Archive
Multimodal & Vision

The Sigmoid Cousin Of CLIP

Archive
Easy
100pts0 solves
Google's 2023 contrastive image-text model that replaces CLIP's softmax with a per-pair sigmoid loss, improving training stability and scalability. Name it (one word). Flag format: CONGRESS{name}. Example: CONGRESS{align}.
Show hint
Sigmoid + Language-Image Pretraining.

Archive — no submissions accepted

This challenge is preserved for reference. Play live challenges at /challenges.