search

mlfoundations / wise-ft

Robust fine-tuning of zero-shot models
https://arxiv.org/abs/2109.01903
Other
644 stars 68 forks source link

Question about Table. 2 in the paper #9

Closed vtddggg closed 2 years ago

vtddggg commented 2 years ago

image

How can we get the results in above figure. Do we need to design text prompts for each task and use them to init the classification head?

I try to add the classification head with random init weights, but get poor results for WiSE-FT.

gabrielilharco commented 2 years ago

Hi @vtddggg, thanks for the interest in our work! We indeed initialize the classification head using text prompts. I recommend using the prompts from OpenAI (see here).

vtddggg commented 2 years ago

Thank for your reply! It has solved my question. Closed