Official models for DiffCLIP: Differential Attention Meets CLIP