Prompt Tuning Params

Problem

Count the trainable parameters of prompt tuning (soft prompts): num_tokens * d_model, since every virtual token is a learnable embedding.

The signature is prompt_tuning_params(num_tokens, d_model). Return an integer.

Hint

Prompt tuning adds no new layers, just a handful of soft token embeddings.

case 1

in: [20,768]

expected: 15360

case 2

in: [1,1]

expected: 1

case 3

in: [100,1024]

expected: 102400

case 4

in: [0,512]

expected: 0

Starter codePython

def prompt_tuning_params(num_tokens, d_model):
    # Your code here
    pass

4 test cases · run your code against them in the editor