Hi.
Documentation for CLIP is really comprehensive and along with a collaborator, we were able to quickly cook something up.
Now, in order to reduce the RAM requirements particularly in App Engine (GCP), we need to use a model that is smaller than openai/clip-vit-base-patch32
. Since OpenAI has pre-trained ResNet50 weights for CLIP (reference), I was wondering if it’s possible to load that into CLIPModel
. If so, someone could help me figure out how?