Extracting CLIP embedding

[TooManyMatts]

Hi, thanks for sharing the code!

Small question about extracting the clip embedding. Your evals/models/clip.py has a forward method that outputs a tensor of shape:
torch.Size([1, 768, 32, 32])

Is there a straightforward way of extracting a feature vector from your forward function that would return something of shape:
torch.Size([1, 512])

i.e. feature vector that is commonly used? For example, using open_clip you would do:
image_features = model.encode_image(image)

Best,
Matt