Are the Are there pomp weights for the ViT-B/32 backbone variants of the CLIP model now?

[taoxinlily]

Thanks for yout great work! Are there pomp weights for the ViT-B/32 backbone variants of the CLIP model now? I'm looking forward to your reply!

[RenShuhuai-Andy]

Hi, sorry for the late reply. Since I have finished my internship in AWS, it takes me much time to re-prepare the datasets and re-train the prompt.

The POMP weights for the ViT-B/32 backbone are available in https://huggingface.co/ShuhuaiRen/POMP-ViT-Base-32/tree/main. The average cross-dataset accuracy for vit_b32_ep5_randaug2_unc1000_16shots_nctx4_cscFalse_ctpend_seed42.pth.tar and vit_b32_ep20_randaug2_unc1000_16shots_nctx16_cscFalse_ctpend_seed42.pth.tar is 62.0% and 61.8%, respectively.