search

DavidHuji / CapDec

CapDec: SOTA Zero Shot Image Captioning Using CLIP and GPT2, EMNLP 2022 (findings)
MIT License
185 stars 19 forks source link

About the type of gpt2 #4

Open zjr2000 opened 1 year ago

zjr2000 commented 1 year ago

Hi,

Thanks for your amazing work! I have a question about the type of GPT2. You have mentioned that you use gpt2 large as your langauge model (In section A.1), But I found your code actually load the GPT2 base model:

self.gpt = GPT2LMHeadModel.from_pretrained('gpt2')

Is there any mistake?

DavidHuji commented 1 year ago

Hi, thanks for paying attention to that. It is a mistake in the paper, we used the base model. Please let me know if you tried both and got better results.