Code attachments to better understand the paper “How Does Diffusion Influence Pretrained Language Models on Out-of-Distribution Data?”
Models finetuned with different ID datasets using diffusion are presented in huggingface: 20ng, sst2, trec, imdb. Please download and move them into the corresponding directory under"./model". The OOD detection results of four different ID datasets are shown in "./ood_datection".