How to get all the features of the image encoder and text encoder

[wuyukun-tong]

I want to extract more features than just the 512 dimensional cls token from the CLIP pre-trained model, how can I modify that？