A collection of Google's SigLIP models split into text and vision towers and config compatible with Huggingface's TEI