Hacker News new | ask | show | jobs
by scalalang 372 days ago
https://arxiv.org/pdf/2407.21771

In this research, they revealed that the VLM can pay more attention to the image simply by chaining attention weights.