the model's weights, but with a small learning rate4
Prompt tuning prepends trainable vectors only to the embedded input5

Timed test