Fetching the paper…

Text encoders bottleneck compositionality in contrastive vision-language models · Around