Explain why text queries…

SigLIP versus CLIP: how a sigmoid loss changes vision encoder pretraining